7th Workshop on Forest Fire Management - EARSeL, European ...

Proceedings of the VII International EARSeL <strong>Workshop</strong> 

Matera (Italy), 2 - 5 september 2009 

Advances on Remote Sensing and GIS applications 

in Forest Fire Management 

Towards an operational use of Remote Sensing in 

Forest Fire Management 

Il Segno 

Edit by 

Emilio Chuvieco 

Department of Geography, 

University of Alcalá (Spain) 

Rosa Lasaponara 

CNR-IMAA, (Italy)

Cover design: Nicola Afflitto (IMAA-CNR) 

Disclaimer: the Editors and the Publisher accept no responsibility for errors or 

omissions in the papers and shall not be liable for any damage to property or 

persons arising from the use of information contained herein. 

Published and distributed by 

Il Segno - Arti grafiche - Soc. coop. sociale 

Via R. Danzi, 10 - tel. 0971.650495 - 340.5424913 

POTENZA (Italy) 

ISBN: 978-88-904367-0-3 

Published for: 

EARSeL 

European Association of Remote Sensing Laboratories

SCIENTIFIC COMMITTEE 

Olivier Arino European Space Agency 

Paulo Barbosa DG Joint Research Centre 

Pietro Alessandro Brivio CNR-IREA, Milano (Italy) 

Andrea Camia Joint Research Centre 

Giuseppe Cavaretta Director DTA CNR (Italy) 

Emilio Chuvieco Department of Geography, 


Claudio Conese CNR-IBIMET, Firenze (Italy) 

Pol Coppin Katholieke Universiteit Leuven (Belgium) 

Vincenzo Cuomo CNR-IMAA, (Italy) 

Mark Danson Telford Institute of Environmental System, 

University of Salford (UK) 

Juan de la Riva Department of Geography and Spatial Management, 

University of Zaragoza (Spain) 

Michael Flanningan Candian Forest Service (Canada) 

Karl Fred Hummeric University of Maryland Baltimore County (USA) 

Ioannis Gitas Department of Forestry and Natural Environmental, 

Aristotle University of Thessaloniki 

Robert Keane Missoula Fire Sciences Laboratory 

Nikos Koutsias GIS - Department of Geography 

University of Zurich (Switzerland) 

Rosa Lasaponara CNR-IMAA, (Italy) 

Bruce D. Malamud King’s College London (UK) 

President of the EGU Natural Hazards (NH) Division 

Marc Paganini European Space Agency 

Pilar Martín Consejo Superior de Investigaciones Científicas 

Jesus San Miguel Joint Research Centre 

Guido Schmuck Joint Research Centre 

Carmine Serio Università degli Studi della Basilicata (Italy) 

Donatella Spano Università degli Studi di Sassari (Italy) 

Joost Vandenabeele Belgian Federal Science Policy Office (Belgium) 

ORGANIZING COMMITTEE 

Rosa Lasaponara CNR-IMAA, (Italy) workshop chair 

Nicola Afflitto CNR-IMAA, (Italy) 

Olivier Arino European Space Agency 

Emilio Chuvieco Department of Geography, 


Rosa Coluzzi CNR-IMAA, (Italy) 

Juan de la Riva Department of Geography and Spatial Management, 

University of Zaragoza (Spain) 

Fortunato De Santis CNR-IMAA, (Italy) 

Ioannis Gitas Department of Forestry and Natural Environmental, 

Aristotle University of Thessaloniki 

Antonio Lanorte University of Basilicata, (Italy)

INDEX 

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

I - PRE-FIRE PLANNING AND MANAGEMENT . . . . . . . . . . . . . . . . 

Remote sensing of fuel moisture content: advances in measurement and 

modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

F.M. Danson 

Using spatial analysis to characterize and explain fire regimes in Spain 

M.V. Moreno & E. Chuvieco 

Fire and climate change in Boreal Forests . . . . . . . . . . . . . . . . . . 

M.D. Flannigan, L.M. Gowman & B.M. Wotton 

Projecting future burnt area in the EU-Mediterranean countries under 

IPCC SRES A2/B2 climate change scenarios . . . . . . . . . . . . . . . . . 

G. Amatulli, A. Camia & J. San-Miguel 

Multiscale characterization of spatial pattern over 1996-2006 wildland 

fire events in the Basilicata Region . . . . . . . . . . . . . . . . . . . . . . 

M. Danese, B. Murgante, A. Lanorte, R. Coluzzi, R. Lasaponara 

Forest fire risk assessment in Ibrahim river watershed-Lebanon . . . . 

C. Abdallah 

A forest fire hazard based on the estimation of tourist hot spot activities 

in Austria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

N. Arndt, A. Arpaci, H. Gossow, P. Ruiz Rodrigo, H. Vacik 

Classification of site and stand characteristics based on remote sensing 

data for the development of Fuel models within a 3D Gap forest stand 

model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

A. Arpaci, N. Arndt, M.J Lexer, M. Matiuzzi, M. Müller, H. Vacik 

Fuel moisture content estimation: a land-surface modelling approach 

applied to African Savannas . . . . . . . . . . . . . . . . . . . . . . . . . . . 

D. Ghent, A. Spessa 

Fuel type mapping using SPOT-5 imagery and object based image analysis 

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

I. Stergiopoulos, A. Polychronaki, I.Z. Gitas, G. Galidaki, K. Dimitrakopoulos, G. 

Mallinis 

Fuel model mapping using ikonos imagery to support spatially explicit 

fire simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

B. Arca, V. Bacciu, G. Pellizzaro, M. Salis, A. Ventura, P. Duce, D. Spano, G. Brundu 

First steps towards a long term forest fire risk of Europe . . . . . . . . 

S. Oliveira, A. Camia & J. San-Miguel 

pag. 9 

pag. 13 

pag. 15 

pag. 21 

pag. 25 

pag. 33 

pag. 39 

pag. 45 

pag. 51 

pag. 57 

pag. 63 

pag. 69 

pag. 75 

pag. 79

Analysis of human-caused wildfire occurrence and land use changes in 

France, Spain and Portugal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

L. Vilar, M.P. Martín, A. Camia 

Fuel moisture content estimation based on hyperspectral data for fire 

risk assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

T.A. Almoustafa, R.P. Armitage & F.M. Danson 

Cyberpark project: Multitemporal satellite data set for pre-operational 

fire susceptibility monitoring and post-fire recovery estimation . . . . 

A. Lanorte, F. De Santis, R. Coluzzi, T. Montesano, M. Monteleone, R. Lasaponara 

Remote Sensing-based Mapping of fuel types using Multisensor, 

Multiscale and Multitemporal data set . . . . . . . . . . . . . . . . . . . . 

R. Lasaponara, A. Lanorte, R. Coluzzi 

Assessing critical fuel parameters using airborne full waveform lidar: the 

case study of Bosco dell’Incoronata (Puglia Region) . . . . . . . . . . . 

R. Lasaponara, A. Guariglia, A. Lanorte, R. Coluzzi 

II - VALIDATION OF RS PRODUCTS FOR FIRE MANAGEMENT . . . . . . 

Assessment of spectral indices derived from modis data as fire risk indicators 

in Galicia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

M.M. Bisquert, J.M. Sánchez, V. Caselles, M.I. Paz Andrade & J.L. Legido 

Relationships between combustion products and their spectral properties 

in fire-affected shrublands . . . . . . . . . . . . . . . . . . . . . . . . . 

R. Montorio, F. Pérez-Cabello, A. García-Martín, V. Palacios & J. de la Riva 

Multi-criteria fuzzy-based approach for mapping burned areas in southern 

Italy with ASTER imagery . . . . . . . . . . . . . . . . . . . . . . . . . . 

M. Boschetti, D. Stroppiana & P. A. Brivio 

Operational use of remote sensing in forest fire management in Portugal 

C.C. DaCamara, T.J. Calado, C. Gouveia 

Estimation of national fire danger rating system 10 hour timelag fuel 

moisture content with MSG-SEVIRI data . . . . . . . . . . . . . . . . . . . 

H. Nieto, I. Aguado, E. Chuvieco, I. Sandholt 

Global monitoring of the environment and security: a comparison of the 

burned scar mapping services of the RISK-EOS project . . . . . . . . . . 

M. Paganini, O. Arino, A. Priolo, G. Florsch, Y. Desmazières, C. Kontoes, I. 

Keramitsoglou, R. Armas, A. Sá 

Effects of fire on surface energy fluxes in a Central Spain Mediterranean 

forest. Ground measurements and satellite monitoring . . . . . . . . . . 

J.M. Sánchez, E. Rubio, F.R. López-Serrano, V. Caselles & M.M. Bisquert 

Daily monitoring of pre-fire vegetation conditions using satellite MODIS 

data: the experience of FIRE-SAT in the Basilicata Region . . . . . . . 

A. Lanorte, R. Lasaponara, R. Coluzzi, G. Basile, G. Loperte, F. Antonucci 

The global MODIS burned area product: validation results . . . . . . . . 

L. Boschetti, D.P. Roy, C.O. Justice 

pag. 85 

pag. 91 

pag. 95 

pag. 99 

pag. 103 

pag. 107 

pag. 109 

pag. 115 

pag. 121 

pag. 127 

pag. 133 

pag. 139 

pag. 145 

pag. 151 

pag. 155

III - FIRE DETECTION AND FIRE MONITORING . . . . . . . . . . . . . . . 

Analysis of CO emissions, by forest fires, in the Iberian Peninsula . . 

A. Calle, J-L. Casanova, J. Sanz & P. Salvador 

Real-time monitoring of the transmission system: watching out for fires 

P. Frost, H. Vosloo, A. Momberg & I.T. Josephine 

Noaa’s operational fire and smoke detection program . . . . . . . . . . . 

M. Ruminski, P. Davidson, R. Draxler, S. Kondragunta & J. Simko, J. Zeng, P. Li 

System for early forest fire detection: firewatch . . . . . . . . . . . . . . 

T. Berna, F. Manassero 

Early warning system for fires in Mexico and Central America . . . . . 

G. López Saldaña, I. Cruz López, R. Ressl 

Advancing the use of multi-resolution remote sensing data to detect and 

characterize biomass burning . . . . . . . . . . . . . . . . . . . . . . . . . . 

W. Schroeder, I. Csiszar, L. Giglio & C. Justice 

Sigri - an integrated system for detecting, monitoring, characterizing 

forest fires and assessing damage by LEO-GEO Data . . . . . . . . . . . . 

F. Ferrucci, R. Rongo, A. Guarino, G. Fortunato, G. Laneve, E. Cadau, B. Hirn, C. Di 

Bartola, L. Iavarone, R. Loizzo 

IV - BURNED LAND MAPPING, FIRE SEVERITY DETERMINATION, AND 

VEGETATION RECOVERY ASSESSMENT . . . . . . . . . . . . . . . . . . 

Improvement of dNBR burnt areas detection procedure by physical considerations 

based on NDVI index . . . . . . . . . . . . . . . . . . . . . . . . 

R. Carlà, L. Santurri, L. Bonora & C. Conese 

Burnt area index using MODIS and ASTER Data . . . . . . . . . . . . . . . 

A. Alonso-Benito, P.A. Hernandez-Leal, A. Gonzalez-Calvo, M. Arbelo, A. Barreto & 

L. Nunez-Casillas 

Satellite derived multi-year burned area perimeters within the national 

parks of Italy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

P. A. Brivio, B. Petrucci, M. Boschetti, P. Carrara, M. Pepe, A. Rampini, D. 

Stroppiana & P. Zaffaroni 

Burn severity and burning efficiency estimation using simulation models 

and GeoCBI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

A. De Santis, G.P. Asner, P.J. Vaughan, D. knapp 

MODIS reflective and active fire data for burn mapping in Colombia . 

S. Merino-De-Miguel, F. González-Alonso & M. Huesca, D. Armenteras, S. Opazo 

A new algorithm for the ATSR World Fire Atlas . . . . . . . . . . . . . . . 

S. Casadio, O. Arino 

Assessment of post fire vegetation recovering in Portugal . . . . . . . 

C. Gouveia, C. Dacamara & R. Trigo 

Some notes on spectral properties of burnt surfaces at sub-pixel level 

using multi-source satellite data . . . . . . . . . . . . . . . . . . . . . . . . 

N. Koutsias & M. Pleniou 

pag. 159 

pag. 161 

pag. 167 

pag. 171 

pag. 175 

pag. 181 

pag. 187 

pag. 193 

pag. 201 

pag. 203 

pag. 209 

pag. 215 

pag. 221 

pag. 227 

pag. 233 

pag. 237 

pag. 243

Monitoring post-fire vegetation regeneration of the 2003 burned areas 

in Portugal using a time-series of MODIS enhanced vegetation index 

P. Malico, J. Kucera, J. San-Miguel Ayanz, B. Mota, J. M.C. Pereira 

Properties of X- and C-band repeat-pass interferometric SAR coherence 

in Mediterranean pine forests affected by fires . . . . . . . . . . . . . . . 

M. Tanase, M. Santoro & U. Wegmüller, J. De La Riva & F. Pérez-Cabello 

Backscatter properties of X- and C-band SAR in A Mediterranean pine 

forest affected by fire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

M. Tanase, J. De La Riva, F. Pérez-Cabello, M. Santoro 

Burned area data time series from LTDR dataset for Canada (1981-2000) 

J.A. Moreno Ruiz, D. Riaño, A. Alonso-Benito, N.H.F. French & S.L. Ustin 

Correction of topographic effects influencing the differenced Normalized 

Burn Ratio’s optimality for estimating fire severity . . . . . . . . . . . . 

S. Veraverbeke, R. Goossens, W. Verstraeten, S. Lhermitte 

Burned areas mapping by multispectral imagery: a case study in Sicily, 

summer 2007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

P. Conte, G. Bitelli 

Post fire vegetation recovery estimation using satellite VEGETATION time 

series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

R. Lasaponara, F. De Santis, R. Coluzzi, A. Lanorte, L. Telesca 

Author index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 

pag. 247 

pag. 253 

pag. 259 

pag. 265 

pag. 271 

pag. 277 

pag. 285 

pag. 291

PREFACE 

In the last decades, the improved spatial and spectral capability of active 

and passive sensors has opened new challenging prospective for the use of 

EO (Earth Observation) technologies for the management and monitoring of 

forest fires. The increasing development of ground, aerial and space remote 

sensing techniques and the tremendous advancement of Information and 

Communication Technologies (ICT) have focused a great interest in the use 

of remote sensing and ICT for supporting forest fire management and monitoring 

in: 

- pre-fire planning and management; 

- fire detection and monitoring; 

- post-fire evaluation and management; 

- post-fire vegetation recovery assessment. 

Additional challenges to this field of research are related to the crucial 

importance of the integration of remote sensing with other traditional 

technologies and ancillary data. Such an integration requires great efforts 

aimed at creating a strong interaction among scientists and managers interested 

in using remote sensing and ICT for forest fire management and monitoring. 

The continuous collaboration among scientists working in different 

fields can strongly contribute to take benefits from the new sensors, techniques 

and methodological approaches for a wide range of investigation 

and application fields. A constructive and complementary multidisciplinary 

approach can open a revolutionary scenario unthinkable several decades 

ago. 

In this cultural framework, the EARSeL Special Interest Group (SIG) on 

Forest Fire acts to establish contacts and foster interaction among scientists 

and managers interested in using remote sensing data (from ground, 

aerial and satellite) and Information Technologies to support and improve 

traditional approach for forest fire management and monitoring. 

In the context of the EARSeL FF-SIG activities, we have been pleased to 

organize the VII International <strong>Workshop</strong> “Advances on Remote Sensing and 

GIS applications in Forest Fire Management: Towards an operational use of 

Remote Sensing in Forest Fire Management”. The workshop is organized by 

the Institute of Environmental Analysis of the National Council Research in 

collaboration with the University of Alcalá and the European Space Agency. 

The event has been carried out with the patronage of the Italian Protection 

9

10 

Agency, Italian National Forestry Service, BELSPO, Comune di Matera and 

the sponsorship of the Italian Space Agency (ASI), University of Basilicata, 

Regione Basilicata-Consiglio Regionale, (Agency for the Tourism) APT- 

Basilicata, Consorzio TERN (Technology for Earth Observation and Natural 

Risk - TEcnologie per le osservazioni della Terra ed i Rischi Naturali) and 

Geocart srl. 

Matera workshop is the latest in a series of technical meetings organised 

by the EARSeL SIG on Forest Fires after its foundation in 1995. Previous 

meetings were held in Alcalá de Henares (1995), Luso (1998), Paris (2001), 

Ghent (2003), Zaragoza (2005) and Thessaloniki (2007) resulted in outstanding 

progress made in forest fire research. 

The Proceeding book includes papers divided in 4 sections which focus the 

following topics: 

- Pre-fire planning and management; 

- Validation of Remote Sensing Products for fire management; 

- Fire Detection and fire monitoring; 

- Burned land mapping, fire severity determination and vegetation recovery 

assessment. 

Rosa Lasaponara Emilio Chuvieco 

Local workshop chair EARSeL FF-SIG Chair

Acknowledgement 

We want to thank the Italian Space Agency (ASI) for funding the publication 

of this book. 

We are grateful to all of those whose help has permitted us to bring the VII 

<strong>Workshop</strong> on “Advances on Remote Sensing and GIS applications in Forest 

Fire Management: Towards an operational use of Remote Sensing in Forest 

Fire Management” to a successful conclusion. 

We are grateful to eveyone responsible for all the organization and arrangements: 

Nicola Afflitto, Rossella Coluzzi, Canio De Bonis, Antonio Lanorte, 

Margherita Santarsiere.

I PRE-FIRE PLANNING 

AND MANAGEMENT

REMOTE SENSING OF FUEL MOISTURE CONTENT: ADVANCES IN 

MEASUREMENT AND MODELLING 

F.M. Danson 

Centre for Environmental Systems Research, School of Environment and Life Sciences, 

University of Salford, Salford, UK 

f.m.danson@salford.ac.uk 

Abstract: Vegetation live fuel moisture content (FMC) is routinely measured 

in areas prone to wildfires as it has an important influence on the probability 

of ignition of the vegetation, and also affects the rate at which fires 

may spread. FMC is easy to measure in the field, requiring only the fresh 

and dry weight of a vegetation sample. Field sampling is often limited both 

in space and time however and recent remote sensing research has attempted 

to develop methods for spatial mapping of FMC. This paper reviews this 

research including, the fundamental physical relationships between FMC 

and other variables, empirical work on FMC estimation, assessing the role 

of canopy reflectance modelling, and conclusions on research priorities. 

1 - Introduction 

Vegetation live FMC is controlled by the interaction of plant physiology, soil 

moisture and atmospheric conditions. It is therefore spatially and temporally 

highly variable and difficult to map using ground measurements which 

are normally sparse in space and time (Chuvieco et al., 2002). 

Dimitrakopoulos & Papaioannou (2001) showed that for many 

Mediterranean species a FMC value of 100% (or less) represents the ‘moisture 

of extinction’, and ignition is possible. Spatial and temporal mapping 

of FMC will therefore provide critical information on when and where the 

moisture of ignition is reached. 

FMC may be described as a ‘composite’ variable because it is determined by 

a combination of two independent physical properties of the vegetation, 

the leaf equivalent water thickness (EWT) and the leaf dry matter content 

(DM), both with units of g cm-2 (equation 1): 

FMC (%) = EWT/DM *100 (1) 

In field sampling however, FMC is more easily estimated from leaf sample 

fresh and dry weights. 

15

16 

I - PRE-FIRE PLANNING AND MANAGEMENT 

Remotely sensed data have been used to estimate vegetation FMC with 

some success in a range of different environments, based known physical 

relationships between leaf water content and spectral reflectance in the 

near- and middle-infrared (Danson et al., 1992). However, remotely sensed 

data are sensitive to the total amount of water in a canopy, often derived 

as the product of EWT and leaf area index (LAI). Hence, when we attempt 

to estimate FMC from remotely sensed data we would expect the relationships 

to be affected by independent variation in EWT, DM and LAI, as well 

as by a range of other spatial and temporal variables. This statement is 

axiomatic and yet it is rarely invoked in explaining variations in the 

strengths of the relationships between FMC and remotely sensed data. 

2 - Remote sensing of fuel moisture content 

Early research on the use of remote sensing to estimate plant water content 

includes the work of Gates et al., (1965), Thomas et al., (1971) and 

Palmer & Williams (1974). Much of this was directed at agricultural crops 

where water stress may be a limiting factor in plant production. The parameter 

measured in these studies was normally gravimetric water content or 

in later work the EWT. Subsequent areas of research include detailed laboratory-based 

studies, application of radiative transfer model inversion and 

a large number of field experiments relating plant water content to remotely 

sensed data (e.g., Danson et al., 1992, Jacquemoud et al., 1993, Baret 

& Fourty, 1997, Jackson et al., 2004). Early publications specifically relating 

to vegetation FMC include the work of Paltridge & Barber (1988) and, 

over a decade later, Hardy & Burgan (1999). This, and most of the subsequent 

research, has been based on establishing empirical relationships 

between field measured FMC, sampled across space or time, and vegetation 

indices (VI) from remotely sensed images. The normalized difference vegetation 

index (NDVI) has often been used as it is easy to compute and can 

be derived from a range of sensors including the NOAA AVHRR, in spite of 

the fact that the neither of the wavebands used in the NDVI is directly sensitive 

to vegetation water content. Recent studies have employed a wider 

range of indices using combinations of wavebands in the shortwave- and 

near-infrared that are directly sensitive to water content. A significant 

number of empirical studies have shown stronger correlations between FMC 

and VI for grasslands than for shrublands or forests. This is due to the simple 

structure of grassland canopies and the co-correlation of FMC with other 

variables like LAI. Few studies have measured other confounding variables 

so that this hypothesis usually remains untested.

Remote sensing of fuel moisture content: advances in measurement and modelling 17 

3 - Use of radiative transfer models 

Canopy reflectance models use radiative transfer theory to describe the 

interaction of radiation with plant canopies and include variables like leaf 

chlorophyll, leaf water content, and canopy leaf area index. Since these 

variables may be correlated in ‘real’ canopies, radiative transfer models 

allow sensitivity analyses to be performed, in which variables may be coupled 

or decoupled, in order to better understand the relationships between 

leaf water content and spectral reflectance. This approach also allows definition 

of the conditions under which FMC may be reliably estimated from 

remotely sensed data. Sensitivity analyses have demonstrated the importance 

of interactions between EWT, DM, LAI and FMC at leaf and canopy 

level (Ceccato et al., 2002, Bowyer & Danson, 2004). A few subsequent 

papers have used radiative transfer models to estimate FMC in laboratory 

and field experiments and it is clear from this work that, when the models 

can be constrained using measured ranges of the relevant variables at a 

given site, the estimation of FMC is more accurate. This work is at an early 

stage and further work combining field measurements and modelling is still 

required. 

4 - Research priorities 

A survey of published journal papers from 2002 onwards, and focused on 

the estimation of FMC from remotely sensed data, was conducted for the 

purposes of this review. A total of 21 papers were identified and classified 

according to the ecosystem within which the work took place and the main 

sensor used. In addition, the papers were classified according to the 

methodology adopted; papers using bivariate or multivariate correlation 

and regression analyses and the application of vegetation indices formed 

one group; papers that primarily used radiative transfer models formed a 

second group. 

Target ecosystem 

Sensor Mediterranean Savanna Other 

Modis 6 (1) (2) 

SPOT VGT 2 1 

AVHRR 3 

Landsat TM 2 

Hyperspectral 3 (1) 

Table 1 - Survey of 21 papers on remote sensing of FMC published between 2002 and 2009. 

Figures indicate number of empirical studies and in brackets model based studies.

18 


The results of the survey (Table 1) show that 16 out of 21 papers (76%) 

focused on Mediterranean ecosystems, 9 out of 21 papers (43%) used Modis 

data, only 4 out of 21 papers (19%) included some element of modelling 

and only 4 out of 21 (19%) made use of hyperspectral data. From this survey 

we could conclude that research priorities include further work using 

hyperspectral sensors, exploration of model-based approaches and extension 

of research on FMC and remote sensing beyond Mediterranean ecosystems. 

References 

Baret, F., & Fourty, T., 1997. Estimation of leaf water content and specific 

leaf weight from reflectance and transmittance measurements. 

Agronomie, 17, 455-464. 

Bowyer, P. & Danson, F.M., 2004. Sensitivity of spectral reflectance to variation 

in live fuel moisture content at leaf and canopy level. Remote 

Sensing of Environment, 92, 297-308. 

Ceccato, P., Gobron, N., Flasse, S., Pinty, B. & Tarantola, S., 2002. 

Designing a spectral index to estimate vegetation water content from 

remote sensing data: Part 1. Theoretical approach. Remote Sensing of 

Environment, 82, 188-197. 

Chuvieco, E., Rianõ, D., Aguado, I. & Cocero, D., 2002, Estimation of fuel 

moisture content from multitemporal analysis of Landsat Thematic 

Mapper reflectance data: Applications in fire danger assessment. 

International Journal of Remote Sensing, 23, 2145-2162. 

Danson, F. M., Steven, M. D., Malthus, T. J. & Clark, J. A., 1992. High-spectral 

resolution data for determining leaf water content. International 

Journal of Remote Sensing, 13, 461-470. 

Danson, F.M. & Bowyer, P., 2004. Estimating live fuel moisture content from 

remotely sensed reflectance. Remote Sensing of Environment, 92, 309- 

321. 

Dimitrakopoulos, A. & Papaioannou, K. K., 2001. Flammability assessment 

of Mediterranean forest fuels. Fire Technology, 37, 143-52. 

Gates, D. M., Keegan, H. J., Schleter, J. C., & Weidner, R., 1965. Spectral 

properties of plants. Applied Optics, 4, 11-20. 

Hardy, C. C., & Burgan, R. E., 1999. Evaluation of NDVI for monitoring live 

moisture in three vegetation types of the Western U.S. Photogrammetric 

Engineering and Remote Sensing, 65, 603-610. 

Jackson, T. J., Chen, D. Y., Cosh, M., LI, F. Q., Anderson, M. & Walthall, C., 

2004, Vegetation water content mapping using LANDSAT data derived 

normalized difference water index for corn and soybeans. Remote 


Jacquemoud, S., Baret, F., Andrieu, B., Danson, F. M., & Jaggard, K. W., 

1995. Extraction of vegetation biophysical parameters by inversion of 

PROSPECT + SAIL model on sugar beet canopy reflectance data.

Remote sensing of fuel moisture content: advances in measurement and modelling 19 

Application to TM and AVIRIS sensors. Remote Sensing of Environment, 

52, 163-172. 

Palmer, K. F., & Williams, D., 1974. Optical properties of water in the near 

infrared. Journal of the Optical Society of America, 64, 1107-1110. 

Paltridge, G. W., & Barber, J., 1988. Monitoring grassland dryness and fire 

potential in Australia with NOAA/AVHRR data. Remote Sensing of 


Thomas, J. R., Namken, L. N., Oerther, G. F., & Brown, R. G., 1971. 

Estimating leaf water content by reflectance measurements. Agronomy 

Journal, 63, 845-847.

USING SPATIAL ANALYSIS TO CHARACTERIZE AND EXPLAIN FIRE 

REGIMES IN SPAIN 

Abstract: The impact of wildland fires on society and the environment is 

strongly associated to fire characteristics, which are classified using the 

concept of fire regimes. The definition fire regimes should consider different 

aspects of fire occurrence, such as fire density, fire frequency, fire size, 

fire seasonality, intensity or severity. This paper presents the classification 

of fire regimes in Spain using the historical fire records database of the 

Spanish forest service, which are collected at a grid size of 10x10 km. Using 

spatial and statistical analysis, several regions of different fire regimes were 

identified. 


M.V. Moreno & E. Chuvieco 

Departamento de Geografía, Universidad de Alcalá, 

Alcalá de Henares, Colegios 2, Spain 

vanesa.moreno@uah.es; emilio.chuvieco@uah.es 

Fire occurrence can not be considered as a binary process (fire/non fire), 

but rather as a combination of different characteristics, which explain the 

environmental and societal impact of fire. Basic aspects to define fire 

regimes are the number of fires, their size, seasonality, persistency and 

causality (Chuvieco et al., 2008). Understanding fire regimes is therefore 

critical to alleviate the negative impacts of fire and improve post-fire management 

(Morgan et al., 2001). 

Spain is one of the most affected countries in Europe by forest fires, since 

it has the largest forested area under Mediterranean conditions (Pausas and 

Vallejo, 1999; Vélez, 2001). Recent changes in socio-economic conditions 

have implied a growing occurrence of fire, while the affected area closely 

depends on climatic cycles. The country has several areas with very different 

fire characteristics, caused by the diversity of both social and climatic 

conditions. The main goal of this paper is to characterize different fire 

regimes in Spain using spatial and temporal analysis of the Spanish fire statistics. 

21

22 


2 - Methods 

2.1 - National Forest Fire Data Base 1980-2004 

The input data was extracted from the Spanish Forest Service (Ministry of 

Environment, Rural and Marine affairs), and includes all fire records from 

1980 to 2004. The fire records are georeferenced to the UTM grid of 10*10 

km. Only those cells with more than 1% of the total area of the grid burned 

in the 25 year period were selected for further analysis. A total of 2,957 

cells were selected, covering around 56% of the total area of Spain. More 

than 320,000 fires were analyzed. 

The following fire metrics were used to characteristic fire regimes for each 

cell: 

• Fire density: ratio of the total number of fires and the cell area. 

• Fire frequency: annual average number of fires. 

• Fire interannual variability: variation coefficient (ratio of the standard 

deviation and the mean of the annual number of fires). 

• Fire seasonality: season with largest number of fires. 

• Mean fire size: ratio of the burned area and the total number of fires. 

• Fire severity: total burned area in different fire sizes. 

3 - Results and discussion 

3.1 - Description of fire regimes characteristics 

Fire density. Average fire density in Spain is 1.28 fires by km 2 , and the standard 

deviation is 3.02 fires/km 2 . The higher densities were found in the 

Northwest, with a predominant Atlantic climate, the Mediterranean coastal 

regions and the central sierras. Medium and low values were found in other 

mountainous areas in the South and the Eastern ranges (fig. 1). Fire frequency 

shows similar patterns, with average values of 4.33 and a Standard 

deviation of 8.98. 

Fire seasonality. Most fires in Spain are summer fires, since 56% of all grid 

cells have their mode in the summer season (June to August). In these 

months, around 40% of all fires occur, and they burn 58% of total affected 

area. The spring season (March to May) is the second highest occurrence, 

with 24% of all fires affecting to 12% of total burned area. Spring fires are 

more common in the northern regions of the country, which also have a second 

mode in winter. Most of those fires are related to agricultural practices. 

Fire interannual variability is higher in the Mediterranean coast and the 

interior mountain areas (fig. 2), since fires are more associated to different 

climatic cycles, while in the Northern regions fires are more closely linked 

to agricultural practices. The coefficient of variation ranges from 39.6% to 

500%. The impact of large fires is quite evident in this regard, which are

Using spatial analysis to characterize and explain fire regimes in Spain 23 

very associated to extended summer drought conditions, with higher 

impact in 1985, 1989, 1991 and 1994. 

Fire severity. Most fires were very small in size (43.8% burned less than 1 

ha), and they affect a fairly limited area (0.8% of total burned area). In 

the opposite side, fires above 500 has only account for 0.4% of the total 

number of fires, but they burned a high proportion of the total affected 

area (41.4%). Fire sizes tend to be bigger in the Mediterranean coast, as a 

result of climatic influences (continental dry winds, heat waves). 

4 - Conclusions and future work 

This paper is a first attempt to characterize fire regimes in Spain using fire 

metrics derived from the historical fire records collected by the Spanish forest 

service. Spatial and temporal variations have been considered, as well 

as different patterns of fire size and seasonality. Few large fires have the 

greater impact, while a large number of fires are very small and have little 

influence. The spatial and temporal patterns need to be explained using 

socio-economic and environmental factors, mainly those related to agricultural 

practices, land abandonment and drought cycles. 

Future work should focus on those explanation factors, based on auxiliary 

variables within a GIS. Additionally, fire metrics should serve as an input 

for a classification of fire regimes in the country, which should be the basis 

for estimation current and future impacts of fire on society and ecosystems. 

References 

Chuvieco, E., Giglio, L., and Justice, C., 2008. Global characterization of fire 

activity: toward defining fire regimes from Earth observation data. Global 

Change Biology, 14, 1488-1502. 

Diego, C., Carracedo, V., García, J.C. and Pacheco, S., 2004. Clima, prácticas 

culturales e incendios en Cantabria. In García, J.C., Diego, C., Fdez. 

de Arróyabe, P., Garmendia, C. y Rasilla, D. (Eds.), 2004. El Clima entre 

el Mar y la Montaña. Asociación Española de Climatología Universidad 

de Cantabria,, Santander. Serie A, nº 4 

Morgan, P., Hardy, C.C., Swetnam, T.W., Rollins, B. and Long, D.G., 2001. 

Mapping fire regimes across time and space: Understanding coarse and 

fire-scale fire patterns. International Journal of Wildland Fire, 10, 329- 

342. 

Pausas, J.G. and Vallejo, V.R., 1999. The role of fire in European 

Mediterranean Ecosystems. In Chuvieco E. (ed.) Remote sensing of large 

wildfires in the European Mediterranean basin. Springer-Verlag. 3-6 

Vélez, R., 2001. Fire situation in Spain. In Global Forest Fire Assessment 

1990-2000. FAO, Roma.

24 


Figure 1 - Fire Density / km 2 . 

Figure 2 - Fire Interannual Variability expressed as the coefficient of variation.

Abstract: Fire is the major stand-renewing agent for much of the circumboreal 

forest, and greatly influences the structure and function of boreal 

ecosystems from regeneration through mortality. Current estimates are that 

an average of 5-15 million hectares burn annually in boreal forests, almost 

exclusively in Siberia, Canada and Alaska. There is a growing global awareness 

of the importance and vulnerability of the boreal region to projected 

future climate change. Fire activity is strongly influenced by four factors – 

weather/climate, vegetation(fuels), natural ignition agents and humans. 

Climate and weather are strongly linked to fire activity which suggests that 

the fire regime will respond rapidly to changes in climate. Global atmospheric 

and oceanic dynamics play a major role in circumboreal fire activity. 

Recent results suggest that area burned by fire is related to temperature 

and fuel moisture. The climate of the northern hemisphere has been warming 

due to an influx of radiatively active gases (carbon dioxide, methane 

etc.) as a result of human activities. This altered climate, modelled by 

General Circulation Models (GCMs), indicates a profound impact on fire 

activity in the circumboreal forest. Recent results using GCMs suggest that 

in many regions fire weather/fire danger conditions will be more severe, 

area burned will increase, people-caused and lightning-caused ignitions 

will increase, fire seasons will be longer and the intensity and severity of 

fires will increase. Changes in fire activity as a result of climate change 

could have a greater and more immediate impact on vegetation distribution 

and abundance as compared to the direct impact of climate change. 

1 - Future fire activity 

FIRE AND CLIMATE CHANGE IN BOREAL FORESTS 

M.D. Flannigan 

Great Lakes Forestry Centre, Canadian Forest Service, Sault Ste Marie, Canada 

mike.flannigan@nrcn.gc.ca 

L.M. Gowman & B.M. Wotton 

Canadian Forest Service, Sault Ste Marie, Canada 

lynn.gowman@nrcan.gc.ca; mike.wotton@nrcan.gc.ca 

Fire activity responds dynamically to the weather/climate, fuels, and people. 

Recently, our climate has been warming as a result of increases of 

radiatively active gases (carbon dioxide, methane etc.) in the atmosphere 

from human activities (IPCC, 2007). Such warming is likely to have a rapid 

25

26 


and profound impact on fire activity, as will potential changes in precipitation, 

atmospheric moisture, wind, and cloudiness (Flannigan et al., 2006; 

IPCC, 2007). Vegetation patterns, and thus fuels for fire, will change in the 

future due to both direct effects of climate change and indirectly as a result 

of changing fire regimes (Soja et al., 2007). Humans will continue to be a 

crucial element of fire activity in the future through fire management, 

human-caused fire ignitions, and land-use. In the future, changes in 

weather/climate, fuels, and people and the non-linear, complex and sometimes 

poorly understood interactions among these factors will determine 

fire activity. 

1.1 - Fire weather 

More severe and extreme fire weather is predicted by 2x and 3xCO 2 climate 

models in the circumboreal forest, with increases in the seasonal severity 

rating (SSR; a rating index to provide a measure of fire control difficulty 

and is a component of the Canadian Forest Fire Weather Index System) of 

up to 50% (Flannigan and Van Wagner, 1991; Stocks et al., 1998; Flannigan 

et al., 2000). Changes in fire weather are predicted to be highly variable 

depending on location. For example, Flannigan et al., (1998) predict that 

the fire weather index (FWI; a rating of fire danger that incorporates temperature, 

humidity, wind speed, and precipitation) may decrease in eastern 

Canada, western Canada, and most of northern Europe (Sweden, Finland, 

and western Russia); increases are expected in southern Sweden and 

Finland and throughout central Canada. Other studies focusing on the 

Canadian boreal forest found that the FWI will decrease for some of eastern 

Canada, but increase for most of the rest of the country in a 2xCO 2 climate 

(Bergeron and Flannigan, 1995; Flannigan et al., 2001). In the 

Russian boreal forest, it has been predicted that areas of maximum fire danger 

risk will double by 2050, and changes will vary spatially across the 

country (Malevsky-Malevich et al., 2008). 

1.2 - Area burned 

Flannigan and Van Wagner (1991) compared SSR from a 2xCO 2 scenario (mid 

21st century) versus the 1xCO 2 scenario (approx. present day) across 

Canada. The results suggest increases in the SSR across all of Canada with 

an average increase of nearly 50%, translating roughly to a 50% increase 

of area burned by wildfire. Bergeron et al., (2004) suggest increases in area 

burned for most sites across Canada by the middle or end of this century, 

although some sites in eastern Canada were projected to have no change 

or even a decrease. Area burned was projected to increase by as much as 

5.7 times the present values, but for many sites the historical area burned 

(1600s to near present) was higher than estimated future fire activity

Fire and climate change in Boreal Forests 27 

(Bergeron et al., 2004). This comparison emphasizes the need to temper 

our thoughts on changes in fire to include a broad temporal context. 

Flannigan et al., (2005) used historical relationships between weather/fire 

danger and area burned in tandem with two GCMs (global circulation models) 

to estimate future area burned in Canada and suggest an increase of 

74-118% in area burned by the end of this century. Using a dynamic global 

vegetation model (DGVM) to examine climate, fire, and ecosystem interactions 

in Alaska, Bachelet et al., (2005) suggest area burned increases of 

14-34% for 2025-2099 relative to 1922-1996. McCoy and Burn (2005) predict 

mean annual area burned in the Yukon to increase 33% and maximum 

annual area burned to increase 227%. In boreal Alberta, Tymstra et al., 

(2007) suggest area burned increases of about 13% and 29% for 2x and 

3xCO 2 scenarios relative to the 1xCO 2 scenario using a fire growth model 

with output from the Canadian Regional Climate Model (RCM). Using air 

temperature and fuel moisture codes from the Canadian Forest Fire Weather 

Index System, Balshi et al., (2008) suggest decadal area burned for western 

boreal North America will double by 2041-2050 and will increase in the 

order of 3.5 to 5.5 times by the last decade of the 21 st century as compared 

to 1991-2000. 

1.3 - Fire occurrence 

Many studies have examined the influence of fire weather and fuel moisture 

on fire occurrence itself (e.g. Martell et al., 1987; Wotton and Martell, 

2005; Drever et al., 2006; Krawchuk et al., 2006). There are few studies that 

have used these fire-weather relationships to then project future fire occurrence. 

For example, in the mixedwood boreal forest of central-eastern 

Alberta, Krawchuk et al., (2009) projected regional increases in lightningcaused 

fire occurrence through the 21 st century, showing an expected 80% 

increase in initiation. Wotton et al., (2003) projected an increase in 

human-caused fires in Ontario of 50% by the end of the 21 st century in 

association with climate change; though an overall increase in fire was projected, 

there were areas where less severe conditions and reduced fire 

occurrence were projected. At a much broader scale, both human and lightning-caused 

fire was examined across the entire forested area of Canada 

and the results suggest that there is significant regional variation in the 

change in future fire occurrence rates compared to current levels. Fire 

occurrence in the boreal forest of Russia has been predicted to increase, 

with up to 12 more days per year with high fire danger (a component of 

which is fire occurrence) (Malevsky-Malevich et al., 2008). Overall, there is 

significant spatial variability in predicted changes in fire occurrence, with 

areas of increases, decreases, or no change expected across the circumboreal 

forest (Bergeron et al., 2004; Balshi et al., 2008).

28 


2 - Fire management 

Wildland fire management problems are increasing across the boreal forest 

for numerous reasons. Common problems include the expansion of the wildland-urban 

interface (Cottrell, 2005), increasing fire suppression costs, and 

increased hazardous emissions causing greater negative human health 

impacts. Fire management agencies worldwide also recognize the consequences 

of, and the contribution of fire to, climate change. Impacts of climate 

change on the fire environment are generally seen with the trend of 

increasing fire activity, particularly in the circumboreal forest (Table 1). The 

obvious question that arises is: can forest fire management agencies adapt 

and mitigate the impacts of this potential increase in fire activity through 

increasing resource capacity? Under climate change a disproportionate 

number of fires may escape initial attack, resulting in very significant 

increases in area burned; the reasoning behind this hypothesis is that there 

tends to be a very narrow range between the suppression system’s success 

or failure (Stocks, 1993). Detailed simulation of the initial attack system of 

Ontario’s fire management agency, which actively manages fire across 

approximately 50 Mha of boreal forest in Canada, showed that to move the 

escape fire threshold down from current levels, very significant investment 

in resources would be required; that is, incremental increases in fire suppression 

resource lead to diminishing gains in initial attack success. A further 

study using Ontario’s initial attack simulation system with future climate 

change scenarios of fire weather and occurrence showed that current 

resource levels would have to more than double to meet even a relatively 

small increase (15%) in fire load. An agency’s fire load threshold is not the 

only physical limit that might play a role in future success and failure of 

fire management objectives under a changed climate. Direct fire suppression 

methods, including high volume airtanker drops, become relatively 

ineffective once fires become somewhat intense crown fires Thus, if fire 

intensities are to increase as suggested earlier in this paper, one can expect 

the number of situations when direct fire suppression activity is ineffective 

to increase as well. Adaptation to new fire climates may require fire agencies 

and the public to re-examine their current tolerance of fire on the landscape, 

or think beyond fire management practices of the 20 th century to 

mitigate unwanted fire. Options such as treating fuels in the immediate 

vicinity of values at risk may be one of the few viable solutions available 

(e.g., Cary et al., 2009), along with strategically-placed landscape fuel 

treatments. 

In the areas of the circumboreal dominated by human-caused fire, one 

might think increased fire prevention campaigns and enforcement of 

restricted fire zones might help reduce the number of starts during high 

fire-potential periods. However, areas with well established fire prevention 

programs such as southern California still tend to have a significant humancaused 

fire load, though this may be due in part to the rise of arson in 

recent years. It is difficult to predict what changes in societal values or


demographics might occur over the next century that will have a direct 

impact on the number of human ignition sources on the landscape. What 

does seem clear is that in areas where fuel is available, environmental conditions 

(i.e. fuel moisture) will be more conducive to ignition in the future. 

Without major changes in patterns of human activity or fuels, the number 

of fires occurring from human causes will likely increase and thus the presence 

of fire in high value areas, and the consequent pressure on fire management 

agencies to deal with this fire, will likely increase. Through most 

of the 20 th century in most of the world, ‘fire management’ organizations 

tended to be fire suppression organizations, focused on fire exclusion. 

Today there is a recognition of the need to balance fire suppression to protect 

values (e.g., in the wildland-urban interface, or in high value timber 

production areas) with the need to let fire in wilderness areas burn. 

Increased fire activity due to climate change, and increased awareness of 

carbon emission from wildfire as well as potential newly available sources 

of carbon emissions, such as boreal peatlands (Flannigan et al., 2009), are 

added pressures that will make achieving a balance between value protection 

and the ecological needs for fire even more difficult for fire management 

agencies. 

3 - Summary 

A great deal of research has been completed on wildland fire and we are 

beginning to understand the relationships between various aspects of fire 

activity and the key factors such as weather/climate, fuels, and people, 

along with their interactions. These factors are dynamic and will continue 

to change as the climate, fuels, and people respond to global change and 

other influences. Overall, we expect that fire activity will continue to 

increase due to climate change. It appears that fire weather, area burned, 

and fire occurrence are generally increasing, but there will be regions with 

no change and regions with decreases in the circumboreal forest. This spatial 

variation highlights the need to view the impact of climate change on 

fire activity in a spatially-dependent context. The length of the fire season 

appears to be increasing already and should continue to lengthen in the 

future. Fire intensity and severity are more difficult to summarize and this 

is an area in need of further research. There could be surprises in the future, 

perhaps even the near future, with respect to fire activity and this is due 

to our limited understanding of the interactions between weather/climate, 

fuels, and people. 

The role of people in global fire regimes needs much more work as policy, 

practices, and behaviour vary across the circumboreal and with time. The 

more physical aspects of wildland fire have received greater attention by 

the research community but there are still areas that need further work 

including global studies that dynamically model weather, vegetation, people, 

fire, and other disturbances. Lastly, we require accurate data sets of fire

30 


activity. The advent of satellite sensors appropriate to monitor wildland fire 

has been a significant advance in terms of area burned but even those data 

provide a wide range of estimates, and we still do not have an accurate 

estimate of circumboreal fire occurrence. 

References 

Bachelet D., Lenihan J., Neilson R., Drapek R., Kittel T., 2005. Simulating 

the response of natural ecosystems and their fire regimes to climatic 

variability in Alaska. Canadian Journal of Forest Research 35, 2244- 

2257. doi: 10.1139/X05-086 

Balshi M.S., McGuire A.D., Duffy P., Flannigan M., Walsh J., Melillo J., 2008. 

Assessing the response of area burned to changing climate in western 

boreal North America using a Multivariate Adaptive Regression Splines 

(MARS) approach. Global Change Biology 14, 1-23. doi: 10.1111/j.1365- 

2486.2008.01679.x 

Bergeron Y., Flannigan M.D., 1995. Predicting the effects of climate change 

on fire frequency in the southeastern Canadian boreal forest. Water, Air 

and Soil Pollution 82, 437-444. 

Bergeron Y., Flannigan M., Gauthier S., Leduc A., Lefort P., 2004. Past, current 

and future fire frequency in the Canadian boreal forest: implications 

for sustainable forest management. Ambio 33, 356-360. 

Cary G., Flannigan M., Keane R., Bradstock R., Davies I., Li C., Lenihan J., 

Logan K., Parsons R., 2009. Relative importance of fuel management, 

ignition management and weather for area burned: Evidence from five 

landscape-fire-succession models. International Journal of Wildland Fire 

18, 147-156. doi: 10.1071/WF07085 

Cottrell A., 2005. Communities and bushfire hazard in Australia: More questions 

than answers. Environmental Hazards 6, 109–114. 

Drever C.R., Messier C., Bergeron Y., Doyon F., 2006. Fire and canopy species 

composition in the Great Lakes-St. Lawrence forest of Témiscamingue, 

Quebec. Forest Ecology and Management 231, 27-37. doi: 

10.1016/j.foreco.2006.04.039 

Flannigan M.D., Van Wagner C.E., 1991. Climate change and wildfire in 

Canada. Canadian Journal of Forest Research 21, 66–72. 

Flannigan M.D., Bergeron Y., Engelmark O., Wotton B.M., 1998. Future 

Wildfire in Circumboreal Forests in Relation to Global Warming. Journal 

of Vegetation Science 9, 469-476. 

Flannigan M.D., Stocks B.J., Wotton B.M., 2000. Climate change and forest 

fires. The Science of the Total Environment 262, 221-229. 

Flannigan M., Campbell I., Wotton M., Carcaillet C., Richard P., Bergeron Y., 

2001. Future fire in Canada’s boreal forest: Paleoecology results and 

general circulation model - regional climate model simulations. 

Canadian Journal of Forest Research 31, 854-864. 

Flannigan M., Logan K., Amiro B., Skinner W., Stocks B., 2005. Future area


burned in Canada. Climatic Change 72, 1-16. doi: 10.1007/s10584-005- 

5935-y 

Flannigan M.D., Amiro B.D., Logan K.A., Stocks B.J., Wotton B.M., 2006. 

Forest Fires and Climate Change in the 21st Century. Mitigation and 

Adaptation Strategies for Global Change 11, 847-859. doi: 

10.1007/s11027-005-9020-7 

Flannigan M.D., Stocks B.J., Turetsky M.R., Wotton B.M., 2009. Impact of 

climate change on fire activity and fire management in the circumboreal 

forest. Global Change Biology 15, 549-560. doi: 10.1111/j.1365- 

2486.2008. 01660.x 

IPCC, (2007) ‘Climate Change 2007: Synthesis Report. Contribution of 

Working Groups I, II and III to the Fourth Assessment Report of the 

Intergovernmental Panel on Climate Change [Core Writing Team, 

Pachauri, R.K and Reisinger, A. (eds.)]. IPCC, (Geneva, Switzerland) 

Krawchuk M.A., Cumming S.G., Flannigan M.D., Wein R.W., 2006. Biotic and 

abiotic regulation of lightning fire initiation in the mixedwood boreal 

forest. Ecology 87, 458-468. 

Krawchuk M.A., Cumming S.G., Flannigan M.D., 2009. Predicted changes in 

fire weather suggest increases in lightning fire initiation and future area 

burned in the mixedwood boreal forest. Climatic Change 92. doi: 

10.1007/s10584-008-9460-7 

Malevsky-Malevich S.P., Molkentin E.K., Nadyozhina E.D., Shklyarevich O.B., 

2008. An assessment of potential change in wildfire activity in the 

Russian boreal forest zone induced by climate warming during the twenty-first 

century. Climatic Change 86, 463-474. doi: 10.1007/s10584-007- 

9295-7 

Martell D.L., Otukol S., Stocks B.J., 1987. A logistic model for predicting 

daily people-caused forest fire occurrence in Ontario. Canadian Journal 

of Forest Research 17, 394-401. 

McCoy V.M., Burn C.R., 2005. Potential Alteration by Climate Change of the 

Forest-Fire Regime in the Boreal Forest of Central Yukon Territory. Arctic 

58, 276-285. 

Soja A.J., Tchebakova N.M., French N.H.F., Flannigan M.D., Shugart H.H., 

Stocks B.J., Sukhinin A.I., Parfenova E.I., Chapin F.S., III, Stackhouse 

P.W. Jr., 2007. Climate-induced boreal forest change: predictions versus 

current observations. Global and Planetary Change 56, 274-296. doi: 

10.1016/jgloplacha.2006.07.028 

Stocks B.J., 1993. Global Warming and Forest Fires in Canada. Forestry 

Chronicle 69, 290-293. 

Stocks B.J., Fosberg M.A., Lynham T.J., Mearns L., Wotton B.M., Yang Q., 

Jin J.-Z., Lawrence K., Hartley G.R., Mason J.A., McKenney D.W., 1998. 

Climate change and forest fire potential in Russian and Canadian boreal 

forests. Climatic Change 38, 1-13. 

Tymstra C., Flannigan M.D., Armitage O.B., Logan K., 2007. Impact of climate 

change on area burned in Alberta’s boreal forest. International 

Journal of Wildland Fire 16, 153-160. doi: 10.1071/WF06084

32 


Wotton B.M., Martell D.L., Logan K.A., 2003. Climate Change and People- 

Caused Forest Fire Occurrence in Ontario. Climatic Change 60, 275-295. 

Wotton B.M., Martell D.L., 2005. A lightning fire occurrence model for 

Ontario. Canadian Journal of Forest Research 35, 1389-1401. doi: 

10.1139/X05-071.

PROJECTING FUTURE BURNT AREA IN THE EU-MEDITERRANEAN 

COUNTRIES UNDER IPCC SRES A2/B2 CLIMATE CHANGE SCENARIOS 

Abstract: The goal of this work is to use the results of statistical modelling 

of historical (1985-2004) monthly burnt areas in European Mediterranean 

countries, as a function of monthly weather data and derived fire danger 

indexes, and to analyse potential trends under present and future climate 

conditions. Meteorological variables were extracted from the ECMWF, and 

the FWI system components were computed from 1961 until 2004. Monthly 

averages of the indexes were used as explanatory variables in a stepwise 

multiple linear regression analysis, to estimate the monthly burnt areas in 

each of the five most affected Mediterranean countries of Europe. 

Significant regression equations and satisfactory coefficient of determinations 

were found, although with remarkable differences among countries. 

Two IPCC SRES climate change scenarios (A2/B2) were simulated using the 

the regional climate model HIRHAM. The multiple regression models were 

than applied to the A2/B2 scenarios results to predict the potential burnt 

areas in each country. The models pointed out tangible changes in the 

potential burnt area extent for the future scenarios compared to the actual 

conditions. 


G. Amatulli, A. Camia & J. San-Miguel 

Joint Research Centre of the European Commission, 

Institute for Environment and Sustainability, Ispra (VA), Italy 

giuseppe.amatulli@jrc.it 

Forest fires have become a main environmental issue in most part of the 

world ecosystems. This phenomenon has direct consequences not only at 

the local scale, but also at the global one, representing a responsible factor 

for many phenomena such as climate change and global warming. The 

impact of climate change and extreme weather events on forest fires have 

received increased attention in the last few years. It is widely recognized 

that weather conditions play a key role in revealing extreme fire potential. 

It is therefore of great interest to analyse projected changes in fire risk 

under climate change scenarios. To date, several studies have appeared that 

made an assessment of forest fire risk trends under climate change scenar- 

33

34 


ios. Most of them there were conducted at world level and some specific at 

USA level. In Europe several studies have been conducted at country level 

but no one considering the overall European Mediterranean region (EU- 

Med) (Portugal, Spain, southern France, Italy and Greece). The goal of this 

work is to use the results of statistical modelling of historical (1985-2004) 

monthly burnt areas in EU-Med, as a function of monthly weather data and 

derived fire danger indexes, and to analyse potential trends under present 

and future climate conditions. 

2 - Dataset 

2.1 - EU fire statistic 

The fire data were extracted from the European Forest Fire Information 

System (EFFIS) EU fire database. This database is a collection of EU member 

states of individual fire event records. For the five EU-Med countries the 

complete available time window expands from 1985 until 2004. The fire 

events are recorded at NUTS3 level (provinces) and aggregated at monthly 

time frame. 

2.2 - Weather dataset 

Daily weather data, temperature, total precipitation, air relative humidity 

and wind speed, were used to calculate Canadian Fire Weather Index (FWI) 

system components. The FWI system is a system of daily meteorologicalbased 

indexes designed for the Canadian forest but amply used and tested 

around the world, and also in EU. The success of this system is based on 

his easy calculation but also in the presence of sub-indexes which summarize 

the fuel moisture. It includes a litter moisture model (Fine Fuel 

Moisture Code - FFMC), an upper organic moisture model (Duff Moisture 

Code - DMC) and a deep organic or soil moisture model (Drought Code - DC). 

These codes directly influence the fire behaviour expressed by four indexes 

(Initial Spread Index - ISI, Build Up Index - BUI, Fire Weather Index - FWI, 

Daily Severity Rating - DSR). 

Meteorological variables were extracted from the European Centre for 

Medium Range Weather Forecast (ECMWF) ERA-40 (40 Years Re-Analysis) 

dataset to compute daily value of fire danger in Europe, with a resolution 

of 1.25° for the period 1961 to 2002. The time series has been extended 

until 2004 with data from the MARS archive of the same ECMWF. The ECMWF 

dataset was manly divided in two time frames. The 1961-1990 was considered 

as control period and the 1985-2004 was used to built up the statistical 

model. 

Instead, the future weather conditions were simulated by the regional climate 

simulation model HIRHAM. The HIRHAM (Christensen et al., 1996) is

Projecting future burnt area in the EU-Mediterranean countries under IPCC SRES A2/B2 climate change scenarios 35 

a regional climate model (RCM) based on a subset of the HIRLAM (Undén 

et al., 2002) and ECHAM models (Roeckner et al., 2003). It combines the 

dynamics of the former model with the physical parametrization schemes of 

the latter. The RCM HIRHAM simulation was performed in the framework of 

PRUDENCE European project (http://prudence.dmi.dk/) by the Danish 

Meteorological Institute. In this project the model was run for two time 

slices: 30-year period corresponding to 1961-1990, called control, and a 

scenario run corresponding to 2071-2100. The future scenario was running 

in according to the A2 and B2 scenarios of the Intergovernmental Pane on 

Climate Change. In the experiments tree horizontal resolutions were adopted, 

12 - 25 - 50 km. For this study the 50 km was used in order to be easily 

combined with the ERA-40 data. Concerning the internal time frame of 

the experiments, each year is composed by 12 month of 30 days, which correspond 

to 360 days for a year. This dose not allows a day-to-day comparison 

with the historical observation. Based on this model, the future prediction 

for the A2 and B2 scenarios correspond to an increment of over the 

EU-Med land surface of +3.87 and +2.46 °C, respectively. 

3 - Methodology 

The FWI code was run at cell level for a daily computation of the sub-indexes 

for the following time frame dataset 

• ERA-40 1961-1990 (control); 

• ERA-40 1985-2004 (observations); 

• RCM-HIRHAM 1961-1990 (control); 

• RCM-HIRHAM A2 2071-2100 (Future Scenario); 

• RCM-HIRHAM B2 2071-2100 (Future Scenario). 

The input parameters were aggregate at time level (monthly average) and 

spatial level (NUTS3 - Country - EU-Med) in order to be crossed with the 

monthly burnt area. The monthly average calculation was computed at cell 

level. Instead, the spatial level aggregation was done at NUTS3 level by 

averaging the monthly value considering the area covered by each cell 

(weighted average). By this operation each NUTS3 has his own monthly 

average condition for each year of observation and can be crossed with the 

monthly burnt area obtained by the EU fire database. A similar operation 

(weighted average) was done to aggregate monthly data at country level 

and at EU-Med level. After, monthly averages of the indexes (240 observations) 

were used as explanatory variables in a stepwise multiple linear 

regression analysis, to estimate the monthly burnt areas in each country 

and at EU-Med level. 

The statistical model can be summarized by the following formula: 

BurntArea sum = aFFMC avg + bDMC avg + cDC avg + dISI avg + f BUI avg + gDSR avg + h 

Two models were computed to summarized Summer-Autumn (May to

36 


November - 140 observations) and Winter-Spring (December to April - 100 

observations) burnt area trends. 

To correct for the bias in the climate change model, the projected future 

fire danger conditions were estimated as the difference between the scenario 

and the control period of the RCM-HIRHAM experiment, and than 

added to the control period conditions assessed with the ERA-40 dataset. 

The multiple regression models were than applied to the A2/B2 scenarios 

results to predict the potential burnt areas in each country. 

4 - Results 

The stepwise multiple linear regression analysis, performed for all the countries 

during the Summer-Autumn period, produces an adjusted r 2 over 0.70. 

On the contrary for Winter-Spring period r 2 was ranging between 0.45-0.69. 

The stepwise selection points out common sub-indexes (DC, ISI) for the 

Summer-Autumn models, on the contrary remarkable differences were presented 

for the selection of the sub-indexes (FFMC, DC, ISI) concerning the 

Winter-Spring models (Table 1). Monthly average prediction and observation 

is depicted in Figure 1. A well noted model fitting is visible for the 

Summer-Autumn period. On the contrary the Winter-Spring period shows a 

not linearity trend. The burnt prediction for the two climate change scenarios 

are extremely evident in August (Figure 1). In Figure 2 is reported 

predicted and observed area trend for the different time periods considering 

the months variability of the ISI and DC indexes. 

5 - Conclusion 

Sub-indexes future projection point out a clear increase of their average 

values al EU-Med level and also at country level. Fire weather and burnt 

area are strongly linked by a robust correlation. Nonetheless, alternative 

no-standard regression models (MARS, GAM) should be tested to smooth 

the highest values of the identified exponential function. More climate 

change simulation models should be tested to come out with an ensemble 

model able to drive internal variability parameters. The understanding of 

future fire activity and consequent burnt area can help the European member 

states to adapt and develop mitigation plans to prevent dramatic fire 

events. 

References 

Christensen, J.H., Christensen, O.B., Lopez, P., Van Meijgaard, E., Botzet, 

M., 1996. The HIRHAM4 Regional Atmospheric Climate Model. DMI 

Scientific Report 96-4, Danish Meteorological Institute, Copenhagen,

Projecting future burnt area in the EU-Mediterranean countries under IPCC SRES A2/B2 climate change scenarios 37 

Denmark 

Unden, P., L. Rontu, H. Jarvinen, P. Lynch, J. Calvo, et al. The HIRLAM-5 

Scientific documentation, December 2002. Available at http:// 

hirlam.knmi.nl 

Erich Roeckner, Klaus Arpe, Lennart Bengtsson, Michael Christoph, Martin 

Claussen, Lydia Dümenil, Monika Esch, Marco Giorgetta, Ulrich Schlese, 

Uwe Schulzweida (1996): The atmospheric general circulation model 

ECHAM-4: Model description and simulation of present-day climate. MPI 

Rep. 218, Max-Planck-Institute for Meteorology, Hamburg, Germany 

Table 1 - Stepwise multiple linear regression models for the EU-med. 

Figure 1 - Monthly average 

trend of the observed 

and predicted burnt area.

38 


Figure 2 - Months variability of ISI and DC indexes and their predicted burnt area.

MULTISCALE CHARACTERIZATION OF SPATIAL PATTERN OVER 1998-2006 

WILDLAND FIRE EVENTS IN THE BASILICATA REGION 

Abstract: In this paper, we analyzed the spatial and temporal patterns of 

fire occurrence in the Basilicata Region during 1998-2006. Spatial indicators 

of autocorrelations, such as Moran’s Index (I) and the Getis & Ord’s 

Index (G) were used to characterize the inter-annual temporal/spatial fire 

distribution and to understand if it is random (autocorrelation equal to 

zero), uniform (negative autocorrelation) or clustered (positive autocorrelation). 


M. Danese 1 , B. Murgante 1 

University of Basilicata, Potenza, Italy 

A. Lanorte 2 , R. Lasaponara 2 

CNR-IMAA, Tito Scalo (PZ), Italy 

a.lanorte@imaa.cnr.it 

The spatial and temporal pattern of wildfires is a key point in the study of 

the dynamics of fire disturbance. It has important implications for vegetation 

patterns and resource management strategies. Nowadays, the growing 

availability of forest fire data archive with detailed information on individual 

fires (where, when, how fire occurs) can allow us to characterize and 

understand (i) how the fire events are distributed across different areas, (ii) 

how the fire events are clustered across space. Thus a number of papers 

have been focused on the characterization of the temporal pattern and 

temporal correlation focusing on the “intra-annual” clusterization of wildfires 

(see for example Tuia et al., 2008). Considerable uncertainty remains 

regarding the characterization of inter-annual variability of fire incidence 

as well as relation between fire activity variability and climate dynamics. 

In this paper, we focus on the inter-annual clusterization of wildfires in the 

Basilicata Region considering the fire events occurred during the 1998- 

2006 time window. Moran’s Index (I) and the Getis & Ord’s Index (G) were 

used for our analysis. 

39

40 


2 - Methods 

Spatial analysis was used in order to characterize the fire spatial distribution 

and to understand if it is random (autocorrelation equal to zero), uniform 

(negative autocorrelation) or clustered (positive autocorrelation). Two 

indicators of spatial autocorrelation were used, Moran’s Index (I) and the 

Getis & Ord’s Index (G). 

Moran’s I (Moran, 1948) is a global indicator of spatial autocorrelation, so 

it measure the degree of autocorrelation of a point pattern. It is defined 

as: 

where N is the number of events, Xi ed Xj are intensity values taken by, 

respectively, the point i and the point j and is the mean of the considered 

variable. Wij is one element of the weights matrix, that contains spatial 

relationships between events because each element of the matrix represents 

the spatial weight of each single event, defined according a fixed 

proximity criterion. One of the most common criterion is the Fixed Distance 

Band Method: if an event is included in the fixed distance band it will be 

considered in the calculation, otherwise it will be excluded. 

Moran’s I is included between the interval [-1; 1]. If the index is less then 

0 the point pattern is negatively autocorrelated; if it converges to 0, the 

spatial distribution is random; if I is greater than 0 there are some cluster 

in the distribution. 

Getis & Ord’s G is instead a local indicator of spatial association. It allows 

us to understand where clusters are located. It is defined by the equation 

(Getis and Ord, 2001): 

where N is the number of events, Xi and Xj are intensity values taken by, 

the point i and the point j, respectively, is the mean of the considered variable, 

wi(d) is one element of the weight matrix. 

Elements with low intensity are autocorrelated if they have also a low G 

value; in the same way, if they have a high intensity, they should have a 

high G, to be clustered.

Multiscale characterization of spatial pattern over 1998-2006 wildland fire events in the Basilicata Region 41 

3 - Study case 

The analysis was performed in Basilicata (9,992 km 2 ) a region in the 

Southern of Italy. It is a mountainous zone with around 47% of in mountains, 

45% is hilly and finally 8% is flat. The climate is variable and strongly 

influenced by three coastlines (Adriatic, Ionian and Tyrrhenian) as well 

as by the complexity of the region’s physical features. The climate is continental 

in the mountains and Mediterranean along the coasts. 

Approximately 35% of the total surface is covered with forest vegetation 

(mainly Oak woods, Beech woods, Mediterranean maquis, Mixed broadleaf 

and/or coniferous woods, Mediterranean scrubs). Prairies, bushes and cultivated 

soil cover approximately 45% of the territory. Between 2001 and 

2008 in the Basilicata Region fire affected more than 20.000 ha (forest and 

non-forest) with 1900 fires generally less than to 10 ha. 

In order to study the spatial autocorrelation of the fire dataset, firstly the 

Moran’s I was applied to the fire occurred for each year. The intensity associated 

to the point pattern was the date when the event occurred. To find 

the ‘best’ fixed distance band, that is the distance where the Moran’s I 

shows the maximum degree of clusterization, an iterative method was used: 

different distance values, varying from a minimum value depending on the 

nearest neighbours of centroids and a maximum value chosen depending on 

the Moran’s I, were introduced in the calculation. Then, we selected the 

distance value which maximizes Moran’s I and at the same time has the 

more significant score values. 

This method was repeated for the dataset of each year (tab. 1), finding that 

only in 1999, 2002, 2003, 2005 the point pattern is strongly clustered, 

while in the others years it is weakly clustered. The cluster obtained identified 

areas with homogeneous fire dynamics and this allows us to better 

characterize and understand fire activity variability. 

Year N of fires Distance Band (m) Moran’s I 

1998 235 2200 0,122375 

1999 132 900 0, 592405 

2000 376 400 0,249357 

2001 300 400 0,184796 

2002 138 400 1 

2003 264 300 0,549425 

2004 205 500 0,145130 

2005 212 200 0,453211 

2006 146 400 0,180571 

Table 1 - Distance bands and Moran’s I found for each year.

42 


The distance bands so found were introduced in Getis and Ord’s G, in order 

to find where clusters were localized. Getis and Ord’s index value was classified 

according to date of occurrence of the event as reported in figure 1 

and in table 2, which shows the example of the 2000 year. 

Figure 1 shows fire location and “seasonal” occurrence for the years 1999, 

2002, 2003, 2005 which exhibited the higher time and spatial clusterization. 

The spatial distribution of fires depends on different types of land use 

and also on different interannual climate variability, which lead to different 

fire susceptibility. It is quite interesting the fact that: (i) for the 1999 

year fires mainly occurred in the summer, as usual for the Mediterranean 

ecosystems (ii) for 2002 fire occurred, as usual, in summer but also during 

winter, and finally (iii) for both 2003 and 2005 years fires happened during 

the spring, summer and winter season. This behavior shows a significant 

variability in the fire “seasonality” for the Basilicata Region. 

Season Day of the year when the event occurred Getis and Ord’s G 

winter 1 79 -4,02502 -2,01534 

356 365 1,89023 2,83819 

springer 80 171 -2,01534 -0,00567 

summer 172 294 -0,00567 0,94228 

autumn 295 355 0,94228 1,89023 

Table 2 - Classification of Getis and Ord’s G for dataset of 2000 year. 

4 - Final remarks 

The aim of this study was to characterize the interannual fire activity in the 

Basilicata Region. Areas of homogeneous fire dynamics were identified with 

cluster analysis. The cluster obtained are physically meaningful and this 

allows us to better characterize and understand fire activity variability. This 

provides useful information to investigate the relative importance of different 

factors influencing short-term variability in fire occurrence at different 

scales and to better understand the contribution of human activities 

and climate variability to the fire incidence. Our results show that rainfall 

variability significantly influences the fire occurrence and areas with different 

types of land use react in a different way to interannual climate variability, 

leading to different fire susceptibility depending on the land use 

type.

Multiscale characterization of spatial pattern over 1998-2006 wildland fire events in the Basilicata Region 43 

Figure 1 - Fire location for 1999, 2002, 2003 and 2005 years which exhibited the higher time 

and space clusterization. 

References 

Moran, P., 1948, The interpretation of statistical maps, Journal of the Royal 

Statistical Society, 10, 243-251. 

Ord, J.K., Getis A., 2001, Testing for Local Spatial Autocorrelation in the 

Presence of Global Autocorrelation, Journal of Regional Science, 41, 411- 

432 

Tuia, D., Lasaponara, R., Telesca, L., Kanevski, M., Emergence of space-clustering 

temporal patterns in forest-fire sequences, in Physica A, 387 

(2008), pp. 3271-3280.

Abstract: Forest fires and other wildfires represent a major threat and cause 

severe damage at environmental, ecological, human, and economical levels 

in the world. The number of fires and the size of the area burnt have 

increased dramatically in Lebanon over the last decades; they are becoming 

larger and more severe than in past fire episodes. Predicting fire hazard 

and risk in the fire-prone ecosystems is critical to the mitigation of the 

effects of fires and to the resulting impact that fires have on the socio-economic 

fabric. In such context, a GIS fire risk model was developed in this 

study for predicting forest fires in Nahr Ibrahim (North Lebanon). It combines 

various influencing parameters, i.e. land cover/use, slope angle and 

aspect, proximity to road network and urban expansion that were extracted 

from satellite imageries and DEMs. The main outcome was the production 

of a fire risk map (scale 1:20,000). The model used seems to be applicable 

to other areas of Lebanon, constituting a tool for land use planning 

and sustainable management. 


FOREST FIRE RISK ASSESSMENT IN IBRAHIM RIVER 

WATERSHED-LEBANON 

C. Abdallah 

National Council for Scientific Research/Remote Sensing Center, 

Beirut, Lebanon 

chadi@cnrs.edu.lb 

Forest fires are considered to be a potential hazard with physical, biological, 

ecological and environmental consequences. Wildfires have become 

ever more destructive throughout the world and the prospects are unfortunately 

that this trend will continue. Numerous GIS fire risk models have 

been proposed (e.g., Mitri and Gitas, 2004) during the last decades that 

can be grouped into: (1) empirical models such as the Canadian Forest Fire 

Behavior Prediction System (cited by ICONA, 1992) and McArthur’s model 

(Weise and Biging, 1997) predicting more probable fire behavior from laboratory 

and outdoor experimental fire, or historical fires; (2) semi-empirical 

models such as the National Fire Danger Rating System, the RERAP Rare 

Event Risk Assessment Process, etc. based on the assumption that the energy 

transferred to the unburned fuel is proportional to the energy released 

45

46 


by the combustion of the fuel; and (3) physical models based on certain 

theoretical and analytical principles having the potential to accurately predict 

the parameters of interest over a broader range of input variables than 

empirically based models. The problem related to the utilization of empirical 

models is their prediction of forest fires according to certain conceptual 

theories, and it is well known that the occurrence of fires shows heterogeneities 

from one location to another. The semi-empirical models must be 

fitted from laboratory fire experimental results; they are more accurate than 

empirical models since incorporating all influencing fire risk parameters, 

however, the main drawback is that the data to those parameters are rarely 

found and roughly precise. 

Some fire risk indexes were developed also worldwide based on vegetation 

indexes like the NDVI, or on Land Surface Temperature (LST) obtained using 

thermal channels. Other climate variables (e.g. wind velocity, humidity, 

etc.) allow a better production of fire indexes. However, the integration of 

all influencing fire variables in a significant risk map is not an easy task, 

and was considered as one of the most important research topics during the 

last years. The lack of data and the difficulty in accessing certain terrain 

variables further underlines the importance of satellite data as the only 

means of generating fire risk maps. In such context, this paper aims to 

develop a forest fire risk model based on a semi-empirical method applied 

on real time field data that would reduce the variability of experimental 

laboratory work. This model is applied in a representative area of Lebanon 

(Nahr Ibrahim watershed). It can be easily extrapolated to other 

Mediterranean areas sharing similar environmental conditions and/or lack 

of detailed spatial data. 

2 - GIS modelling of forest fires 

2.1 - GIS factor collection and preparation 

The proposed model is based on the integration of several influencing fire 

risk factors under a GIS environment shown below. 

a. Land cover/use: Land cover/use is a significant dynamic parameter 

affecting the fire process. Various plant species have a different sensitivity 

towards fire. Each fuel or vegetal type burns more or less quickly than the 

others. The extraction of land cover/use categories was performed through 

visual interpretation of pan-sharpen Landsat TM – IRS imagery (6 m) referring 

to Land Cover methodology adopted by CORINE program. The diverse 

land cover/use modes were classified in function of their sensitivity to fires 

depending on field observations and consultation of several studies (e.g., 

Bonazoutas et al., 2006). 

b. Slope gradient and aspect: Slope gradient and aspect were calculated 

from the constructed digital elevation model (DEM) generated for the study 

area with a planimetric resolution of 10 m. Slope affects both the rate and

Forest fire risk assessment in Ibrahim river watershed-Lebanon 47 

direction of the fire spread. Considering the histogram of equalization 

between distribution of slope gradient and the corresponding number of 

pixels, the slope gradient was divided into ten classes (ranging between 

less 5° and 90°) with various fire sensitivities ranging from very low sensitivity 

on the bottom of the hills to very high at the peaks. Aspect is useful 

for visualizing the amount of solar illumination the vegetation receives, 

which influences both the type of fuels and their moisture condition. 

Aspect is divided into the eight major directions plus the non-oriented flat 

areas, each one having a different sensitivity to fires. 

c. Urban expansion: The urban expansion provides an important indication 

of the presence of people affected by forest fire risks. It is commonly 

accepted that the denser the people in a given area, the higher the forest 

fire risk and vice versa. The automatic extraction of the urban expansion 

was performed from the 1:20,000 land cover/use map (LNCSR-LMoA, 2002). 

A buffer zone of 500 m above the urban area was created using the 

Euclidian distance, and ten classes were considered. 

d. Proximity to road network: Roads are usually sites inducing fires (e.g., 

cigarettes). For this reason, roads were included in this study, and extracted 

from IKONOS imageries (1 m) through visual interpretation. Thus, a 

buffer zone of 50 m above the road (maximum height of a talus cut created 

by the construction of a road) was created. 

2.2 - GIS data analysis and manipulation 

Not all the considered factors have the same effect on fire, but the estimation 

of their weights is not an easy task. Therefore, a regional parameterization 

was done in giving weight to all fire parameters using a 

weight/rate approach consisting of applying primary and secondary-level 

weights. The primary-level weights are rule-based in that ratings are given 

to each class of a parameter on the basis of logical referred system. The 

secondary-level (factor) weights are, however, opinion-based scores, which 

determine the degree of tradeoff of one parameter/against another. This 

leads to the following percentages of each parameter effect on the fire risk 

occurrence as follows: land cover/use (50%), slope gradient (10%), slope 

aspect (15%), urban expansion (10%), and proximity to road network 

(15%). 

3 - Discussion and conclusion 

3.1 - Production of forest fire risk map 

In the produced fire risk map with eight classes (Figure 1), class 8 (high 

fire risk) covers the largest area (34%), and is mostly distributed in the 

centre of the studied region and on the side of the mountain hills. As one

48 


moves towards the crests, the intensity of fire is decreasing due to the 

absence of trees. 

3.2 - Advantages and problems of the fire risk built model 

The established model with 5 factors has defined a map of fire intesity with 

eight classes. Such a map was unavailable in the study area, although at 

this stage the map still depends on relatively subjective procedures. The 

map represents the result of modelling from available knowledge and data, 

can serve needs of different researchers, and is important for policies and 

socio-economic decision-makers. It can prioritize the anti-fire protection 

measures according to the obtained fire risk level. The model has certain 

advantages over fully-automated fire risk models which are too-data hungry 

to be applied at a regional scale. This model can be easily extrapolated 

to all the country if the functional capacities of GIS are used, because 

they allow model integration with additional basic and factorial data and 

code modification in order to analyse the data. The factorial morphological 

maps (slope gradient and slope aspect) are more persistent with time than 

the other used maps, i.e. land cover/use, urban expansion and proximity to 

road network which must be updated and can change drastically through 

years due to human activities. However, many difficulties have been 

encountered. The use of one-lumped value for land cover/use in fire risk 

modelling is sometimes an oversimplification of the real situation in which 

fire shows complex spatial and temporal patterns. Future study should focus 

on parameters [e.g., fuel loading (kg/m 2 ), surface to volume ratio, fuel 

moisture content, fuel bed depth, particle density (kg/m 3 ), etc.] and methods 

to capture this variability and to incorporate data in fire risk models. 

References 

Bonazountas, M., Kallidromitou, D., Kassomenos, P., 2006. A decision support 

system for managing forest fire casualties. Journal of Environmental 

Management, 84, 412-418. 

ICONA, 1992. Los incendios forestales en Espana durante 1991. Instituto 

Nacional para la Conversacion de la Naturaleza, MAPA, Madrid. 

LNCSR-LMoA, 2002. Land cover/use maps of Lebanon at 1:20,000 scale. 

Lebanese National Council for Scientific Research-Lebanese Ministry of 

Agriculture. 

Mitri, G.H., Gitas, I.Z., 2004. A semi-automated object oriented model for 

burned area mapping in the Mediterranean region using Landsat-TM 

imagery. International Journal of Wildland Fire, 13, 367-376. 

Weise, D.R, Biging, G.S., 1997. A qualitative comparison of fire spread models 

incorporating wind and slope effects. Forest Science, 43(2), 170-180.

(a) Forest fire risk map. 

(b) Gis model design. 

Forest fire risk assessment in Ibrahim river watershed-Lebanon 49

50 


(c) Study area location 

Figure 1 - Forest fire management in ibrahim watershed.

A FOREST FIRE HAZARD BASED ON THE ESTIMATION OF TOURIST HOT 

SPOT ACTIVITIES IN AUSTRIA 

N. Arndt 1 , A. Arpaci 1 , H. Gossow 1 , P. Ruiz Rodrigo 2 , H. Vacik 1 

1 University of Applied Life Sciences Vienna, Austria 

2 Polytechnic University of Valencia, Spain 


Austria is a predominantly alpine Central European Country with a size of 

83.871 km 2 and a forest cover of 47,2%. The main tree species is Norway 

Spruce (Picea abies L. K.) with a total share of 53,7% (ÖWI, 2002). Austrian 

forests do not fulfil the characteristics of fire prone ecosystems, nor have 

they seriously been fire-impacted so far. Due to the debate on probable climate 

change it is hypothesized, that the risk of forest fires will increase in 

the coming decades (Flannigan et al., 2005). Forest fires are a result of 

complex interactions between ecological factors such as weather, fuel type, 

forest structure, topography and socio-economic factors such as population 

density, infrastructure and tourism activities (Chuvieco et al., 2009; 

Kalabokidis et al., 2002; Marchi et al., 2006). However, in Austria more than 

90% of forest fires in the course of the last 50 years are a result of human 

interference. Therefore it is necessary to identify forest fire “hot spots” in 

Austria in order to develop basic information for an emergency strategy for 

Austrian fire brigades. In this context a forest fire hazard model will be 

developed combining a socio-economic risk model with drought indices and 

a fuel classification approach within the frame of the Austrian Forest Fire 

Research Initiative (AFFRI). This contribution describes a part of the fire 

hazard approach determining the role of tourist activities for the ignition 

risk of forest fires in Austria. 


The idea is to estimate the effect of tourist activities along touristic infrastructures 

on fire ignition by mapping the seasonality, frequency, type and 

intensity of tourism activities. 

The bases of our research are records of forest fires from Austrian fire 

brigades, statistics on tourism, namely overnight stays per district and sea- 

51

52 


son for the period between 2002 and 2008 in combination with the information 

on existing touristic infrastructure, namely hiking trails, mountain 

huts and cable cars. Several data sources were used to fix the location of 

cable cars, touristic huts, hiking trails and forest fires in a map with ArcGIS. 

Additionally we created maps for the mean number of overnight stays per 

district and a map showing the share of forested area (km 2 ) on district level 

within the district boundaries. 

Figure 1 - Touristic infrastructure of Austria with forest fires (Statistik Austria). 

Figure 2 - Mean overnight stays of Austria (district level) 2002-2008 (Statistik Austria).

A forest fire hazard based on the estimation of tourist hot spot activities in Austria 53 

The risk rating assumes that the fire risk is very high close to touristic 

infrastructure and decreases with distance from the infrastructure together 

with a decrease of intensity. 

For the estimation of the influence of tourist activity on forest fires in a 

district a rating has been done in merging the touristic infrastructure with 

the tourist intensity, the percentage of the forest cover and the area of a 

district with the following formula: 

∑ζ i = (λ*ζ ic /DS + λ*ζ ihut /DS + λ*ζ ihi /DS) * Ø overnight * FC/DS 

where λ represents the weight for touristic infrastructure per district, ζic represents the weight rating for the cable cars per district, ζihut the weight 

for the huts in a district and ζihi the weight for the hiking trails in a district. 

Øovernight symbolizes the mean number of overnight stays per district; 

FC/DC indicates the percentage of Forest Cover in a district. 

We defined different weights for the touristic infrastructures considered - 

namely cable cars, huts and hiking trails - and calculated four different scenarios 

in order to identify the relative touristic intensity in the individual 

districts. In scenario I we assigned an equal weight for the touristic infrastructure. 

In scenario II a weight of 10% was assumed for the cable cars, 

for the huts a weight of 40% and for the hiking trails a weight of 50%. In 

scenario III the weight of the cable cars was assumed to be 50%, the 

weight of the huts 40% and the weight of the hiking trails 10%. In scenario 

IV the weight assigned to the cable cars is 20%, the weight assigned 

to the huts is 40% and the weight assigned to the hiking trails is 40% 

respectively. Next we grouped the districts into five categories relative to 

the number of forest fires that occurred between 2002 and 2008 and 

opposed them to the calculated mean overall value as well as the mean 

minimum and maximum value of the calculated scenarios. 

N° forest fires Mean overall value Minimum Maximum 

0 6.085 2,5 91.692 

1 - 3 11.464 5 270.591 

4 - 6 17.336 21 305.961 

7 - 10 10.920 18 80.527 

11 - 15 38.904 44 252.198 

16 - 20 47.941 1.191 223.258 

21 - 30 57.018 1.108 192.332 

> 30 12.979 1.707 36.496 

Table 1 - Mean overall value, minimum and maximum for four different scenarios.

54 


3 - Results 

The districts around the big Austrian cities do not show a fire risk because 

of tourism. The reason for that is that forest fires are absent within the district 

boundaries of the cities. Exceptions are the districts of the towns of 

Villach with eight forest fires, Wiener Neustadt and Krems with 3 forest fire, 

Waidhofen an der Ybbs with 2 forest fires and Salzburg and Sankt Pölten 

with 1 forest fire. 

Touristic activities are found to influence fire risk in the provinces of Tyrol, 

Carinthia, Salzburg and Styria in all four scenarios significantly. In those 

provinces the highest density of touristic infrastructure as well as the highest 

number of overnight stays is found. Being popular tourist destinations 

with an ideal site development makes those provinces susceptible to a 

higher fire risk. 

Compared to the influence of tourism in the mountainous regions of Austria 

the influence of tourism on fire risk turns out to be fairly marginal in 

Vorarlberg, Upper Austria, Burgenland and Lower Austria. Even though the 

number of overnight stays is quite elevated in some parts of those 

provinces, tourism does not play a significant role for fire ignition, since 

the degree of site development with touristic infrastructure is relatively low 

compared to the mountainous regions of Austria. In spite of this the results 

of the districts of Wiener Neustadt and Neunkirchen in Lower Austria stand 

out notably. Here the influence of tourism on fire risk is relatively high. 

Additionally the importance of the area as a local recreation area for Wiener 

Neustadt and Baden increases fire risk remarkably in those two districts. 

4 - Conclusions 

In regions with a high density of touristic infrastructure fire risk is influenced 

by touristic activities to a great extent. Areas in the direct vicinity 

of cities did not proof to be susceptible to fire risk caused by tourism at 

all. Regions with a medium or low density of touristic infrastructure had a 

comparatively low fire risk connected to tourism. However the influence of 

touristic activities cannot be excluded completely. Another important factor 

needing further investigation in this context is the influence of local 

recreation on fire risk. 

References 

Chuvieco, E.; Aguado, I.; Yebra, M.; Nieto, H.; Salas, J.; Pilar Martín; Vilar, 

L.; Martínez, J.; Martín, S.; Ibarra, P.; de la Riva, J.; Baeza, J.; 

Rodríguez, F.; Molina, J.R.; Herrera, M.A.; Zamora, R.: Development of a 

framework for fire risk assessment using remote sensing and geographic

A forest fire hazard based on the estimation of tourist hot spot activities in Austria 55 

information system technologies, in: Ecological Modelling (2009); in 

press 

Flannigan, M.D.; Logan, K.A.; Amiro, B.D.; Skinner, W.R.; Stocks, B.J.: 

Future area burned in Canada, in: Climatic Change 72, 1 - 16 (2005) 

Kalabokidis, K.D., Konstantinidis, P., Vasilakos, C.: GIS analysis of physical 

and human impact on wildfire patterns, in: Forest fire research and wildland 

fire safety, IV International Conference on Forest Fire Research, 

Portugal, 2002 (Viegas, D.X. (ed.)) 

Marchi, E.; Tesi, E.; Brachetti Montorselli, N.; Bonora, L.; Checcacci, E.; 

Romano, M.: Forest fire prevention: developing an operational difficulty 

index in firefighting, in: V International Conference on forest fire 

research 2006 (Viegas, D.X., ed.)) 

Österreichische Waldinventur 2002 (Austrian Forest Inventory 2002): 

http://web.bfw.ac.at/i7/oewi.oewi0002

CLASSIFICATION OF SITE AND STAND CHARACTERISTICS BASED ON 

REMOTE SENSING DATA FOR THE DEVELOPMENT OF FUEL MODELS 

WITHIN A 3D GAP FOREST STAND MODEL 

A. Arpaci 1 , N. Arndt 1 , M.J. Lexer 1 , M. Matiuzzi 2 , M. Müller 1 , H. Vacik 1 

1 University of Natural resources and Applied Sciences, Vienna 

Department of Forest-and Soil Sciences - Institute of Silviculture 

2 University of Natural resources and Applied Sciences, Vienna 

Department of Landscapes, Spatial and Infrastructure sciences 

Institute of Surveying, Remote Sensing and Land information 

alexander.arpaci@boku.ac.at 


Austria is not known for its wild land fire history. But still in the last decade 

wild land fires have draw attention by forest and landscape managers due 

to their increasing numbers. With increasing temperatures and changing 

precipitations patterns landscape might become even more vulnerable to 

fire hazards. Hence fire has to be added to disturbance regimes which are 

considered as drivers of ecosystem processes. Especially forest stands as 

part of landscape patterns are driven by disturbance regimes like bark beetle 

diseases, wind storms and fire. Important processes in forest development 

like regeneration and inter species competition are therefore dependent 

on the prevailing disturbance regime (Flemming, 2000, Ulanova, 2000, 

Lundquist and Beatty, 2002). The Austrian Forest Fire Research Initiative 

(AFFRI) tries to contribute to that fact. With the development of a fire disturbance 

module for the forest stand /gap model PICUS (Lexer and 

Hönninger, 2001; Seidl, 2005) a forest management decision support tool 

will be designed. The currently most robust and used algorithm in fire modelling 

the Rothermel model (Rothermel, 1972) for fire spread and behaviour 

will be used to calculate the spread and behaviour of wild land fires under 

different conditions for Austrian fire prone forest types. In this context fuel 

models need to be developed, which allow an accurate prediction of forest 

fire behaviour under different climatic and site conditions. In this contribution 

we will present the process of the localization of Austrian forest 

stands which are threatened by a high risk of wild land fires and will therefore 

be selected for fuel sampling. 

2 - Material & methods 

The AFFRI project does benefit from previous research activities about 

Austrian fire history (Gossow et al., 2008) over the last decade which gives 

57

58 


an overview about the general situation, like when and where typically wild 

land fires occur in Austria. Additionally a comprehensive database was created, 

where information, provided by fire fighters and forest managers 

about fires in their region, is collected. Estimations about the exact coordinates 

of the fire out break and the uncertainties linked to that are established. 

These coordinates build the foundation for the fire risk mapping 

approach. They are mapped with their information attached in an attribute 

table and can be accessed via the fire ID. In a first step fire “hot spots” in 

Austrian forest ecosystem are localized. 

To describe similarities and frequencies in wild land fire incidents forest 

parameter collected by the national forest inventory campaign (ÖWI, 2002, 

Gabler & Schadauer, 2002) are used. The inventory data on forest parameters 

exist as a point grid with stored attributes per sampling point. For the 

region Tirol the inventory parameters where extrapolated (Mattiuzzi, 2008). 

Using a LANDSAT image the spectral 

reflectance from the inventory 

points where used as ground truth 

for a reflectance similarity classification. 

With this method parameters like 

forest type, crown cover, forest composition 

and stage where derived. 

Forest ground vegetation communities 

were also derived; here the thesis 

is that similar crown reflectance 

has similar ground vegetation communities. 

In Austria forest communities 

are classified according to the 

Picture 1 - Geo climate units, the region 

Tirol, forest fires and their error buffer. 

eco-regions in Killian, et al., 1993). 

These units define what kind of forest 

communities can be found under 

a certain range of altitude and 

aspect and other abiotic conditions. This eco-regions are an important 

information for the stratification of forest communities. 

For fire behaviour modelling or risk mapping slope and aspect play an 

important part (Han et al., 2003). Slope and aspect was derived from a DEM 

of the Alpine arc. All data sets were re-projected to WGS 84 UTM zone 32N. 

In order to indentify the most forest fire prone forest communities we wanted 

to analyze the forest and topographic characterisation near to the forest 

fires occurred and identify the highest frequencies in order to make a 

design for the fuel sampling. The uncertainties in the localisation of the 

forest fires lead to the approach of buffering the most likely site of a forest 

fire. Depending on the uncertainty in the localisation of the coordinates 

of each forest fire record the buffer was classified large (> 2 km) or small 

(< 100m). 

In the case study for the region Tirol we had to face the problem that sev-

Classification of site and stand characteristics based on remote sensing data 59 

eral information got lost when approaching them by zonal statistic. This 

Lost was due to the overlay from buffers and the large variance of possible 

combinations of forest topographic parameters in one buffer. 

To solve this problem we decided for a map algebra approach. Here every 

parameter is reclassified in a single raster layer. 

All raster layers were projected and have the same LANDSAT extent and resolution. 

To collect information of all rasters in one raster every raster got an 

index number and every class within each raster got a value within the range 

of this index number. The resulting layers were than added via map algebra. 

So a raster layer was created were every single pixel contains an index number 

and information from 5 layers collected in one raster image. This process 

was repeated twice to collect all parameters in two raster layers. 

Example:

60 


Raster layer 1 

Raster layer 2 

Pixel containing the value 2236634 

Where 2 = geoclimate unit (central alpine) 

2 = deep montane (900 – 1100 m) 

3 = crown closure (moderate) 

6 = vegetation type (gras) 

211= fire ID 

Pixel containing the value 2711634 

Where 2 = slope (20 - 30 °) 

7 = aspect (west) 

1 = deciduous (10%) 

1 = coniferous (10%) 

634 = fire ID 

These two obtained raster layers were then used to highlight patterns of 

similar forest and topographic parameters within the fire buffer. To do so, 

raster values and their frequencies had to be exported into a database system 

where the single parameters could be accessed via SQL language. With 

SQL querries it was possible to identify parameters which occurred more 

often than others. 

One problem arise in this scenario; in very heterogeneous topographic landscapes 

the size of a buffer directly influences the amount of pixels and 

hence the variance of the value in that buffer. So in very big buffer which 

exist when localisation was only very roughly possible information get lost 

due to no possible identification of trends. To solve that problem classes of 

buffer size were build and individually analyzed. It showed out that with 

smaller buffer (< 300m) trends were becoming more visible and for further 

analysis only buffers till 300 meter were considered. 

Diagram 1 - Vegetation 

composition in burned 

areas.

Classification of site and stand characteristics based on remote sensing data 61 

It was necessary to test if our results were fire related, or just being a sample 

of the whole raster layer. The test of composition of vegetation types 

in the whole LANDSAT image and the composition in the fire buffer showed 

that there is clearly trend to herbs and grass cover. 

With that test it seems likely that the extrapolation of surface cover derived 

from crown cover reflectance similarities has some verification. 

3 - Results 

To identify fire prone forest communities a decision tree was build were certain 

characteristics of forest communities were listed and compared to the 

trends in our analysis. The trends were the following: 

Forest stands consisting from mostly coniferous trees with a dense structure 

from 900 till 1700 meter on a south-facing exposition and grass or dry 

herb under-storey were more likely to be burned in Tirol. With the result of 

the frequency analysis it was possible to identify four forest communities 

which are more likely to be threatened by fire incidents because they occur 

in the area of the eco-region 1.2 and fulfil the above mentioned characterizations. 

In Tirol this forest communities are Larici - Piceetum, Picealuzulo 

nemerosae, Pinus sylvestris-erico pinetum, Picea - calamagrosti var. 

Picetum. 

Percent off Pixel with in fire buffer 

Elevation montane 65% 

crwon closure dense 66% 

Veg-type herbs/gras 79% 

slope 0-20° 62% 

aspect southfacing 42% 

coniferous 70-100% 42% 

4 - Conclusion & outlook 

The above listed method and the results can be used to identify and map 

forest stands with a higher fire risk. But risk identified with this method 

does not account the human aspect neither does it analyse the hazard scenario. 

The human aspect is part of another research approach in the AFFRI 

project. Hazard has to be analyzed with a broader approach in mind 

because risk and hazard have to be analysed individually (Allgoewer, 

Bachmann, 2001). For hazard rating accurate modelling of fire behaviour is 

a main factor. The next task will be the development of a Austrian fuel 

model, which allows to classify forest communities not from a static but 

dynamic perspective. This fuel models will be linked to other disturbance 

regimes as well. To approach fire and fuel models we will sample fuels not

62 


only along for the strata of the forest community but along a temporal gradient 

as well. Fuel accumulation should then be possible to be simulated 

with the forest model PICUS. From that perspective it would be possible to 

test forest management activities for its consequences from the fire perspective 

during succession simulations also. 

References 

Gossow et al., 2008; Gossow, H., Hafellner, R., Arndt, N.: More forest fires 

in the Austrian Alps - a real coming danger?, in: Borsdorf, A. et al., 

Managing Alpine Future, Proceedings 

Han et al., 2003; Han G.J., Keun H.R., Kwang H.C., Yeon K.Y, Statistics 

based Predictive Geo Spatial data mining: Forest Fire Hazardous Area 

mapping 

Killian, Müller, Starlinger 93; Killian W., Müller F., Starlinger F., Die 

Forstlichen Wuchsgebiete Österreichs, Eine Naturraumgliederung nach 

waldökologischen Gesichtspunkten; BWF - Berichte Schriftenreihe der 

Bundesforschungs - und Ausbildungszentrums für Wald, Naturgefahren 

und Landschaft, Wien, 2006; Nr. 82; 60 pp 

Lexer & Hönninger, 1998; M.J. Lexer, K. Hönninger; A modified 3D patch 

model for spatially explicit simulation of vegetation composition in heterogeneous 

landscapes; For. Ecol. Manage. 144 (2001) pp. 43 – 65 

Lundquist JE; Beatty JS., 2002. A method for characterizing and mimicking 

forest canopy gaps caused by different disturbances. Forest Science. 48: 

582-594. 

ÖWI, 2002, Gabler Schadauer; Gabler K., Schadauer K.; Methoden der Österreichischen 

Waldinventur 2000/02 - Grundlagen der Entwicklung, 

Design, Daten, Modelle, Auswertung und Fehlerrechnung; BWF- Berichte 

Schriftenreihe der Bundesforschungs- und Ausbildungszentrums für 

Wald, Naturgefahren und Landschaft, Wien, 2006; Nr. 135; 132 pp. 

Rothermel, 1972; Rothermel R.C., A mathematical model for predicting fire 

spread in wildland fuels; GTR-INT-115; Odgen, UT: USDA Forest Service, 

Intermountain Forest and Range Experiment Station. 

Seidl et al., 2005. Seidl R., Lexer M.J., Jäger D., Hönninger K., 2005. 

Evaluating the accuracy and generality of a hybrid patch model; Tree 

physiology 25 939-951. 

Ulanova, 2000. Ulanova N.G., The effects of windthrow on forests at different 

spatial scales: a review, For. Ecol. Manage. 135 (2000), pp. 155- 

167. 

Volney and Fleming, 2000. W.J.A. Volney and R.A. Fleming, Climate change 

and impacts of boreal forest insects, Agric. Ecosyst. Environ. 82 (2000), 

pp. 283-294. 

Allgower Bachmann, 2001, Allgoewer B., Bachmann A., A consistent wildland 

fire risk terminology is needed!; Fire Management Today; 64; pp. 

4.

FUEL MOISTURE CONTENT ESTIMATION: A LAND-SURFACE MODELLING 

APPROACH APPLIED TO AFRICAN SAVANNAS 

Abstract: The aim of this study is to reduce the uncertainty associated with 

modelling the fire regime in Africa, by providing a more robust estimation 

of fuel moisture content (FMC) through model simulation. By constraining 

model predicted FMC through the assimilation of remotely sensed land-surface 

temperature (LST) and Normalised Difference Vegetation Index (NDVI) 

data into a leading land-surface model, the findings presented here show 

that a combination of satellite data and biophysical modelling provides a 

viable option to adequately predict FMC over continental scales at high 

temporal resolution. 


D. Ghent 1 , A. Spessa 2 

1 Department of Geography, University of Leicester, 

Leicester, UK 

2 Walker Institute for Climate System Research, Department of Meteorology, 

University of Reading, UK 

Despite the importance of fire to the global climate system, in terms of 

emissions from biomass burning, ecosystem structure and function, and 

changes to surface albedo, current land-surface models do not adequately 

estimate key variables affecting fire ignition and propagation. FMC is considered 

one of the most important of these variables (Chuvieco et al., 

2004). However, the complexity of plant-water interactions, and the variability 

associated with short-term climate changes, means it is one of the 

most difficult fire variables to quantify and predict. Previous studies 

(Sandholt et al., 2002; Snyder et al., 2006) have represented FMC as a surface 

dryness index, expressed as the ratio between NDVI and LST; with the 

argument that this ratio displays a statistically stronger correlation to FMC 

than either of the variables considered separately. This is the approach 

taken in this study. 

Previous modelling studies of fire activity in Africa savannas, such as 

Lehsten et al., (2009), have reported significant levels of uncertainty associated 

with the simulations. This uncertainty is important because African 

savannas are among some of the most frequently burnt ecosystems and are 

a major source of greenhouse trace gases and aerosol emissions (Scholes et 

63

64 


al., 1996). The simulation of realistic fire disturbance regimes with biophysical 

and biogeochemical models is a prerequisite for reducing the 

uncertainty of the African carbon cycle, and the feedbacks associated with 

this cycle and the global climate system. 

2 - Materials and methods 

2.1 - Model description 

The model applied in this study is JULES (Joint UK Land Environment 

Simulator), which is the community version of the UK Met Office’s MOSES 

land-surface model. JULES is described in more detail by Cox et al. (1999), 

but in essence was developed for coupling to a GCM, in order to calculate 

the surface-to-atmosphere fluxes of heat and water. Each surface gridbox is 

represented as mixture of five plant functional types: broadleaf trees, 

needleleaf trees, C 3 grasses, C 4 grasses, and shrubs; and four non-vegetation 

types: urban, inland water, bare soil and ice. 

2.2 - Experimental setup 

For this study, JULES was run from 1982 to 2003 at 0.5° x 0.5° spatial resolution 

for IGBP land classes representing savannas and grasslands for 

southern hemisphere continental Africa. Meteorological input data were 

taken from 6-hourly NCEP reanalysis (Kalnay et al., 1996) and TRMM precipitation 

(Kummerow et al., 1998) datasets. The selected IGBP land classes 

were mapped onto the nine JULES surface tiles in accordance with 

Dunderdale et al. (1999), and soil parameters were derived from the ISLSCP 

II soil data set (Global Soil Data Task, 2000). Prior to the main run the 

model was spun-up for a period exceeding 200 years, to enable the soil 

temperature and soil moisture to reach a state of equilibrium. 

An archive of FMC field measurements from four sites within Kruger National 

Park (KNP), South Africa covering the years 1982 to 2003 were used to calibrate 

and validate the model. These measurements were obtained by comparing 

the sample weights before and after oven-drying. Measurements 

acquired from Skukuza and Pretoriuskop sites were used for calibration, 

with measurements from from Satara and Mopani sites used for validation. 


For calibration of the model from measurements obtained from Skukuza and 

Pretoriuskop sites for 1982 to 2003 (Fig. 1, left), the determination coefficient 

(r 2 ) produced from the nonlinear multiple regression equation was 

0.568 (p < 0.001):

Fuel moisture content estimation: a land-surface modelling approach applied to African Savannas 65 

( NDVI 

FMC = 84549.236 * ------------- ) - 96.037 

LST max 

This model for FMC was applied over all savanna regions of southern Africa, 

with validation of the model at Satara and Mopani sites for 1982 to 2003 

(Fig. 1, right) giving r 2 = 0.604 (P < 0.001); the standard error being 21.19. 

Over all southern Africa savanna regions the model simulated high values 

during the wet season, declining as the dry season progressed (Fig. 2). This 

dependence on precipitation was evident at the four KNP sites, whereby the 

trajectory of modelled values was consistent with the temporal variability 

in precipitation. 

Figure 1 - Scatterplots of observed vs. predicted FMC values for Skukuza and Pretoriuskop sites 

(left) and Satara and Mopani (right) for measurements from 1982 to 2003. 

Figure 2 - Mean FMC modelled at each 0.5° gridbox over southern Africa during March (left) 

and September (right) of 2001.

66 



Moisture content is a complex characteristic of vegetation that is a function 

of numerous surface variables. This study has illustrated that a sophisticated 

land surface model, such as JULES, is capable of modelling FMC over 

fine temporal resolutions for large geographical regions. An advantage of 

this method is that modelled FMC does not suffer from missing data limitations 

associated with direct Earth Observation derivations, such as sensor 

technical difficulties and excess cloud cover. Furthermore, a land surface 

model has the ability to seamlessly pass this variable to a coupled fire 

module for estimating fire characteristics, such as rate-of-spread and 

burned area. 

Acknowledgements 

The authors would like to thank Navashni Govender, Programme Manager for 

Fire Ecology at Kruger National Park, South Africa for the provision of FMC 

measurements from 1982 to 2003. 

References 

Chuvieco, E., Aguado, I. and Dimitrakopoulos, A.P., 2004. Conversion of 

fuel moisture content values to ignition potential for integrated fire 

danger assessment. Canadian Journal of Forest Research-Revue 

Canadienne De Recherche Forestiere 34(11): 2284-2293. 

Cox, P.M., Betts, R.A., Bunton, C.B., Essery, R.L.H., Rowntree, P.R. and 

Smith, J., 1999. The impact of new land surface physics on the GCM simulation 

of climate and climate sensitivity. Climate Dynamics 15(3): 183- 

203. 

Dunderdale, M., Muller, J.P., and Cox, P.M., 1999. Sensitivity of the Hadley 

Centre climate model to different earth observation and cartographically 

derived land surface data-sets. The Contribution of POLDER and New 

Generation Spaceborne Sensors to Global Change Studies, Meribel, France, 

pp 1–6. 

Global Soil Data Task., 2000. Global Gridded Surfaces of Selected Soil 

Characteristics (IGBPDIS). International Geosphere-Biosphere Programme 

- Data and Information Services. Available online [http:// 

www.daac.ornl.gov/] from the ORNL Distributed Active Archive Center, 

Oak Ridge, Tennessee, U.S.A. 

Kalnay, E., et al., 1996. The NCEP/NCAR 40-year reanalysis project. Bulletin 

of the American Meteorological Society 77(3): 437-471. 

Kummerow, C., Barnes, W., Kozu, T., Shiue, J. and Simpson, J., 1998. The 

Tropical Rainfall Measuring Mission (TRMM) sensor package. Journal of 

Atmospheric and Oceanic Technology 15(3): 809-817.

Fuel moisture content estimation: a land-surface modelling approach applied to African Savannas 67 

Lehsten, V., Tansey, K.J., Balzter, H, Thonicke, K., Spessa, A., Weber, U., 

Smith, B. and Arneth, A., 2009. Estimating carbon emissions from 

African wildfires. Biogeosciences 6(3): 349-360. 

Sandholt, I., Rasmussen, K. and Andersen, J., 2002. A simple interpretation 

of the surface temperature/vegetation index space for assessment 

of surface moisture status. Remote Sensing of Environment 79(2-3): 

213-224. 

Scholes, R.J., Ward, D.E. and Justice, C.O., 1996. Emissions of trace gases 

and aerosol particles due to vegetation burning in southern hemisphere 

Africa. Journal of Geophysical Research-Atmospheres 101(D19): 23677- 

23682. 

Snyder, R.L., Spano, D., Duce, P., Baldocchi, D., Xu, L.K. and Kyaw, T.P.U., 

2006. A fuel dryness index for grassland fire-danger assessment. 

Agricultural and Forest Meteorology 139(1-2): 1-11.

FUEL TYPE MAPPING USING SPOT-5 IMAGERY AND OBJECT BASED 

IMAGE ANALYSIS 

Abstract: Judicial wildland fire prevention and management requires precise 

information on fuel characteristics and spatial distribution of the various 

vegetation types present in an area. Remote sensing, along with 

Geographic Information Systems (G.I.S.), are important components of fuel 

mapping efforts and fire hazard mitigation. The aim of this work was to 

investigate the potential use of SPOT-5 imagery in fuel type mapping by 

employing object based image analysis. The specific objects were: i) to 

develop an object based classification model for fuel type mapping in 

Chalkidiki (Northern Greece), and ii) to evaluate the transferability of the 

developed model in Attica (Central Greece). A modified version of the 

Prometheus fuel classification scheme was developed, considering information 

content of the SPOT imagery. Originally, an object based model including 

multi-scale image segmentation and classification, was developed and 

applied in Chalkidiki. Following, the model rule set was transferred and 

applied, after minor adjustments, in Attica. Extensive field work was carried 

out in both test sites for model development, training and evaluation. 

The overall accuracy of the fuel type map produced for the Chalkidiki study 

area was more than 85% while the results obtained for the Attica area were 

also satisfactory (overall accuracy 80%). Considering the reliability and 

robustness of the developed approach as well as the characteristics of SPOT- 

5 imagery, the proposed methodology has the potential to be used operationally 

for generating large scale fuel type maps over extended areas. 


I. Stergiopoulos, A. Polychronaki, I.Z. Gitas, G. Galidaki, 

K. Dimitrakopoulos, G. Mallinis 

Laboratory of Forest Management and Remote Sensing, 

School of Forestry and Natural Environment, 

Aristotle University of Thessaloniki, Greece 

thor@for.auth.gr 

Forest fires are an integral part of the Mediterranean ecosystems and constitute 

a real threat to natural environment, given the spectacular increase 

in number of fires events observed in recent decades (Pausas and Vallejo, 

1999) 

In order to improve forest fire prevention and suppression, fire managers 

69

70 


need accurate knowledge/information concerning the description and the 

spatial distribution of the fuels. Since the description of all fuel properties 

is a very complex and difficult task, a classification system is usually used 

by fire managers, where all the physical attributes of fuels are grouped in 

different classes, known as fuel types, according to their fire behavior. 

Specifically a fuel type is defined as “an identifiable association of fuel elements 

of distinctive species, form, size arrangement and continuity that 

will exhibit characteristic fire behavior under defined burning conditions” 

(Merrill and Alexander, 1987). 

Remote sensing and GIS constitute an important and useful tool in mapping 

fuel types. Several studies have tried to accurately map fuels using 

satellite imagery with different spatial and spectral characteristics 

(Giakoumakis et al., 2002; Gitas et al., 2006; Lasaponara et al., 2007; Riaño 

et al., 2002). The aim of this study was a) to develop an object based classification 

model for fuel type mapping in Chalkidiki and b) to evaluate the 

transferability of the developed model in Attica. 

2 - Study Area 

Two study areas were selected for this research. The first one (Figure 1a), 

the Perfecture of Chalkidiki, is about 2.920 km 2 and it is located in 

Northern Greece. The climate is characterized as Mediterranean with short 

periods of drought, hot summers and mild winters. Common forest species 

are Pinus halepensis, Pinus nigra, Quercus conferta, Fagus moesiaca 

Juniperus communis and Quercus coccifera. 

The study area of the Perfecture of Attica (Figure 1b) is about 3.800 km 2 

and it is located in Central Greece. The climate of the area is characterized 

with dry period during hot summers and belongs in the semi dry 

Mediterranean zone. The more extended forests in Attica are the forests of 

Pinus halepensis. There are also, forests of the endemic Greek fir (Abies 

cephallonice) and the area is composed by Quercetum ilicis, Crataegus 

monogyna, Juniperus oxycedrus and many other endemic plants. 


3.1 - Imagery and anchillary data 

In total six SPOT5 satellite images, aquired on April and May 2007, were 

the primary source of information. In addition, anchillary data were used: 

1:5.000 B&W orthophotographs, vegetation maps and a 20m DTM of the 

study areas.

Fuel type mapping using SPOT-5 imagery and object based image analysis 71 

3.2 - Field measurements and fuel classification scheme 

The Prometheus classification system was adapted to the vegetation conditions 

of the study areas and 9 fuel type classes were created: i) rural/no 

vegetation areas, ii) agricultural areas, iii) very sparse vegetation, iv) 

sparse shrublands, v) dense shrublands, vi) sparse broadleaves, vii) dense 

broadleaves, viii) sparse conifers and ix) dense conifers. Homogenous 

patches (30x30m), representative for all fuel conditions, were located using 

GPS device and extensive field measurements were collected both for training 

and assessing the classification process. Stratified sampling method 

was used to select the samples. Main vegetation species, species composition, 

canopy cover and density, terrain information were recorded for every 

sampling area. 

Figure 1 - a) Study area of Chalkidiki and b) Study area of Attica. 

3.3 - Satellite data preprocessing and image enhancements 

All SPOT5 images were orthorectified using the mosaic of the B&W 

orthophotographs and the 20m DTM. Due to the temporal difference of their 

acquisition, radiometric enhancement (histogram match) was performed to 

all SPOT5 images. Two mosaics were derived from the enhanced data. NDVI 

values were calculated and integrated to them.

72 


4 - Object-oriented classification 

The basic processing units of object-oriented image analysis are image 

objects or segments and not single pixels. The object-oriented image analysis 

procedure involved two steps: image segmentation and image classification. 

The bottom-up segmentation approach was used, and the multiresolution 

segmentation algorithm was employed. Two different levels of 

objects were created. For the image classification the nearest neighbor 

method was used to the spectral layer mean values of the image and the 

NDVI. The same methodology was applied to Attica, after minor adjustments. 

Figure 2 shows the results of the two fuel classifications. 

Figure 2 - a) Fuel type map of Chalkidiki, b) Fuel type map of Attica. 


Accuracy assessment in both cases was achieved by comparing the map 

with field data collected from the study areas. Furthermore, in the area of 

Attica the accuracy of the maps was checked by fire managers, and the map 

is already used in pre-fire planning. An approximate overall accuracy of 

85% and 80% was reached for Chalkidiki and Attica respectively. The problem 

was between “sparse broadleaves with dense regenaration” and “very 

dense and tall shrublands”. 


Object-oriented image analysis was applied to enhanced SPOT5 imagery in 

order to map fuel types in Chalkidiki. The methodology was transferred and 

applied to Attica. The accuracy of the final fuel type maps shows that the 

methodology is transferable. Further assessment is needed before the model 

can be used operationaly the fire managers.

References 

Fuel type mapping using SPOT-5 imagery and object based image analysis 73 

Giakoumakis, N.M., Gitas, I.Z., San-Miguel, J., 2002. Object-oriented classification 

modeling for fuel type mapping in the Mediterranean, using 

LANDSAT TM and IKONOS imagery-preliminary results. In: Viegas (Eds.), 

Forest Fires Research & Wildland Fire Safety, Millpress, Rotterdam. 

Gitas, I.Z., Mitri, G.H., Kazakis, G., Ghosn, D., Xanthopoulos, G., 2006. Fuel 

type mapping in Anapolis, Crete by employing QuickBird imagery and 

object-based classification. Forest Ecology and Management 234 (S1), 

S228. doi:10.1016/j.foreco. 2005.08.255. 

Lasaponara, R., Lanorte, A., 2007. On the capability of satellite VHR 

QuickBird data for fuel type mapping characterization in fragment landscape. 

Ecological Modelling 204, pp. 79-84. 

Merril, D.F., Alexander, M.E., 1987, Glossary of Forest Fire Management 

Terms, fourth edn. National Research Council of Canada, Canadian 

Committee on Forest Fire Management, Ottawa, Ontario. 

Pausas, J.G., Vallejo, V.R., 1999. The role of fire in European Mediterranean 

ecosystems. In: Chuvieco, E. (Ed), Remote Sensing of Large Wildfires. 

Springer, Berlin, pp. 3-16. 

Riaño, D., Chuvieco, E., Salas, F.J., Palacios-Orueta, A., Bastarrica, A., 

2002. Generation of fuel type maps from Landsat TM images and ancillary 

data in Mediterranean ecosystems. Canadian Journal of Forest 

Research 32, 1301-1315.

FUEL MODEL MAPPING USING IKONOS IMAGERY TO SUPPORT SPATIALLY 

EXPLICIT FIRE SIMULATORS 

B. Arca 1 

V. Bacciu 2, 3 , G. Pellizzaro 1, 3 , M. Salis 2, 3 , A. Ventura 1, 3 , 

P. Duce 1, 3 , D. Spano 2, 3 , G. Brundu 4 

1 Istituto di Biometeorologia, National Research Council, 

Sassari, Italy, B.Arca@ibimet.cnr.it 

2 Department of Economics and Woody Plant Ecosystems (DESA), University of Sassari, Italy 

3 Euro-Mediterranean Center on Climate Change (CMCC), Italy 

4 Corpo Forestale della Regione Sardegna, Italy 

Abstract: The effect of fire environment on fire spread and behaviour can 

be adequately simulated by using different models, mainly based on semiphysical 

approaches. Effects on fire behaviour can be integrated at various 

scales using spatially and temporally explicit fire spread and behaviour simulators. 

Criticisms of fire simulators frequently concern the need of high 

resolution environmental data, in particular data on fuel types, fuel model 

characteristics and weather variables. The aim of this work was to evaluate 

the capabilities of IKONOS imagery to accurately map fuel types and fuel 

model for the main Mediterranean maquis associations in Northern Sardinia 

(Italy). We also evaluated the sensitivity of the predicted fire spread and 

fire behaviour to variation in spatial resolution of fuel model maps. The 

results showed a sensitivity of the predicted burned areas and rate of 

spread to the accuracy and resolution of fuel model maps, providing a clear 

insight for the use of fire simulators in fire management applications. 


The availability of accurate fuel data at different spatial and temporal 

scales is essential for fire management applications. A reasonable prediction 

of fire potential can help fire managers in planning and prioritizing 

activities and in both fire hazard and fire risk assessment. Fuel physical 

characteristics related to different fuel complexes are frequently parameterized 

into “fuel models”. Fuel model maps are generally supplied as inputs 

to spatially explicit fire simulators to achieve fire spread and behaviour 

information for fuel management and fire management applications. To 

obtain realistic simulations of fire spread and behaviour, fuel maps must be 

developed at fine resolutions. In this context, the use of remotely sensed 

multispectral data can be successfully adopted to develop fuel maps at local 

scale. Recently, the availability of high spatial and spectral satellite data 

has provided useful tools for capturing fine scale fuel distribution. In this 

75

76 


study, we explored the use of IKONOS data to develop fine resolution fuel 

maps. We also evaluated the potential of the fuel model maps for predicting 

fire spread and behaviour using spatially and temporal explicit fire simulators. 

The general aim of this work was to evaluate the capabilities of 

IKONOS imagery to accurately map fuel types and fuel model for main 

Mediterranean maquis associations in Northern Sardinia (Italy). 


The study was conducted in North-West Sardinia. The one-point stratified 

sampling was adopted to identify the sampling sites for the estimation of 

fuel load and other fuel model characteristics by destructive measurements. 

Sampling sites were selected from the analysis of the Corine Land Cover 

map, the IKONOS satellite images of the area, and the map of the potential 

vegetation. The following variables were collected at each sampling site: 

live and dead fuel load, depth of the fuel layer, plant cover. Dead and live 

fuel load were inventoried following the standardized classes (1h, 10h, 

100h) of the USDA National Fire-Danger Rating System. A cluster analysis 

was applied to classify the different sites in terms of fuel types and the 

results were reclassified using an adaptation of Prometheus classification 

system. A supervised classification by the Maximum Likelihood algorithm 

was performed on IKONOS images to identify and map the different types 

of maquis vegetation. The sample of ground truth points used for training 

the algorithm and validating the accuracy consisted of 132 points collected 

by destructive measurements, non destructive measurements, and data 

provided by visual classification of IKONOS images. The accuracy of the 

classification was evaluated using the following statistical indicators 

derived from an error matrix: overall accuracy, user’s accuracy, producer’s 

accuracy, and Cohen’s kappa coefficient. Custom fuel models were associated 

to each vegetation type to obtain fuel model map. This derived map was 

re-sampled by the nearest neighbour algorithm using three different resolutions 

(5m, 10m, 15m), and then used into the FARSITE fire area simulator 

(Finney, 2004) to estimate the effect of the spatial resolution of fuel 

maps on fire spread and behaviour. 


The custom fuel models derived by both the cluster analysis and 

Prometheus classification system are presented in Table 1. The total fuel 

load and the fuel load of different fuel size classes data result similar to, 

experimental data obtained in other studies conducted on shrubland vegetation. 

Fuel model CM4 appears quite similar to fuel model FM4 (35.93 Mg 

ha -1 ; Anderson, 1992) and fuel model SH7 (32.28; Scott and Burgan, 2005). 

The ICONA fuel model key (1990) and Dimitrakopoulos (2002) provided data

Fuel model mapping using ikonos imagery to support spatially explicit fire simulator 77 

similar to CM2 and CM3 (respectively 22.2 and 25.50 Mg ha -1 ). Table 2 

shows the accuracy coefficients as well as the omission and commission 

errors, obtained from the supervised classification of IKONOS images. The 

achieved overall accuracy was 72.73%, with a Kappa coefficient of 0.67. 

The main source of error among all classes was due to the misclassification 

of the ‘‘Broad-leaf” class (user’s accuracy of 37.50%); the 12% of its pixels 

were classified as “High and close maquis”, due to the similar spectral characteristics 

of leaves. Regarding to the maquis, the major classification 

problems come from the high mixing between “Medium” and “Low and 

open” maquis, and “Agriculture and pasture” fuel type. This was probably 

due to both the limited spectral resolution of the sensor and the high spatial 

resolution that increased the spectral within-field variability. The fuel 

model maps derived from IKONOS images were imported into FARSITE. 

Results from FARSITE simulations (Table 3) showed that both the average 

rate of spread and the burned area values were affected by the different 

resolutions of fuel model maps. In particular, the burned area was highly 

sensitive to changes on fuel map resolution for moderate wind speed (from 

45% to 66% of increase relatively to the 5m reference map) compared to 

the rate of spread, that was more sensitive (from 30% to 32% of increase) 

for low values of wind speed. 


Results showed that the use of remotely sensed data at high spatial resolution 

achieves high values of accuracy. The sensitivity analysis showed 

that changes in fuel map resolution affect the predictive capabilities of the 

fire behaviour simulators. In conclusion, the analysis of IKONOS data represents 

a valuable tool to obtain fuel model maps for spatially explicit modelling 

applications. 

References 

Anderson, H.E., 1982. Aids to Determining Fuel Models for Estimating Fire 

Behaviour. USDA Forest Service, Intermountain Forest and Range 

Experiment Station General Technical Report, INT-122. 

Dimitrakopoulos, A.P., 2002. Mediterranean Fuel Models and Potential Fire 

Behavior in Greece. International Journal of Wildland Fire 11, 127-130. 

Finney, M.A., 2004. FARSITE: Fire Area Simulator-model development and 

evaluation. Research Paper RMRS-RP-4, Ogden, UT: U.S. Department of 

Agriculture, Forest Service, Rocky Mountain Research Station. 47 p. 

ICONA, 1990. Clave fotografica para la identificación de modelos de combustible. 

Defensa contra incendios forestales, MAPA, Madrid. 

Scott J.H., Burgan R.E., 2005. Standard fire behavior fuel models: a comprehensive 

set for use with Rothermel’s surface fire spread model. Gen.

78 


Tech. Rep. RMRS-GTR-153. Fort Collins, CO: U.S. Department of 

Agriculture, Forest Service, Rocky Mountain Research Station; 72 p. 

Custom fuel model CM 1 CM 2 CM 3 CM 4 

Dead 1 hr (Mg ha-1) 3.89 5.68 10.07 12.81 

Dead 10 hr (Mg ha-1) 2.43 8.95 3.85 5.33 

Dead 100 hr (Mg ha-1) 0.07 0.57 0.62 0.04 

Live Herbac. (Mg ha-1) 0.33 0.3 0.36 0.19 

Live Woody (Mg ha-1) 2.05 7.89 10.48 12.76 

Dead SAV (m-1) 1964 2.427 22.9 2906 

Live SAV (m-1) 4464 5609 5847 5578 

Fuel Complex depth (m) 0.48 0.7 0.84 1.65 

Table 1 - Custom fuel model parameters grouped by cluster analysis. 

Fuel type class Producer’s Accuracy (%) User’s Accuracy (%) 

Agriculture and pasture 90.00 87.80 

High and close maquis 73.68 73.68 

Medium height and close maquis 71.43 75.00 

Medium maquis 40.00 60.00 

Low and open maquis 53.85 58.33 

Broad-leaf 60.00 37.50 

No Fuel 85.71 85.71 

Overall accuracy 72.73% Kappa coefficient 0.67 

Table 2 - Confusion matrix. 

Resolution map Wind speed 5 km h -1 Wind speed 10 km h -1 

(m) Area (ha) ROS (m min -1 ) Area (ha) ROS (m min -1 ) 

5 73.60 4.77 121.80 7.89 

10 94.70 6.20 177.30 9.56 

15 111.00 6.31 203.40 9.95 

Table 3 - Values of total burned area and mean rate of spread (ROS) provided by FARSITE simulator 

for different fuel model map resolution and wind speed.

FIRST STEPS TOWARDS A LONG TERM FOREST FIRE RISK OF EUROPE 

S. Oliveira, A. Camia & J. San-Miguel 



sandra.santos-de-oliveira@jrc.ec.europa.eu; andrea.camia@jrc.ec.europa.eu; 

jesus.san-miguel@jrc.ec.europa.eu 

Abstract: Forest fires are a major disturbance in Europe, particularly in the 

Mediterranean region. Long term forest fire risk assessment is an important 

tool for supporting the responsible authorities in setting up suitable fireprevention 

measures and allocating fire-fighting resources. This work provides 

the current status of a research effort aimed at developing a long 

term fire risk map of Europe, which will be included as a component of the 

European Forest Fire Information System (EFFIS). The fire risk model adopted 

for the assessment is based on the approach that combines fire occurrence 

and fire outcome, thus encompassing probability of ignition, estimated 

fire behavior and expected consequences, and aiming to integrate 

physical, biological and socio-economic factors. 

The first step has been the enhancement of the fire occurrence data stored 

in the European Fire Database of EFFIS, in which recorded fire ignitions 

exhibit a certain degree of geo-location uncertainty. Location of fire ignition 

points is given in most cases as administrative district without geographical 

coordinates. Therefore methods to approximate density estimations 

of the spatial distribution of fire ignition points are needed. One of 

the options tested in this study is the use of land cover data to constrain 

the geo-location of the ignition points recorded in a given administrative 

district inside the boundaries of the fire spatial domain (i.e. forested and 

wildland areas). The point distribution is made randomly or with a weighted 

probability filtering, and a continuous surface is then created by kernel 

density methods. 

A second step is the analysis of potential variables affecting fire occurrence. 

The list of these variables is being compiled on the basis of extensive 

literature review and experts’ knowledge. The final selection of the 

variables to be used in the model will be based on data availability and 

exploratory statistical analysis. To assess the significance of the predictor 

variables in fire occurrence, several alternative methods are being explored, 

among which are logistic regression and geographically weighted regression. 

The methodology firstly developed for the Euro-Mediterranean region 

79

80 


will then be applied to the other European countries, making the necessary 

adjustments to cope with the specific conditions of each area. 

1 - The European Forest Fire Information System (EFFIS) 

Since 2000, the European Forest Fire Information System supports the forest 

fire services of the EU countries in forest fire prevention and management. 

EFFIS is a web-based system that provides reliable and updated information 

on several issues related to wildland fires in Europe, such as mapping 

of burnt areas, identification of hotspots and fire danger forecast. 

The fire danger maps provided by EFFIS follow a dynamic and short-term 

approach, based on meteorological variables, and are being mainly used in 

the pre-suppression and suppression phases. 

Further improvements of EFFIS, currently under development, are intended 

to include information on long term forest fire risk at the European scale. 

Long term forest fire risk will allow the assessment of the vulnerability of 

forested/wildland areas to fire, the analysis of the probability of occurrence 

and of the potential fire behaviour in case of ignition. 

2 - Long term forest fire risk: first steps in modelling forest fire ignition 

2.1 - The European Fire Database 

The European Fire Database was established in 2000, on the basis of the 

pre-existing database referred to as “common core of information on forest 

fires” that was set up in the context of the first EU Regulation on forest 

fire prevention established in 1992. It currently includes data of 21 countries, 

about 1.87 million individual fire event records. The European 

Mediterranean countries have the longest series of fire data, starting from 

1980 (Portugal), 1983 (Greece) and 1985 (Spain, Italy, France). Most of the 

fire ignition points are provided with a descriptive location based on the 

administrative districts of each country. Only in recent years geographical 

coordinates have been added to the database. 

2.1 - Location of ignition points (IP) 

Since most of the ignition points in the database lack an accurate geo-location, 

it is necessary to estimate the spatial distribution of ignition points. 

One of the possible approaches is the random distribution of the number of 

ignition points in the wildland area of each administrative division, selected 

through Corine Land Cover (Amatulli et al., 2007, de la Riva et al., 

2004), followed by the application of a kernel density estimation method.

First steps towards a long term forest fire risk of Europe 81 

Martinez et al. (2009) applied the historical values of the cumulative number 

of fires divided by the forest area of municipalities. Alternatively, 

records of fire events with coordinates can be used to analyse point location 

patterns (Catry et al., 2007) and apply a weighted probability to the 

distribution of the points for which there are no coordinates. 

For the purpose of this study, the distribution of ignition points with coordinates 

in Portugal, the country with the longest series of fire data, was 

analysed in view of the land cover. Between 2001 and 2007, Portugal 

recorded 133411 IP. The land cover of each IP was retrieved based on 

Corine Land Cover (CLC) 2000 (25 m). Additionally, it was also obtained the 

correspondent category of the Pan-European Forest/Non-Forest Map 2000 

(25 m) (Pekkarinen et al., 2009). 

Considering Corine Land Cover data, it was found the following: 

• The number of ignition points in each CLC category does not correspond 

to the surface area occupied by each land cover in the country. Chisquare 

test was applied in order to verify if there was any difference 

between the observed and the expected number of ignition points per 

land cover class (χ2 = 895.006, df=40, p90% than 

expected) and Agricultural Areas (heterogeneous agricultural areas - 

>74% than expected). 

• CLC classes were assembled into larger groups based on the fire domain 

defined in EFFIS. IP density was calculated based on the number of 

points per km2 in each fire domain group. Artificial surfaces show the 

higher density, followed by agriculture (Fig. 1). 

Figure 1 - Density of ignition points per land use group.

82 


According to the Pan-European Forest/Non-Forest Map 2000, only 10.7% of 

the ignition points fall into the Forest category. It must be considered that 

the Forest class excludes woodlands with trees smaller than 5 m height, 

burnt areas, forest roads and forest nurseries and regeneration (with less 

than 30% canopy closure). 

Data limitations derived from the incorrect insertion of coordinates in the 

original file and the posterior transformation of projection systems must be 

taken into account. Nevertheless, these results provide an important 

insight about the fire domain concerning ignitions; land covers such as 

artificial surfaces that represent the urban/rural interface and the agricultural 

areas, shouldn’t be excluded from the analysis, since fires often start 

in these areas and propagate to wildlands. These findings are in accordance 

with previous studies, which show that around 95% of forest fires in 

Mediterranean Europe are human-caused (e.g., Ne’eman et al., 2004). 

To confirm the influence of human variables in the occurrence of fires, an 

additional preliminary analysis was carried out, based on the distribution 

of ignition points in relation to the roads network. Data of Portuguese 

roads were obtained from TeleAtlas (2006-2009 TeleAtlas) and analysed 

according to the type of roads (national roads and local roads). Buffers of 

50, 100, 200 and 500 meters were created around the roads and the total 

number of ignition points falling into each buffer was calculated. 

It was found that the density of ignition points (IP/km2 /10) increases in 

the classes up to 100 m distance from the roads, being more evident in the 

local roads dataset (Fig. 2). 

Figure 2 - Density of ignition points according to distance from the roads.

3 - Conclusion and further research 

First steps towards a long term forest fire risk of Europe 83 

These preliminary results point out to the potential influence of land use 

and road network in the distribution of fire ignition points in Portugal. 

Further analysis of these and other variables is under development, as part 

of the methodology to assess long-term forest fire risk in Europe. 

References 

Amatulli, G., Perez-Cabello, F., de la Riva, J., 2007. Mapping lightning/ 

human-caused wildfires occurrence under ignition point location uncertainty. 

Ecological Modelling 200: 321-333 

Catry, F.X., Damasceno, P., Silva, J.S., Galante, M., Moreira, F., 2007. Spatial 

Distribution Patterns of Wildfire Ignitions in Portugal. International 

Conference Wildfire 2007, Sevilla, Spain 

de la Riva, J., Perez-Cabello, F., Lana-Renault, N., Koutsias, N., 2004. 

Mapping wildfire occurrence at regional scale. Remote sensing of 

Environment 92: 288-294 

Martinez, J., Vega-Garcia, C., Chuvieco, E., 2009. Human-caused wildfire 

risk rating for prevention planning in Spain. Journal of Environmental 

Management 90: 1241-1252 

Ne’eman, G., Goubitz, S., Nathan, R., 2004. Reproductive traits of Pinus 

halepensis in the light of fire - a critical review. Plant Ecology 171, pp. 

69-79 

Pekkarinen, A., Reithmaier, L., Strobl, P., 2009. Pan-European Forest/Non- 

Forest mapping with Landsat ETM+ and CORINE Land Cover 2000 data. 

ISPRS Journal of Photogrammetry and Remote Sensing, 64 (2), 171-183

ANALYSIS OF HUMAN-CAUSED WILDFIRE OCCURRENCE AND LAND USE 

CHANGES IN FRANCE, SPAIN AND PORTUGAL 

L. Vilar & M.P. Martín 

Centre for Human and Social Sciences. Spanish Council for Scientific Research, 

Madrid, Spain 

lara.vilar@cchs.csic.es; mpilar.martin@cchs.csic.es 

A. Camia 



andrea.camia@jrc.ec.europa.eu 

Abstract: In Southern European countries, where anthropogenic activity 

plays an important role in altering natural dynamics of ecosystems, wildfire 

is a dominant feature of the landscape. This paper explores the relationship 

between human caused fires and land use changes in France, Spain and 

Portugal. The Fisher exact test has been applied to examine the significance 

of this association at provincial level (NUTS level 3). The increase in urban 

use between 1990 and 2000 was found to be associated with the decrease 

of negligent fires in France and Spain. The preliminary regional view provided 

in this work can contribute to the understanding of main driving factors 

of biomass burning at continental and global scales. 


Mediterranean ecosystems can not be fully understood without considering 

the role of fires. Natural fires have been essential to maintain biodiversity 

in the area. Fire has been also a widely used tool to manage the territory. 

However, in the last decades, natural fire regimes have experienced significant 

alterations (Westerling et al., 2006, FAO, 2007). Understanding the 

mechanisms behind fire occurrence is expected to improve fire prevention 

activities, supporting fire risk assessment and reducing the negative 

impacts of wildland fires. The aim of this paper is to explore the relationship 

between human caused wildland fires and land use changes in the 

European Mediterranean countries during the last decades (1986-2006). 

2 - Methods 

2.1 - Study areas 

The study areas are three European Mediterranean countries: France, Spain 

and Portugal. In these countries fire plays an important role, causing 

85

86 


extensive environmental damage in forest and agriculture areas. Historical 

fire trends in these countries show an increase in the total number of fires 

from the 1986-1995 to 1996-2006 period. Regarding fire causes, in the 

three countries more than 90% of the fires are due to causes related to 

human activities. Concerning land cover changes, the three countries show 

a similar trend characterized by increase of urbanized area and decrease of 

agricultural lands. This trend is sharper in France and Portugal and less evident 

in Spain (after analysis of Corine land cover changes -CLC1990 - 

CLC2000- dataset, http://dataservice.eea.europa.eu/, May 2009). 

2.2 - Data 

The fire data source used in this work is the fire database of the European 

Forest Fire Information System (EFFIS, http://effis.jrc.ec.europa.eu/, May 

2009) which integrates harmonized fire statistics from the EU countries. 

This database is stored at the Institute for Environment and Sustainability 

of the Joint Research Centre (JRC). It contains information on individual 

fire events related to the location of the ignition point (NUTS3), the fire 

date, presumed cause and burned area. The fire causes are classified in the 

following four broad categories: unknown, lightning, accident or negligence 

and deliberate. The fire data used in the study cover 22 years, split in two 

periods: 1985-1995 and 1996-2006. We have used Corine Land Cover 1990 

(CLC1990) and 2000 (CLC2000) to analyze the landscape changes in the 

study areas. The Corine land cover changes (CLC1990 - CLC2000) dataset 

(http://dataservice.eea.europa.eu/dataservice/, May 2009) combines 

CLC1990 and CLC2000. From this map, which includes all the category 

changes from CLC level 3, we have aggregated the main land covers and the 

changes between them: urban, agricultural, forest, grasslands and shrublands. 

We have carried out the analysis at NUTS level 3 (province). The vector 

layer with the NUTS3 boundaries has been taken from the 

Eurostat/GISCO (Geographic Information System of the European 

Commission) database (http://epp.eurostat.ec.europa.eu/portal/page/portal/gisco/introduction, 

May 2009). 

2.3 - Statistical analysis 

For the fire data, we have analyzed the relative increase and decrease of 

negligence and deliberate caused fires by NUTS3 as percentage of the total 

number of fires with known cause. For the land cover data we have calculated 

the extent of each land cover change category and referred to the 

total extent of the NUTS3. To examine the significance of the association 

between the increase or decrease of human-caused wildfires and land cover 

changes we have applied the Fisher exact test of significance. This test can 

be used instead of the chi-square test in 2 by 2 tables particularly for small

Analysis of human-caused wildfire occurrence and land use changes in France, Spain and Portugal 87 

samples (Garson 2006). This test has been applied to analyze potential 

connection between wildfire occurrence trends and (i) the land cover 

change category that covers the largest area in the NUTS3 (ii) the land 

cover change category (that covers the largest area in the NUTS3) grouped 

in four main categories, meaning the increase in each cover to the detriment 

of the other covers: increase of urban area (urbanization), increase of 

forest areas (forest expansion), increase of agricultural areas and agricultural 

abandonment. 

3 - Results 

In France, in those areas where forest fire causes have changed in the study 

period, the main land cover changes observed have been from agricultural 

to urban, from forest to shrubland and from grassland to shrubland. 

However, we have found statistically significant relationships only between 

the decrease of fires due to negligence and the agricultural to urban land 

cover change, the null hypothesis of no relationship has been rejected in 

this case with p=0.035. When agricultural to urban is the main change, the 

decrease of negligence has been the 58.7% of the NUTS3. When agricultural 

to urban is not the main change, the increase in negligence fires has 

been the 63.8%. Figure 1 shows the NUTS3 where these relationships have 

been found: 

Figure 1 - Decrease in fires caused by negligence between the periods 1986-1995 and 1996- 

2006 and agricultural to urban land cover change in France by NUTS3.

88 


In Spain, we have found statistically significant relationships between the 

decrease of negligent fires and urbanization as main land cover change. 

When urbanization is the main change, then the decrease in negligence 

fires has been the 60%. In Portugal, none of the land cover changes had a 

statistically significant relationship with the changes in fire trends. 

4 - Discussion and conclusions 

We have found significant relationships between land cover changes and 

the occurrence of wildland fires due to negligence in France and Spain. The 

increase of urbanization was associated with a decrease of negligence fires 

in both countries. In France, the relationship was statistically significant 

where the main land cover change between the two study periods was from 

agricultural areas to urban, whereas in Spain was the increase in urban land 

cover to detriment to the other land covers. The socioeconomic changes in 

the last decades in the European countries have lead to the abandonment 

of rural activities and the increase of urban areas. This phenomenon has 

several effects: forest management is disappearing because of the abandonment 

of the rural activities, driving to an accumulation of fuel in the 

forests. This accumulation favors the probability of fire ignition. We have 

found that the gaining in urbanization is related to the decrease in fire due 

to negligence actions. This effect might be explained because if there are 

less forest areas the probability of a fire may decrease. In both countries 

we have not found significant relationships with the land cover changes 

and the fires due to deliberate actions. In Portugal, we have not found significant 

relationships with the fire occurrence trends. These results show 

some preliminary findings of the analysis of the relation between humancaused 

fire trends and land cover changes in three South-Western European 

countries. Because of the difference between countries, it would be needed 

a more detailed analysis of the fire cause types as the exact meaning of 

the 4 broad fire cause categories may differ from country to country. 

Besides, a more detailed spatial scale such as e.g. NUTS5 level, would 

improve the analysis, because the trends and changes could be better 

depicted. Also, other data sources of land cover change should be tested. 

References 

Food and Agricultural Organization of the United Nations (FAO), 2007. Fire 

Management Global Assessment. A thematic study prepared in the 

framework of the Global Forest Resources Assessment 2005. FAO Forestry 

Paper 151. Rome, Italy. 320 pp. Available at: http://www. fao.org/ 

forestry/ fra2005/en/ (last accessed May 25, 2009). 

Garson, D., 2006. Statnotes: Fisher Exact Test of Significance. Available at: 

http://faculty.chass.ncsu.edu/garson/PA765/fisher.htm (last accessed

Analysis of human-caused wildfire occurrence and land use changes in France, Spain and Portugal 89 

May 25, 2009). 

Westerling, A.L., Hidalgo, H.G., Cayan, D.R. and Swetnam, T.W., 2006. 

Warming and earlier spring increase western US forest wildfire activity. 

Science 313, 940-943.

FUEL MOISTURE CONTENT ESTIMATION BASED ON HYPERSPECTRAL DATA 

FOR FIRE RISK ASSESSMENT 

T.A. Almoustafa, R.P. Armitage & F.M. Danson 

Centre for Environmental Systems Research, School of Environment and Life Sciences, 

University of Salford, Salford, UK 

T.A.Almoustafa@pgr.salford.ac.uk 

Abstract: This study aims to investigate whether the relationships between 

fuel moisture content (FMC) and vegetation spectral reflectance identified 

in previous studies, can be used to improve fire risk assessment for fireprone 

upland vegetation in the UK. Two hyperspectral images from the AISA 

Eagle and Hawk sensors were collected over a UK test site in 2008. In-situ 

field spectra were collected concurrently with the flights, together with 

ground data on live FMC and other relevant variables. The results show that 

FMC varies temporally and with vegetation type. Reflectance variability 

with FMC is expressed most strongly in the SWIR and NIR. The sensitivity 

of SWIR and NIR reflectance to FMC variations is much greater when considering 

each vegetation type individually. The results also compare first 

derivative and broad band vegetation indices (VI) with FMC. 


Live fuel moisture content (FMC) is controlled by the interaction of plant 

physiology and soil moisture conditions, and is spatially and temporally 

highly variable (Chuvieco et al., 2002). Empirical and model-based studies 

have examined relationships between broad band spectral reflectance and 

vegetation FMC, but few studies have used hyperspectral data. This technology 

provides fine spectral resolution that facilitates vegetation biophysical 

and biochemical information mapping. In addition, there is a lack 

of research on the relationship between spectral reflectance and FMC in 

temperate environments, in particular upland areas. This research aims to 

test the use of airborne hyperspectral data for mapping FMC in semi-natural 

upland areas. 

91

92 



2.1 - Data collection 

Burbage Moor, located south west of Sheffield (United Kingdom) on the 

edge of the Peak District National Park, was the test site. It is characterised 

by flat topography and composed of homogenous Calluna vulgaris stands of 

different ages and conditions, areas of grassland dominated by Eriophorum 

angustifolium, bog and areas of Pteridium aquilinum invasion. A field campaign 

was carried out in 2008, focusing on 10 plots consisting of Calluna 

stands of different ages, Eriophorum angustifolium and mixed Eriophorum 

angustifolium and Calluna stands. At each plot canopy height and vegetation 

composition were recorded and vegetation samples were collected for 

FMC estimation. In addition, the spectral reflectance was measured using a 

field spectroradiometer. The spectral measurements of the vegetation and a 

series of light and dark calibration targets were acquired from a nadir position 

2m above each plot. This field work was concurrent with airborne 

hyperspectral data acquisition. Hyperspectral images from the AISA Eagle 

and Hawk instruments (390-2450 nm) were collected by the U.K. Natural 

Environment Research Council (NERC) Airborne Remote Sensing Facility 

(ARSF) over three parallel North/South flight lines on May 6 th and July 1 st 

2008. 

2.2 - Laboratory work and image processing 

The vegetation samples were stored in well-sealed plastic bags to avoid 

moisture loss and transferred to the laboratory for gravimetric moisture 

analysis on the day of data collection. The fresh weight (Wf) of each sample 

was measured using an electronic balance and then they were dried for 

48 hours at 60°C, as recommended by Chuvieco et al. (2002), and their dry 

weight (Wd) determined (equation1). 

FMC= [Wf-Wd/Wd]*100 (1) 

The radiance values for the study plots were extracted from the airborne 

images of the two dates and corrected to reflectance using the field spectroradiometer 

measurements from the calibration targets, using a band-byband 

linear regression approach. The first derivative was calculated using a 

simple difference function, where a 7nm gap was taken either side of a 

given wavelength. Landsat TM-equivalent reflectance was calculated for 

each plot and was used to calculate broad band vegetation indices (VI). 

Linear correlation coefficients were calculated between FMC and spectral 

reflectance, the first derivative and with the VIs for all the plots and for 

Calluna plots only.

3 - Results 

Fuel moisture content estimation based on hyperspectral data for fire risk assessment 93 

FMC was found to be lower in May than July for all the plots, with an average 

of 75 % and 116% respectively. The largest temporal variation in FMC 

values was for regenerated Calluna, where FMC increased from 66% to 136% 

between May and July. The smallest variation in FMC was for building 

Calluna, where values changed from 84% in May to 107% in July. 

Statistically significant negative correlations were found between FMC and 

spectral reflectance at 685nm, 689nm and 704nm when all plots were considered. 

When considering only Calluna plots, there were significant negative 

correlations with visible wavelengths and positive correlations in the 

NIR between 728nm and 915nm. Over the SWIR there were significant negative 

correlations between 1428nm and 1782nm and between 2034nm and 

2337nm. The strongest correlation was in the visible region and the highest 

value was in the blue region, where r = 0.8. When the first derivative 

was considered, a greater number of wavelengths were significantly correlated 

with FMC. The main areas of significant positive correlation were 

between 678nm and 764nm and between 1782nm and 1807nm, and additionally 

between 506nm and 538nm, and at 1416nm, 1668nm and 2375nm 

for Calluna-only plots. The main areas of significant negative correlations 

were at 594nm and 934nm and between 1081nm and 1182nm, and additionally 

between 1258nm and 1315nm, and at 1454nm for the Calluna-only 

plots. In terms of VIs, the strongest correlation was with the moisture 

stress index (MSI= TM5/TM4) where (r= -0.98). 


FMC showed important variation with vegetation type and it exhibited 

important temporal variations, where FMC values increased from May to 

July. This may be due to the larger portion of green material in the canopy 

in July compared to May. There was no significant correlation between FMC 

and the spectral reflectance in the visible region with the exception of a 

few wavelengths which had significant negative correlation with FMC. This 

corresponds to the findings of Bowyer and Danson (2004), who showed that 

FMC had no direct impact on spectral reflectance in the visible domain. This 

is expected as the main parameter that controls the leaf spectral variation 

for visible wavelengths is leaf chlorophyll content (Zhang et al., 2008). 

However, when considering Calluna plots separately, FMC was correlated 

with a large number of visible wavelengths, which may be due to the cocorrelation 

between FMC and canopy chlorophyll content. This correlation 

became stronger and positive for wavelengths in the NIR region, which may 

be due to the co-correlation between FMC and leaf area index (LAI). Higher 

LAI and biomass produces more scattering within and between leaves, 

increasing the reflectance due to the secondary effects that FMC has on the 

internal cell structure of the leaves (Liu et al., 2004). In the SWIR, the

94 


higher the moisture, the lower the reflectance, this was expected as this 

region is strongly affected by water absorption, which is known to lower 

reflectance. These results correspond to the findings of Liu et al. (2004). 

Using the first derivative enhanced the spectral signal produced by FMC 

variation and highlighted more wavelengths that were significantly correlated 

with FMC. This may be due to the fact that the first derivative reduces 

the variability caused by changes in other factors such as soil reflectance 

and solar zenith angle. A range of VIs showed strong correlations with FMC, 

which corresponds to the findings of Chuvieco et al. (2002) and Danson and 

Bowyer (2004) who found a significant correlation between FMC and NDII 

and WI. To summarize, FMC exhibits important variations which affect the 

spectral reflectance measured by airborne hyperspectral instruments, particularly 

in the NIR and SWIR regions. Using the first derivative and specific 

VIs improves the correlation between hyperspectral data and FMC. This 

study indicates that airborne hyperspectral data may be used for FMC mapping 

in semi-natural upland areas. 

Acknowledgements - This study is funded by the Syrian government 

through Tishreen University, Lattakia, Syria. The authors would like to 

thank the Natural Environment Research Council Field Spectroscopy Facility 

(NERC-FSF), Alasdair Mac Arthur, Natural England, Moors for the Future, and 

Sheffield City Council for their support. The airborne imagery was supplied 

by the UK Natural Environment Research Council (NERC) Airborne Remote 

Sensing Facility (ARSF) (Grant Award GB08/3). 

References 

Bowyer, P. and Danson, F.M., 2004. Sensitivity of spectral reflectance to 

variation in live fuel moisture content at leaf and canopy level. Remote 


Chuvieco, E., Riaño, D., Aguado, I. and Cocero, D., 2002. Estimation of fuel 

moisture content from multitemporal analysis of Landsat Thematic 

Mapper reflectance data: Applications in fire danger assessment. 

International Journal of Remote Sensing, 23, 2145–2162. 

Danson, F.M. and Bowyer, P., 2004. Estimating live fuel moisture content 

from remotely sensed reflectance. Remote Sensing of Environment, 92, 

309-321. 

Liu, L., Wang, J., Huang, W., Zhao, C., Zhang, B. and Tong, Q., 2004. 

Estimating winter wheat plant water content using red edge parameters. 

International Journal of Remote Sensing, 25, 17 3331-3342. 

Zhang, Y., Chen, J.M., Miller, J.R. and Noland, T.L., 2008. Leaf chlorophyll 

content retrieval from airborne hyperspectral remote sensing imagery. 

Remote Sensing of Environment, 118, 3234-3247.

CYBERPARK PROJECT: MULTITEMPORAL SATELLITE DATA SET FOR 

PRE-OPERATIONAL FIRE SUSCEPTIBILITY MONITORING AND POST-FIRE 

RECOVERY ESTIMATION 

Abstract: This paper describes a system for the monitoring of natural protected 

areas in the Foggia Province (Apulia Region South of Italy) which was 

developed in the framework of the Cyberpark 2000 project (2007-2008). The 

project was funded by the UE Regional Operating Program of the Apulia 

Region (2000-2006). The system fulfills three main aims: it acts as a preventive 

tool by monitoring fire susceptibility, it backs up the forest fire 

detection using and integrated approach, and it assists in planning the recuperation 

of the burned areas and monitoring post-fire vegetation recovery. 


A. Lanorte 1 , F. De Santis 1 , R. Coluzzi 1 , T. Montesano 1 

M. Monteleone 2 , R. Lasaponara 1 

1 CNR-IMAA, Tito Scalo (PZ), Italy 


2 Università di Foggia, Facoltà di Agraria, Italy 

This paper presents the knowledge-based system we developed in the context 

of Cyberpark 2000 project funded by the UE Regional Operating 

Program of the Apulia Region (2000-2006).The project aim was to support 

decision making for the management and monitoring of the natural protected 

areas located in the Foggia Province (Apulia Region). The system 

includes four different modules, from A to D. In this paper we will focus on 

modules A and D. 

A satellite time series of high spatial resolution data for supporting the 

analysis of fire static danger factors through land use mapping and 

spectral/quantitative characterization of vegetation fuels; 

B satellite time series of MODIS for supporting fire dynamic risk evaluation 

of study area – Integrated fire detection by using thermal imaging 

cameras placed on panoramic view-points; 

C integrated high spatial and high temporal satellite time series for supporting 

studies in change detection factors or anomalies in vegetation 

covers; 

D satellite time-series for monitoring: (i) post fire vegetation recovery 

and (ii) spatio/temporal vegetation dynamics in unburned and burned 

vegetation covers. 

95

96 


2 - Methodological Approach and results 

The knowledge-based system we developed for fire susceptibility assessment 

and fire recovery monitoring is mainly based on the use satellite time 

series. In particular, the analysis of the MODIS data set 1999 to 2008 

(available free of charge from NASA web site) provides reliable information 

for the characterization of vegetation variations at different temporal and 

spatial scales. This is crucial for (i) fire susceptibility and (ii) vegetation 

fire recovery. 

2.1 - Fire susceptibility assessment 

The assessment of fire susceptibility is performed (as described in Lanorte 

et al., 2009 in this book) integrating fuel property (type and loading) with 

Greenness and fuel moisture daily maps. Fuel property is obtained from 

Landsat TM data processed using supervised classification techniques and 

spectral analysis methodologies performed at sub-pixel level. Using Landsat 

TM images we mapped: (i) Vegetation type, (ii) Fuel type (Prometheus system), 

(iii) Fuel model (NFFL system), (iv) Fuel load. 

Figure 1 - Study area and fuel types. 

The analysis was performed in the Capitanata land (Apulia region - 

Southern of Italy) which is shown in figure 1. The estimation of fire susceptibility 

was performed as in the case of Basilicata (FIRE-SAT) using 

Landsat TM for fuel type mapping and MODIS for fuel greenness and moisture 

estimation. Figures 2 show fire danger maps for February, May, July, 

and October 2007 respectively.

Cyberpark project: Multitemporal satellite data set for pre-operational fire susceptibility monitoring and post-fire recovery estimation 97 

Figure 2 - Danger map obtained for February, May, July, and October 2007. 

2.2 - Post-fire vegetation recovery assessment 

The assessment of vegetation fire recovery is performed using satellite time 

series. MODIS Normalized Difference Vegetation Index (NDVI) from 2001 to 

2007 was used to examine the recovery characteristics of fire affected vegetation 

at different temporal and spatial scales. In order to eliminate the 

phenological fluctuations, for each decadal composition, we focused on the 

normalized departure NDVIdn = (NDVI-NDVIm)/σndvi where NDVIm is the 

decadal mean and σndvi is the decadal standard deviation see figure 3. The 

decadal and the standard deviation are calculated for each decade, e.g., 

first decade of January, by averaging over all years in the record. We analyzed 

both: (1) Post-disturbance NDVI spatial patterns on each image date 

were compared to the pre-disturbance pattern to determine the extent to 

which the pre-disturbance pattern was re-established, and the rate of this 

recovery. (2) time variation of NDVI for pixels fire-affected and fire-unaffected 

areas (Tuia et al., 2008).

98 


Figure 3 - Time series for Bosco of Orsara affected by fire: from top to down: (i) NDVI, (ii) 

(NDVI-NDVIm), and (iii) NDVIdn trend. Red arrows show time of fire occurrence (2001, and 

2005). 

3 - Final remark 

Results obtained suggest that the satellite MODIS time series provides reliable 

information for the characterization of vegetation variations at different 

temporal scale: (i) short term to monitor fire susceptibility, (ii) medium 

and long term to assess post fire vegetation recovery. 

References 

Telesca L., Lasaponara R. and Lanorte A., Intra-annual dynamical persistent 

mechanisms in Mediterranean ecosystems revealed SPOT-VEGETATION, 

Time Series, Ecological Complexity, 5, 151-156, 2008.

REMOTE SENSING-BASED MAPPING OF FUEL TYPES USING 

MULTISENSOR, MULTISCALE AND MULTITEMPORAL DATA SET 

Abstract: The characterization of fuel types is very important for computing 

spatial fire hazard and risk, for simulating fire growth and intensity 

across a landscape. However, due to the complex nature of fuel characteristic 

a fuel map is considered one of the most difficult thematic layers to 

build up. The advent of sensors with increased spatial and spectral resolution 

may improve the accuracy and reduce the cost of fuels mapping. The 

objective of this research is to evaluate the accuracy and utility of imagery 

from Quickbird, Advanced Spaceborne Thermal Emission and Reflection 

Radiometer (ASTER), Landsat Temathic Mapper, and MODIS. 


R. Lasaponara, R. Coluzzi, A. Lanorte 

CNR-IMAA, Tito Scalo (PZ), Italy 


In the context of fire management, fuel maps are essential information 

requested at many spatial and temporal scales for managing wild-land fire 

hazard and risk and for understanding ecological relationships between wildland 

fire and landscape structure. Remote sensing data provide valuable 

information for the characterization and mapping of fuel types and vegetation 

properties at different temporal and spatial scales including the global, 

regional, landscape levels, down to a local scale level. This study aims to 

ascertain how well remote sensing data can characterize fuel type at different 

spatial scales in fragmented ecosystems. For this purpose, different 

approaches have been adopted for fuel type mapping: the well-established 

classification techniques performed at the pixel level and spectral mixture 

analysis (SMA) which allows a sub-pixel analysis. Among the classifications 

performed at the pixel level, two different techniques, parametric and non 

parametric were considered in this study. The K NN (nearest neighbor) was 

applied to obtain a non parametric classifications of fuel types; whereas the 

maximum likelihood classification (MLC) was adopted for performing a parametric 

analysis. The K NN and MLC were adopted because both of them are 

“conventional supervised classifiers” widely used for remote sensed imagery. 

99

100 


Field work 

Remote sensed 

dataset 

Geometric 

correction 

Punctual reconnaissance of the study area 

developed in many years and directed towards 

identify the vegetation typologies and structural 

characteristics of vegetation cover 

Fuel types recognitions performed before, 

during and after the acquisition 

of remote sensing data 

Prometheus adaptation 

The seven fuel type classes standardized 

in the context of Prometheus system were 

datailed identifield and carefully verified 

for the study area on the basis of ground 

field recognitions 

Model construction 

Selection of Regions of 

Interest point (Ground-Truth 

dataset) for the seven 

classes (fuel types) + no 

Fuel and unclassified pixels 

Supervised Classification 

using training dataset 

Fuel type maps 

Post classification 

Confusion matrix and testing 

dataset 

Accuracy assessment 

Figure 1 - Shows the work flow of the methodological approach adopted. 

Spectral signature 

characterization

Remote Sensing-based Mapping of fuel types using Multisensor, Multiscale and Multitemporal data set 101 

The selection of Fuel type classes, training data, and classification procedure 

is a key step that strongly influence the final results and the accuracy 

levels. 

Fuel type classes considered in this study are an adaptation of the 

Prometheus model as in Lasaponara and Lanorte (2007), Fieldwork fuel type 

recognitions, performed before, after and during the acquisition of remote 

sensing data, were used as ground-truth dataset to assess the results 

obtained for the considered test area. Results from different classification 

strategy must compared in order to assess which performs better. 

The method comprised the following three basic steps: (I) adaptation of 

Prometheus fuel types for obtaining a standardization system useful for 

remotely sensed classification of fuel types and properties in the considered 

Mediterranean ecosystems; (II) model construction for the spectral 

characterization and mapping of fuel types based on supervised classification 

techniques (III) evaluation for Quickbird, Landsat TM, ASTER, and 

MODIS -based results with ground-truth. 

Figure 2 - Quickbird - based MLC classification for a test area in Calabria.

102 


The analysis has been performed for some test areas in the south of Italy. 

Using two different approaches: (I) classification performed using a single 

date images, and (ii) classification performed using multidate images. 

In the case of single date data processing, we observed that as a whole, 

SMA results obtained from Quickbird (overall accuracy 77%) and Aster 

(overall accuracy 82%) did not show any significant improvements, whereas 

results from TM and MODIS have shown that the use of unmixing technique 

allows us to improve at around 7% and 12% the accuracy level for 

TM (k coefficient from 57% to 64%) and MODIS (k coefficient from 67% to 

79%) respectively compared to the MLC and KNN. These results confirmed 

the effectiveness of SMA in handling spectral mixture problems, especially 

in fragmented ecosystems as those considered for our analysis. 

Results from KNN showed improvements around 3% and 5% for Quickbird 

and Aster, whereas no significant improvements have been observed for TM 

and MODIS data set. 

In the case of multidate processing, improvements around 4% were 

obtained for the different data set we considered. 


In this paper, imagery from Quickbird, Advanced Spaceborne Thermal 

Emission and Reflection Radiometer (ASTER), Landsat Temathic Mapper, and 

MODIS, have been evaluated in terms of accuracy and utility for mapping 

fuel type and load. Different supervised classification techniques were compared. 

Results showed that each data set processed using different classification 

provided satisfactory accuracy levels. In particular, KNN well performed 

for Quickbird (81%) and Aster (87%), whereas SMA provided the 

best results for TM (64%) and MODIS (79%). Additional improvement can 

be achieved using data fusion approach to merge the spatial and spectral 

characteristics of different satellite data set. 

References 

Lasaponara R., and Lanorte A., 2007. On the capability of satellite VHR 

QuickBird data for fuel type characterization in fragmented landscape. 

Ecological Modelling (ECOMOD845R1).

ASSESSING CRITICAL FUEL PARAMETERS USING AIRBORNE FULL 

WAVEFORM LIDAR: THE CASE STUDY OF BOSCO DELL’INCORONATA 

(APULIA REGION, SOUTHERN ITALY) 

Abstract: The aim of this paper is to develop the use of airborne lidar 

(LIght Detection and Ranging) remote sensing for accurately and effectively 

assessing fuel critical parameters, (such as closeness, height, density, 

load) for both canopy and understory. The study was performed in a natural 

protected area (Bosco dell’Incoronata) located in the Apulia Region 

(Southern Italy). Lidar data acquisition was carried out on April 2007. 

Estimates of fuel properties were compared with in-situ data collected at 

the same time (more or less) as the LIDAR data acquisition. 


R. Lasaponara 1 , R. Coluzzi 1 , A. Guariglia 2 , 

M. Monteleone 3 , A. Lanorte 1 



2 Geocart s.r.l., Potenza, Italy 

3 University of Foggia, Foggia, Italy 

Airborne laser scanning is a remote sensing approach developed to obtain 

a high-precision and complete vertical profile of the height of objects by a 

laser pulse. Thought its efficient data sampling capabilities Airborne Laser 

Scanning (ALS) has completely revolutionized the area of bathymetric and 

topographic surveying. The applications of airborne ALS have been increasing 

rapidly over recent years. ALS data are used for corridor mapping, 

coastal and urban area monitoring, land cover classification, estimation of 

forest tree height, assessment of the seasonal canopy differences, collecting 

forest inventory data. Recent studies have examined the possibility of 

using ground LIDAR in fuel type mapping, see for example Riano et al. 

(2004), but study based on ALS are quite rare. The latest generation of airborne 

ALS is the Full-Waveform (FW) scanner which offers improved capabilities 

compared to the conventional system, especially in areas with close 

canopy and dense vegetation. Nowadays the majority of published studies 

(for different applications fields) are based on data collected by conventional 

ALS. In this paper, critical fuel parameters were assessed for both 

canopy and understory by combining data from FW ALS and ortophotos. 

103

104 


2 - Methods 

There are two different types of ALS: (i) conventional scanners based on 

discrete echo and (ii) full-waveform scanners. The conventional or discrete 

echo scanners detect a representative trigger signal for each laser beam, 

whereas the full-waveform (FW) laser scanning systems permit one to digitize 

the complete waveform of each backscattered pulse. FW ALS data 

allows us to have more control in the interpretation process of the physical 

measurements, this enables the extraction of additional information 

about the structure and the physical backscattering characteristics of the 

illuminated surfaces. The FW ALS sensor can effectively penetrate vegetation 

canopies and therefore it allows us to identify and better classify the 

understory. 

3 - Data analysis 

The ALS survey was operated by GEOCART on the 26 th April 2007 and carried 

out using a RIEGL Airborne Laser Scanner LMS-Q560 mounted on an 

helicopter. The average point density value of the dataset is about 30 

points/m2. The accuracy is 25 cm in xy and 10 cm in z. 

Usually, laser scanning products can be classified by: (i) height; (ii) intensity; 

(iii) RGB colours if an ortophoto is available; (iv) echo width. Herein, 

we will focus on the elaboration we performed using both height, obtained 

from the 3D point clouds, and ortophoto acquired at the same time as ALS 

survey. Due to its ability to pass between plants or tree branches, ALS is 

suited to assess both canopy and understory, but it is necessary to process 

the ALS point cloud using appropriate numerical filters. For the case study 

the elaborations were performed using a commercial software TerraScan 

(Terrasolid, www.terrasolid.fi), which represents a high standard for the 

laser data processing. TerraScan classification is based on a parametric 

approach and develops according to an orderly sequence of stages decided 

by an operator. In this case study, the classification of laser data was performed 

using a strategy based on a set of “filtrations of the filtrate”. 

Appropriate criteria for the classification and filtering were set to gradually 

refine the intermediate results in order to obtain fuel height identification 

and the discrimination between canopy and understory. The workflow 

can be summarized as follows: 

- Classify low point; 

- Classify ground; 

- Classify points below surface; 

- Classify points by class; 

- Classify points by height from ground for different heights; 

- Classify isolated points; 

- Shape identification and load computation.

Assessing critical fuel parameters using airborne full waveform LIDAR: the case study of Bosco dell’Incoronata 105 

The critical aspect of this classification is the appropriate choice of the 

threshold values. 

Figure 1 - (left) Study area Location and (right) orthophoto of the study area acquire at the 

same time as lidar data set. 

Figure 2 - LIDAR classification by vegetation height: (upper) a section and (lower) a spatial 

mapping.

106 


The estimation of forest fuel loads (currently a work in progress) is performed 

using a workflow based on: (i) the identification of shape, (ii) the 

estimation of variables which describe fuel structure (height, cover, volume) 

and dimension, and finally, (iii) the fuel load is determined from the 

literature references. 

Figure 3 Shape identification: examples of customized models based on TerraScan software 


The preliminary results obtained for the test site Bosco dell’Incoronata 

pointed out that FW ALS can be fruitfully used for a detailed and reliable 

mapping of critical fuel parameters for both canopy (including canopy bulk 

density, canopy height, canopy fuel weight, and canopy base height) and 

understory over extensive forested areas. The elaborations were performed 

using a commercial software TerraScan (Terrasolid, www.terrasolid.fi), which 

represents a high standard for the laser data processing. In this test area, 

Lidar survey allowed us to: (i) assess and map fuel heights for wood as well 

as for shrub under tree; (ii) to obtain detailed Digital Terrain Model (DTM) 

and Surface (DSM), (iii) to estimate details for a single tree, (iv) to estimate 

and map fuel load 

References 

Reutebuch, S.E., Andersen H.E., et al., 2005. Light detection and ranging 

(LIDAR): An emerging tool for multiple resource inventory, Journal of 

Forestry 103(6): 286-292. 

Riano, D., Chuvieco E., et al., 2004. Generation of crown bulk density for 

Pinus sylvestris L. from lidar, Remote Sensing of Environment 92(3): 

345-352.

II VALIDATION OF RS 

PRODUCTS FOR FIRE 

MANAGEMENT

ASSESSMENT OF SPECTRAL INDICES DERIVED FROM MODIS DATA AS 

FIRE RISK INDICATORS IN GALICIA 

M.M. Bisquert 1 

1 Earth Physics and Thermodynamics Department, University of València, 

Burjassot, España 

Maria.Mar.Bisquert@uv.es 

J.M. Sánchez 1-2 , V. Caselles 1 , M.I. Paz Andrade 3 & J.L. Legido 4 

2 Applied Physics Department, University of Castilla-La Mancha, 02071, Albacete, España 

3 Applied Physics Department, University of Santiago, Santiago de Compostela, España 

4 Applied Physics Department, University of Vigo, Vigo, España 

Abstract: The Galicia region, placed at the northwest of Spain, is an area 

specially affected by the action of fire. It is known that fires can be related 

to the water status of the vegetation. Variations in water content produce 

changes in the visible and infrared regions of the spectrum. These 

characteristics permit us to monitor the changes in the vegetations status 

and relate them to fire probability using remote sensing. There are many 

vegetation and moisture indices that have been used to monitor the vegetation 

status. However, the different indices do not work the same in different 

vegetation species. Thus, a study is necessary to determine the most 

appropriate index to monitor the vegetation status in our study area. For 

the present work we have selected eight indices from the bibliography, and 

we have compared their feasibility as fire risk indicators. Overall, results 

show that we can use either, EVI or GEMI, to monitor fire risk in Galicia 

with a relative error of about 15%. 


Forest fires are an extremely destructive phenomenon for nature. In Spain, 

forest fires are very abundant. Prevention is an important issue in the fight 

against fire. There are a wide number of models which try to obtain a forest 

fire risk indicator using multiple variables. The aim of this work is to 

analyze the isolate potential of the information about the vegetation status 

as a new parameter to be included in the fire prediction systems in the 

Galicia region by the use of remote sensing techniques. 

Many works have studied the relation between the vegetation status and 

forest fires using spectral indices. Nevertheless, a universal relation 

between these indices and fire probability has not been found. Therefore it 

is necessary to obtain, for each case, the most appropriate index which best 

characterizes the changes in the vegetation status, and so, better represents 

the relation between forest fires and the vegetation status. 

109

110 

II - VALIDATION OF RS PRODUCTS FOR FIRE MANAGEMENT 

In this work we compare eight different spectral indices from the bibliography, 

obtained from MODIS sensor on board Terra satellite, as fire risk indicators 

in the Galicia region. 

2 - Study site and materials 

The study site in this work is the Galicia region placed at the Northwest of 

Spain. About the 70% of the Galician territory is occupied by forest areas. 

Despite its humid and rainy climate, Galicia is the Spanish region with the 

largest number of forest fires. 

The organisms for the management of fire prediction and extinction tasks 

use a grid as a basis dividing Galicia in 360 10-km side squares (UTM 29). 

Also, historic fire data are provided at the same spatial resolution. 

Therefore, information on vegetation stage has been adapted to this spatial 

resolution, using the same grid. As seen in Sánchez et al. (2009), the 

huge majority of fire events take place from February to October. Thus, the 

analysis is constrained to this period. The temporal resolution chosen is 16 

days, as the vegetation stage does not change significantly within this twoweek 

period. The vegetation information is obtained from the products 

MOD09 A1 and MOD13 Q1 provided by the MODIS service. 


3.1 - Pre-processing of the MODIS images 

In this work we use images from the MOD09 A1 product, (8 days temporal 

resolution and 500 m spatial resolution), as well as images from the MOD13 

Q1 product (16 days temporal resolution and 250 m spatial resolution). 

Within a satellite scene, there can be pixels with wrong values due to the 

presence of clouds or because of some problem with the detector. Therefore 

it is essential to perform a filtering process. We use the information contained 

in the quality band of each product and information of the Corine 

Land Cover for the filtering process. Afterwards, we carry out a filling. Then 

we make a composition of two consecutive images in order to obtain one 

image each 16 days. Finally we apply the 10x10 km grid and calculate a 

value for each cell. 

3.2 - Relation between fire frequency and the index variations 

The spectral indices chosen for the present study are: NDVI (Rouse et al., 

1974), SAVI (Huete, 1988), NDII (Hunt and Rock, 1989), GEMI (Pinty and 

Verstraete, 1992), NDWI (Gao, 1996), VARI (Gitelson et al., 2002), EVI 

(Huete et al., 2002) and GVMI (Ceccato et al., 2002). This selection is

Assessment of spectral indices derived from modis data as fire risk indicators in Galicia 111 

based on a bibliographic compilation of different works that have used 

these indices as indicators of the vegetation conditions. As shown in 

Sánchez et al. (2009), the fire frequency is represented versus the variation 

suffered by the indices during de previous period. 

The 50% of the data (odd years) are used for obtaining the relation, and 

the other 50% (even years) for the validation. 


From the previous analysis, only three of the indices were shown to fit a 

linear regression, these are: EVI, GEMI, and SAVI. Figure 1 shows the linear 

regression of these three indices. 

Figure 1 - Linear adjustment between the percentage of fire-affected cells and the index variations 

in the previous two weeks, for odd years. 

For the validation we use the even year data. We apply the equations 

obtained for each index to the variations suffered by the index, and compare 

the results obtained with the real fire data. Results of these adjustments, 

as well as a statistical analysis are included in table 1. Based on 

these results we conclude that both, GEMI and EVI, can be used to estimate 

the fire probability in a cell with an error about 15%. 

The organisms for the management of fire prediction and extinction tasks 

use graduated scales for fire risk prediction. After some proofs we observed 

that the most optimum classification is: High risk (∆index

112 


a b R 2 Bias RMSE RMSES RMSEU MAD MADP 

GEMI 0.12±0.18 -0.04±0.05 0.66 -0.004 0.06 0.010 0.06 0.04 14% 

EVI (MOD 09) 0.88±0.19 0.04±0.06 0.51 -0.0019 0.07 0.010 0.07 0.05 16% 

SAVI 1.5±0.2 -0.13±0.07 0.74 0.002 0.06 0.04 0.003 0.05 17% 

EVI (MOD 13) 0.93±0.17 0.01±0.05 0.61 -0.008 0.06 0.012 0.06 0.05 16% 

Table 1 - Statistics of the validation for the three indices. 


In this work we present a comparative study of spectral vegetation indices 

with the aim of obtaining a simple empirical model for the estimation of 

fire risk in the Galicia region. Eight different indices have been selected. 

The variations suffered by each one, in a two weeks period, are used as the 

entry parameter for the model. Three of the indices are shown to fit a linear 

regression with the fire probability: EVI, GEMI and SAVI. From the validation 

results we may conclude that the most appropriate indices for the 

fire risk estimation are GEMI and EVI, with a relative error of about 15%. 

In future works we will apply this model to other regions and include the 

surface temperature as an additional input of the model. 


This work has been financed by the Science and Innovation Spanish 

Ministry (Projects CGL2007-64666/CLI, CGL2008-03668/CLI, and Juan de la 

Cierva research contract of J.M. Sánchez). 

References 

Ceccato, P., Gobron, N., Flasse, S., Pinty, B., Tarantola, S., 2002. Designing 

a spectral index to estimate vegetation water content from remote sensing 

data: Part 1. Theoretical approach. Remote Sensing of Environment, 

82, 188-197. 

Gao, B.C., 1996. NDWI - a normalized difference water index for remote 

sensing of vegetation liquid water from space. Remote Sensing of 


Gitelson, A.A., Kaufman, Y.J., Stark, R., Rundquist, D., 2002. Novel algorithms 

for remote estimation of vegetation fraction. Remote Sensing of 


Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing 

of Environment, 25, 295-309. 

Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., & Ferreira, L.G., 

2002. Overview of the radiometric and biophysical performance of the

Assessment of spectral indices derived from modis data as fire risk indicators in Galicia 113 

MODIS vegetation indices. Remote Sensing of Environment, 83, 195-213. 

Hunt, E.R., & Rock, B.N., 1989. Detection of changes in leaf water-content 

using near-infrared and middle-infrared reflectances. Remote Sensing of 


Pinty, B. y Verstraete, M.M., 1992. GEMI: a non-linear index to monitor 

global vegetation from satellites. Vegetatio, 101, 15-20. 

Rouse, J.W., Hass, R.H., Schell, J.A. and Deering, D.W., 1974. Monitoring 

vegetation systems in the Grat Plains with ERTS. Proceedings, third Earth 

Resources Technology Satellite-1 Symposium, Greenbelt, NASA SP-351, 

pp. 309-317. 

Sánchez, J.M., Caselles, V., Bisquert, M.M., Paz Andrade, M.I., Legido, J.L., 

2009. Fire risk estimation from MODIS Enhanced Vegetation Index data. 

Application to Galicia region (northwest Spain). International Journal of 

Wildland Fires (under review).

RELATIONSHIPS BETWEEN COMBUSTION PRODUCTS AND THEIR SPEC- 

TRAL PROPERTIES IN FIRE-AFFECTED SHRUBLANDS 

R. Montorio 

University of Zaragoza, Zaragoza, Spain 

montorio@unizar.es 

F. Pérez-Cabello, A. García-Martín, V. Palacios & J. de la Riva 

University of Zaragoza, Zaragoza, Spain 

fcabello@unizar.es; algarcia@unizar.es; palacios@unizar.es; delariva@unizar.es 

Abstract: Fire severity is considered an influencing factor in the post-fire 

recovery dynamic of burnt areas. New research works point out the usefulness 

of studying the individual combustion products which constituted the 

burnt areas. This work aims for assessing the relationship between the main 

combustion products and hyperspectral data and also for comparing the 

sensitivity of the original reflectance values against the transformed data 

(first derivative and absorption features analysis). Statistically significant 

relationships are observed between the vegetation remains product and the 

three spectral datasets. The black carbon and ash products are found to be 

highly related with the first derivative dataset but not so good relationships 

are observed with the other two spectral datasets. These results indicate 

the higher sensitivity of the first derivative transformed data to combustion 

products and the usefulness of this spectral information to the 

assessment of fire severity. 


Forest dynamic is highly influenced by fire, explaining the need for analyzing 

the variables controlling postfire dynamic (Pérez and Moreno, 1998). 

Within these variables, severity has been commonly considered as a critical 

one to assess postfire effects due to its greater control in vegetation 

response and erosion processes (Miller and Yool, 2002). Satellite data have 

been proved suitable for detecting and mapping this variable because fire 

disturbances in vegetation and soil produce detectable changes in their 

spectral properties. In opposite to the traditional evaluation from spectral 

intervals of severity indices new research works point out the usefulness of 

estimating the presence of individual combustion products, especially those 

associated with known fire severity levels (Smith et al., 2005). In this 

framework, ground-level research works are a necessary first step to study 

the actual spectral properties of the solid combustion products and the sen- 

115

116 


sitivity of the spectral data to their presence. This work aims for assessing 

the relationship between the main post-fire surface materials which are 

consequence of the combustion process (black carbon, ash and vegetation 

remains) and hyperspectral data. In the same way, this work compares the 

sensitivity of the original reflectance values against the transformed data 

(first derivative and absorption features analysis). 

2 - Study area 

The specific fire study site is located in Peñaflor experimental station (PES) 

(Zaragoza, Spain). It is a south-facing slope of 12° placed in a semiarid 

environment with a Mediterranean shrubland. A 15x3m section in the lowest 

sector of the slope was burnt in an experimental fire allowing the fire 

to spread naturally to the remaining slope. 


3.1 - Obtaining of field data 

Two different techniques were used for the obtaining of field data: (1) high 

spatial resolution photography using a Reflex Nikon D100 digital camera 

and (2) field spectrometry with the field spectrometer Avantes AvaSpec 

which registers reflectance in the 400-1800 nm bandwidth with a 0.57 nm 

spectral sampling in the VIS-NIR range and a 3.5 nm spectral sampling in 

the SWIR range. 

The obtaining of both informations was made using a metallic structure 

(3x3x2m) with a mobile system to hold both devices in such a way that 

both registered, from the nadir, the same surface. To control the surface 

registered by the spectrometer, its field of view was restricted to a 10º 

angle thus generating a circular surface of capture of 30 cm of diameter 

(Figure 1). From every photograph we retained only this central surface 

avoiding distortion problems and metallic structure shadows. 

We applied a regular sampling in the rectangular section and a random one 

in the remaining burnt slope, building a database of 305 points.

Relationships between combustion products and their spectral properties in fire-affected shrublands 117 

Figure 1 - Experimental design. 

3.2 - Post-treatments of field data 

To apply subsequent treatments in a homogeneous manner we built a mosaic 

file from the 305 circular-shaped files. To quantify the combustion products 

we applied a supervised classification process selecting a maximumlikelihood 

method. From this classification we obtained the percentages of 

the three combustion products studied in this research: (1) ash, where fuel 

had undergone a complete combustion; (2) charcoal, where an unburned 

fuel component remains; and (3) vegetation remains, vegetation non 

affected by the fire. 

From direct fieldwork we obtained reflectance data from the VIS and NIR 

regions. Applying hyperspectral techniques to this original information we 

obtained derived information: (1) the standard first derivative spectra 

(FDS) and (2) the absorption band depth (BD). 

The FDS transformation can be defined as the reflectance change rate for a 

specific spectral distance along the different wavelengths considered 

(Dawson and Curran, 1998). This technique emphasizes the wavelengths 

were the spectral curve has rough changes in its form. 

FDS λ(j) = (R λ(j+1) – R λ(j) ) / (λ (j+1) - λ (j) ) 

where FDS λ(j) is the first derivative spectra in the wavelength j; R is the 

reflectance value; (j) and (j+1) are the wavelengths. 

The continuum removal (CR) is the technique applied to obtain the BD data. 

From the continuous reflectance spectrum we identified nine absorption 

features, five from a vegetation spectrum and four from a charcoal spectrum. 

By applying the CR equation the continuum-removed spectra were calculated 

and BD (li) was obtained in each wavelength (Mutanga et al., 2004): 

CR (λi) = R (λi) / Rc λ(i) 

BD (λi) = 1 – CR (λi)

118 


where CR (λi) is the differential absorption in each wavelength (λi), R (λi) is 

the reflectance, Rc λ(i) is the reflectance of the continuous tendency. 

4 - Results 

The three sets of spectral data, reflectance (R), FDS and BD, were analyzed 

using a stepwise multiple linear regression routine. The number of points 

of the data set (n=305) allowed choosing until six independent variables. 

Results are summarized in the Table 1. 

Statistically significant relationships are observed between the vegetation 

remains product and the three spectral datasets (adjusted r 2 > 0.8). The 

results of the charcoal product show more differences between spectral 

datasets although the values are quite good in all of them: adjusted r 2 near 

0.6 in the R and BD datasets and higher than 0.7 in the FDS. Differences 

between datasets are also observed in the ash product although with this 

product the analysis with the BD dataset does not achieve good results 

(adjusted r 2 = 0.46), relationships with R and FDS shown values of 0.67 and 

0.74, respectively. 

Table 1 - Summary of the regression results. 


This research has shown the validity of the combined used of high spatial 

resolution photography and field spectrometry to better the understanding 

of fire severity variable. According to the results we can state some conclusions 

related to the usefulness of the different spectral dataset. The calculation 

of transformed hyperspectral data is justified only for ash products 

and especially for the charcoal because vegetation showed the same adjust-

Relationships between combustion products and their spectral properties in fire-affected shrublands 119 

ed values with the three dataset (R, FDS and BD). Considering the results 

attained in the FDS dataset we can state that vegetation, charcoal and ash 

can be adequately estimate from spectral data. As these products are representative 

of different fire severity levels, we can conclude that from FDS 

we are able to improve the evaluation of this variable. 

References 

Dawson, T.P., Curran, P.J., 1998. Technical note: A new technique for interpolating 

the reflectance red edge position. International Journal of 

Remote Sensing, 19: 2133-2139. 

Miller, J.D., Yool, S.R., 2002. Mapping forest post-fire canopy consumption 

in several overstory types using multi-temporal Landsat TM and ETM+ 

data. Remote Sensing of Environment, 82: 481-496. 

Mutanga, O., Skidmore, A.K., 2004. Hyperspectral band depth analysis for 

better estimation of grass biomass (Cenchrus ciliaris) measured under 

controlled laboratory conditions. International Journal of Applied Earth 

Observations and Geoinformation, 5: 87-96. 

Pérez, B., Moreno, J.M., 1998. Methods for quantifying fire severity in 

shrubland-fires, Plant Ecology, 139: 91-101. 

Smith, A.M.S., Wooster, M.J., Drake, N.A., Dipotso, F.M., Falkowski, M.J. 

and Hudak, H.T., 2005. Testing the potential of multispectral remote 

sensing for retrospectively estimating fire severity in African Savannahs, 

Remote Sensing of Environment, 97: 92-115.

MULTI-CRITERIA FUZZY-BASED APPROACH FOR MAPPING BURNED 

AREAS IN SOUTHERN ITALY WITH ASTER IMAGERY 

Abstract: Burned area mapping algorithms developed for satellite images 

often rely on the use of spectral/vegetation indices for discriminating 

between burns and other surfaces. Since the choice of one or more indices 

over the others might be subjective, we propose a semi-automated 

approach for integrating indices into a synthetic indicator (score) of likelihood 

of burn based on fuzzy set theory. The mapping method is based on 

a region growing algorithm that uses seed pixels identified by a conservative 

threshold on the synthetic score. The algorithm was tested on ASTER 

images and validated with an independent data set. Burned area maps for 

the Calabria region are presented and discussed. 


M. Boschetti 1 , D. Stroppiana 1 & P.A. Brivio 1 

1 CNR-IREA, Institute for Electromagnetic Sensing of the Environment, 

Milan, Italy 

boschetti.m@irea.cnr.it; stroppiana.d; brivio.pa 

Forest fires in Italy destroy more than 50.000 ha of natural vegetation 

every year. In this environment most of the burned surfaces are smaller 

than ten hectares limiting the use of widely available moderate resolution 

data and well accepted multi-temporal mapping methods. It is therefore 

necessary in supporting fire monitoring activity to exploit high/very high 

resolution (HR) multispectral data. Spectral Indices (SIs) are often proposed 

as a suitable mapping method with single post fire HR data. 

However, no agreement exists on the index that performs better than the 

others and in which situation as to be preferred (Stroppiana et al., 2009; 

Lasaponara et al., 2006). The objective of our research was to evaluate the 

performance of some widely used spectral indices in separating burned surfaces 

from other targets and to develop a semi-automated algorithm for 

mapping fire affected areas based on the integration of different SIs and 

on fuzzy set theory. 

121

122 



The analyses were carried out on AST07 surface reflectance in the VIS/NIR 

with a resolution of 15 m and SWIR wavelengths with a spatial resolution 

of 30 m. Three sets of ASTER data acquired over southern Italy in different 

sites were used: 

• Four scenes (08/17/02 - 06/10/03 - 06/24/03 - 08/11/03) to perform 

SIs separability analysis by selecting pixels for burn (~1200) and unburn 

classes (~800) (Stroppiana et al., 2009). 

• Four scenes (20/07/01 - 08/09/01 - 05/09/03 - 14/09/04) to train the 

algorithm by selecting 15,000 pixels of burn by photo-interpretation. 

• One scene (28/07/2003) to assess method performance. 

2.1 - SI selection and separability analysis 

The analysis was conducted on five different Spectral Indices commonly 

used for fire monitoring, see table, using images of the first data set. We 

also included the NIR ASTER band 3 that is very sensitive to the damage 

caused to vegetation by fire. 

Index Formula 

NBR Normalized Burn Ratio (ρNIR - ρSWIR 8 ) /(ρNIR - ρSWIR 8 ) 

BAI Burned Area Index [(0.1 - ρRED) 2 +(0.6 - ρNIR) 2 ] -1 

MIBI Mid-Infrared Burn Index 10ρSWIR 5 - 9.5ρSWIR 4 + 2 

CSI Char Soil Index ρNIR/ρSWIR 8 

SAVI Soil-Adjusted Veg. Index (ρNIR - ρRED)(1.5)/(ρNIR + ρRED + 0.5) 

ρRED, ρNIR, ρSWIR 4 , ρSWIR 5 , and ρSWIR 8 is the reflectance in ASTER bands 2, 3, 4, 5, and 8, respectively 

Table 1 - Spectral indices used in this study. 

Separability (S) (Kaufman and Remer, 1994) between burns and other surfaces 

was computed using the formula S = |µ i,b - µ i,u |/(σ i,b + σ i,u ) where µ i,b 

and σ i,b are the mean and standard deviation of burned areas and µ i,u and 

σ i,u are the mean and standard deviation of unburned surfaces, respectively. 

2.2 - Fuzzy functions definition and mapping method 

In the adopted partially data-driven approach (Robinson, 2003) the histograms 

of the values of each SI for the burn training pixels (data set 2) 

were interpolated with a sigmoid curve. These functions map SI values into 

the [0,1] domain where the higher likelihood of burn provides a score closer 

to 1. The membership degrees of the indices are then combined into a

Multi-criteria fuzzy-based approach for mapping burned areas in southern Italy with ASTER imagery 123 

synthetic score by applying a weighted average (WA) operator with weights 

derived from results of separability analysis. The WA map is used to extract 

burn seeds (WA>0.7 and cluster size >0.5 ha) and the final burned areas 

map is obtained by applying a region growing algorithm (WA_RG). All the 

maps produced are then filtered with a 3x3 median filter and only polygons 

greater that 1 ha are retained as fire affected areas. Validation was performed 

on the third data set. WA_RG maps were compared to polygons identified 

from visual interpretation to derive the error matrix and the accuracy 

measurements (Congalton, 1991). 


Separability analysis showed that NBR is the index that better separates the 

burn class from other surfaces and consequently presents the highest weight 

(21%) in the final score (WA). CSI (19%) and SAVI (17%) resulted also 

important while the lower weight was assigned to MIRBI (13%). Finally BAI 

and NIR complete the weight vector with about 15% of importance each. 

NBR BAI NIR CSI SAVI MIRBI 

Vegetation 1.99 1.63 1.55 1.74 1.87 1.38 

Shadow 1.74 0.53 0.42 1.67 1.19 0.20 

Soil 1.10 1.42 1.50 1.01 0.92 1.54 

AVG 1.61 1.20 1.16 1.47 1.33 1.04 

Weight 21% 15% 15% 19% 17% 13% 

Table 2 - Single classes and average (AVG) separability score for each SI. Relative importance 

and weights to be used in the WA operator were derived from average values. 

The parameters of the sigmoid fuzzy functions interpolating the SI histograms 

are reported in table. The functions have been constrained to f=0 

for values above or below which pixels are not considered burned (see 

Stroppiana et al., 2009). 

Sigmoid function: * SI - µ −1 ** SI - µ −1 

f = 1 + exp [( ---------------)] f = 1 + exp [( ---------------)] 

NBR* BAI** NIR* CSI* SAVI* MIRBI** 

µµ 0.20 63.90 0.20 1.34 0.17 1.49 

σσ 0.05 7.62 0.00 0.13 0.01 0.05 

threshold ≤ -0.3 - ≤ 0.1 ≤ 0.55 ≤ 0.05 ≥ 2.0 

σ σ 

Table 3 - Parameters of the membership functions derived by interpolation of training data.

124 


Figure shows the final burn map for the Aspromonte region (28/07/2003) 

produced by the region growing algorithm on the basis of seed pixels identified 

from the weighted average synthetic score map. It is interesting to 

notice that the fires in this area occurred mainly in the west coast in a 

agro-environment close to urban settlements. Fires impact on both natural 

vegetation (forest and shrub) and agricultural land with almost the same 

percentage. The Aspromonte National Park, that is the central part of the 

study area, resulted less affected by fires (112.5 ha vs 1960.5 ha) probably 

as a consequence of protection activity and remoteness of some areas. 

Validation shows very satisfying results: the WA_RG map has OA and Kappa 

coefficient respectively of about 99.5% and 0.73 and presents a low commission 

error of 2.7%. 

Burned area map for Aspromonte. Burn polygons are overlaid to the Corine Land Cover. 


The proposed fuzzy aggregation method allowed to exploit the ability of 

different SI in detecting burned areas. When the approach is applied to 

independent data the proposed method performs well with high accuracy.

Multi-criteria fuzzy-based approach for mapping burned areas in southern Italy with ASTER imagery 125 

The approach described here allows in a more automatic way to map fire 

affected areas in a conservative approach that produces very few false 

alarms (low commission error). 

References 

Congalton, R.G., 1991. A review of assessing the accuracy of classifications 

of remotely sensed data. Remote Sens. Environ., 37, 35-46. 

Kaufman, Y.J. and Remer L.A., 1994. Detection of forest using Mid-IR 

reflectance: an application for aerosol studies. IEEE Trans. Geosci. 

Remote Sens., 32, (3), 672-683. 

Lasaponara R., 2006. Estimating spectral separability of satellite derived 

parameters for burned areas mapping in the Calabria region by using 

SPOT-Vegetation data. Ecological Modelling, 196, 265-270. 

Robinson, P.B., 2003. A perspective on the fundamentals of fuzzy sets and 

their use in Geographic Information Systems. Transactions on GIS 7(1), 

3-30. 

Stroppiana, D., Boschetti M., Zaffaroni P. and Brivio, P.A., 2009. Analysis 

and interpretation of spectral indices for soft multi-criteria burned area 

mapping in Mediterranean regions. IEEE Geosci. Remote Sens. Letters, 6 

(3), 499-503.

OPERATIONAL USE OF REMOTE SENSING IN FOREST FIRE MANAGEMENT 

IN PORTUGAL 

Abstract: Like other southern regions of Europe, Portugal has experienced 

a dramatic increase in fire incidence during the last few decades that has 

been attributed to modifications in land-use (e.g. land abandonment and 

fuel accumulation) as well as to climatic changes (e.g. reduction of fuel 

humidity) and associated occurrence of weather extremes. Wildfire activity 

also presents a large inter-annual variability that has been related to 

changes in the frequency of occurrence of atmospheric conditions 

favourable to the onset and spreading of large-fires. The aim of the present 

study is to provide evidence that levels of fire risk prior to the beginning 

of the fire season may be anticipated by using information about vegetation 

stress at the beginning of the fire season (e.g. as derived from vegetation 

indices based on remote-sensed data) and combining it with longrange 

weather forecasts (e.g. in the form of fire risk indices based on summer 

climatic outlooks). 


C.C. DaCamara 

CGUL, IDL, University of Lisbon, Lisbon, Portugal 

cdcamara@fc.ul.pt 

T.J. Calado, C. Gouveia 

University of Lisbon, Lisbon, Portugal 

mtcalado@fc.ul.pt; cmgouveia@fc.ul.pt 

In the European context, Portugal presents the highest number of fire 

occurrences and has the largest areas affected by wildfires (Trigo et al., 

2006). In this respect, vegetation is known to play an important role since 

it drives fuel accumulation, which exerts a control on fire behavior, especially 

in the case of fires burning under less severe weather conditions. 

However, meteorological and climatic factors play a crucial role in fire 

behavior. As pointed out by Pereira et al. (2005), more than 2/3 of the 

inter-annual variability of burned area is explained by meteorological factors 

namely i) the temperature and precipitation regimes of the spring preceding 

the fire season and ii) the occurrence during the fire season of circulation 

patterns of short-duration that induce extremely hot and dry spells 

over western Iberia. 

Using remote sensed information to monitor vegetation stress during the 

127

128 


pre-fire season may therefore represent an added value for fire management 

in Portugal (Gouveia et al., 2009). In fact, the Iberian Peninsula is recurrently 

affected by drought episodes and a combined effect of lack of precipitation 

over a certain period with other climatic anomalies, such as high 

temperature, high wind and low relative humidity over a particular area may 

result in reduced green vegetation cover. In fact, major fire events in Iberia 

are frequently preceded by drought periods, reflected by the decrease of 

vegetation activity before the fire episode. This effect is further aggravated 

by a quick response of dead fuels to changing meteorological conditions, 

such that even short periods of drought may lead to an exponential 

increase in fire risk. 

2 - Results 

Burned area (BA) data covering the period 1982-2006 were obtained from 

the official data supplied by the National Forest Authority (AFN). Figure 1 

presents the inter-annual variability of burned area over Portugal during the 

considered period. The close resemblance between the annual and the summer 

curves implies that the annual fire regime in Portugal is mostly dominated 

by events occurring in July and August. Vegetation stress prior to the 

begining of the fire season was assessed using averaged values in May of 

the Normalized Difference Vegetation Index (NDVI) as obtained from the 

Global Inventory Modelling and Mapping Studies (GIMMS) database (Zhou 

et al., 2001), covering the considered 25-year period. Values of NDVI in May 

were then spatially averaged over the forested regions of Portugal as identified 

by downgrading the Global Land Cover (GLC) for the year 2000 

(Bartholome et al., 2002) to the GIMMS scale. Meteorological fire danger in 

summer 1982-2006 was estimated using July + August means of the Daily 

Sevirity Rating (DSR), an index that may be derived from the so-called 

Canadian Fire Weather Index (FWI) system (van Wagner, 1987). DSR is a 

numeric rating of the difficulty of controlling fires and reflects the expected 

efforts required for fire suppression (CFS, 2007). 

Regression tree analysis was used to assess how yearly amounts of BA in 

the fire season could be related with vegetation stress prior to the fire season 

and meteorological conditions during the fire season. Regression tree 

(Breiman et al., 1984) is a nonparametric technique that is very effective 

in selecting from given variables the interactions among them that are 

most important in determining the outcome variable to be explained. 

Figure 2 presents a scatter plot of NDVI in May versus DSR in summer 1982- 

2006 (left panel) together with the recorded amounts of BA during the 

same period(right panel). Figure 3 provides a schematic overview of the 

obtained regression tree. It may be noted that years of weak activity are 

associated to both low values of NDVI in May (i.e. low biomass) and DSR in 

summer (i.e. meteorological conditions that do not favour the onset and 

spreading of fire). The largest event (in 2003) is, on the contrary, associ-

Operational use of remote sensing in forest fire management in Portugal 129 

ated to large values of both variables. The role of NDVI is also particularly 

important in controlling the impact of high values of DSR on the BA 

amounts. Finally the apparent tendency of not having high values of NDVI 

associated to high values of DSR may reflect the role of soil moisture in 

controlling the impact of extreme hot and dry spells on the onset and 

spreading of wildfires. 

References 

Bartholomé, E., Belward, A., Achard, F., Bartalev, S., Carmona-Moreno, C., 

Eva, H., Fritz, S., Gregoire, J., Mayaux. P., Stibig, H., 2002. GLC 2000: 

Global Land Cover mapping for the year 2000. EUR 20524 EN, European 

Commission, Luxembourg. 

Breiman, L., Friedman, J. H., Olshen, R. A., Stone, C. J., 1984. Classification 

and regression trees. Monterey, Calif., U.S.A., Wadsworth, Inc. 

CFS, 2007: Canadian wildland fire information system. 

http://fire.nofc.cfs.nrcan.gc.ca/en/background/bi_FWI_summary_e.ph. 

Gouveia C., Trigo, R.M., DaCamara, C.C., 2009. Drought and Vegetation Stress 

Monitoring in Portugal using Satellite Data. Nat. Hazards Earth Syst. Sci., 

9: 185-195. 

Pereira, M.G., Trigo, R.M., DaCamara, C.C., Pereira, J.M.C., Leite, S.M., 2005. 

Synoptic patterns associated with large summer forest fires in Portugal. 

Agr. Forest Met., 129: 11-25. 

Trigo R.M., Pereira, J.M.C., Pereira, M.G., Mota B., Calado, M.T., DaCamara 

C.C., Santo, F.E., 2006. Atmospheric conditions associated with the exceptional 

fire season of 2003 in Portugal. Int. J. of Climatology 26 (13): 

1741-1757. 

van Wagner, C.E., 1987. Development and structure of the Canadian Forest 

Fire Index System. Canadian Forestry Service, Ottawa, Ontario, Forestry 

Technical Report 35. 

Zhou, L., Tucker, C.J., Kaufmann, R.K., Slayback, D.N., Shabanov, V., 

Myneni, R.B., 2001. Variations in northern vegetation activity inferred 

from satellite data of vegetation index during 1981 to 1999. J. Geoph. 

Res., 106: 20069-20083.

130 


Figure 1 - Inter-annual variability of burnt area respecting to annual values (thick line) and 

summer (i.e. July + August) values (thin line) over Portugal, during 1982-2006. 

Figure 2 - Scatter plot of NDVI in May vs. DSR in summer over forested areas in Portugal for 

the period 1982-2006 (left panel); summer amounts of burned areas in Portugal sorted in 

descending order (right panel) during the same period. Dotted lines (left panel) indicate cut 

points as obtained from regression tree analysis (see Figure 3). Symbols (both panels) identify 

the different classes of fire activity according to the different fitted response values of the 

regression tree model (see Figure 3); black inverted triangles weak activity [log(BA)=3.74]; 

grey circles moderately low activity [log(BA)=4.70]; black circles moderately high activity 

[log(BA)=4.86]; grey triangles high activity [log(BA)=5.24]; black triangles extremely 

high activity [log(BA)=5.56].

Operational use of remote sensing in forest fire management in Portugal 131 

Figure 3 - Results from regression tree analysis. The regression tree predicts the response values 

at the rectangular leaf nodes based on a series of questions inside the diamonds.

ESTIMATION OF NATIONAL FIRE DANGER RATING SYSTEM 10 HOUR 

TIMELAG FUEL MOISTURE CONTENT WITH MSG-SEVIRI DATA 

H. Nieto 1 , I. Aguado 1 , E. Chuvieco 1 , I. Sandholt 2 

1 Department of Geography, University of Alcalá. Alcalá de Henares, Spain 

hector.nieto@uah.es; inmaculada.aguado@uah.es; emilio.chuvieco@uah.es 

2 Department of Geography, University of Copenhagen. Copenhagen, Denmark 

is@geo.ku.dk 

Abstract: The moisture content of fuels is a key factor in both fire ignition 

and propagation. Fire meteorological indices are commonly based on temperature, 

relative humidity, solar radiation and wind speed, which are only 

measured in selected sites that usually are sparsely distributed. In this 

study we propose to use remote sensing data to estimate meteorological 

data at an adequate spatial and temporal resolution. We will make use of 

the Spinning Enhanced Visible and Infrared Imager (SEVIRI) sensor, 

onboard the Meteosat Second Generation (MSG) satellites, to estimate air 

temperature and relative humidity on an hourly basis. Air temperature and 

vapour pressure are combined to calculate Simard’s Equilibrium Moisture 

Content and the NFDRS 10 hour timelag fuel moisture content. A meteorological 

station located in the Cabañeros National Park (Spain) has been 

used to calibrate and validate the results during the year 2005. 


One of the factors in fire danger management systems is fuel moisture content 

(FMC), since it is a critical variable in fire ignition and propagation 

(Dimitrakopoulos et al., 2001; Rothermel, 1972). Unlike live vegetation, 

which can regulate water losses through stomatal closure, dead fuels tend 

to gain or lose moisture until equilibrium with surrounding atmosphere is 

achieved. This steady moisture content is called Equilibrium Moisture 

Content (EMC), and is primarily affected by temperature and relative humidity 

(Viney et al., 1991). 

A revision of different dead fuel moisture content models can be found in 

Viney (1991). Traditionally, these models have been applied with observed 

data from meteorological stations or with forecasted data from Numerical 

Weather Prediction (NWP) models. The spatial representation of meteorological 

stations is limited, since the network usually is very sparse and stations 

are usually located in agricultural or urban areas. On the other hand, 

133

134 


NWP can provide spatially distributed data at a reasonable resolution by 

downscaling and interpolating the surface meteorological variables 

(Aguado et al., 2007), but they have a limited accuracy due the downscaling 

and interpolation tasks, as well as that forecasted data may deviate 

from observations due to the stochastic nature of the atmosphere. 

Thermal infrared data has been proven to be useful to estimate air temperature 

and water vapor pressure (Goward et al., 1994), and thus our hypothesis 

is that it is possible to retrieve dead fuel moisture content from remote 

sensing data. Goward et al. (1994) proposed an algorithm to estimate air 

temperature with AVHRR data. This algorithm (hereafter called TVX) is 

based on the observed linear relationship between the Land Surface 

Temperature (LST) and a vegetation index (NDVI), as a measure of vegetation 

cover. If a linear regression can be obtained between LST and NDVI, 

the air temperature can be retrieved by extrapolating this line to a maximum 

cover NDVI (NDVI max ) since vegetation canopies barely deviates from 

air temperature in a few degrees. For more details about this algorithm the 

reader is addressed to Goward et al. (1994). On the other hand, most of the 

water vapor is concentrated in the lowest layers of the atmosphere, since 

the decrease of water vapor through the atmosphere follows a power law 

(Smith, 1966). Although several authors pointed out that the reliability of 

estimates with daily data decreases compared to longer periods (Bolsenga, 

1965; Schwarz, 1968), total precipitable water (W) has been related to 

daily surface humidity with remote sensing data (Goward et al., 1994). 

2 - Objective 

This study aims to estimate dead fuel moisture content through the 

retrieval or air temperature and relative humidity from remote sensing data. 

Calibration and validation data has been acquired from a meteorological 

station located in the National Park of Cabañeros (39.319758ºN, 

4.394824ºW). This station provides hourly data of temperature and relative 

humidity. Only data during spring and summer were used for calibration and 

validation since it is the most critical season for wildfire assessment. 

3 - Methods 

3.1 - Satellite processing 

Images from Meteosat Second Generation-Spinning Enhanced Visible and 

Infrared Imager (MSG-SEVIRI) were selected since it provides an excellent 

temporal resolution (15 minutes) at an adequate spatial sampling (3km at 

sub-pixel nadir). Bands centered in 10.8µm and 12.0µm for the Iberian 

Peninsula have been used together with the EUMETSAT cloud mask to produce 

daily estimates of W. We have exploited the air temperature daily cycle

Estimation of national fire danger rating system 10 hour timelag fuel moisture content with MSG-SEVIRI data 135 

of MSG-SEVIRI applying the algorithm proposed by Sobrino et al. (2008). 

With this approach, the dependence of both surface and air temperature is 

eliminated for the estimation of W. 

Air temperature was retrieved by means of the TVX algorithm. The NDVI was 

calculated from SMAC corrected bands centered in 0.6µm and 0.8µm. LST 

was retrieved with a split window algorithm (Sobrino et al., 2004). The 

algorithm was applied in a window of 7x7 pixels centered in our station, 

and adopting a NDVI max of 0.996. 

3.2 - Calculation of Equilibrium Moisture Content and 10h FMC 

The Equilibrium Moisture Content proposed by Simard (1968) was chosen 

due to its simplicity and the wide spectrum in which it has been applied 

(Aguado et al., 2007). The formulation only requires as inputs temperature 

and relative humidity and it does not take into account hysteresis effects. 

The calculation of the 10hr timelag FMC is straightforward by multiplying 

the EMC by 1.28 (Bradshaw et al., 1983). 

Relative humidity requires water vapor pressure e a as well as saturation 

vapor pressure e s . The latter is related to air temperature T following an 

exponential relationship (Allen et al., 1998). e a was estimated from remote 

sensing data, by empirical fitting between the retrieved W and the vapor 

pressure measured at the meteorological station during the year 2005. 


A linear function was found to be the most suitable between e a and W. Eq.2 

shows the fitted relationship (R 2 =0.56, p

136 



We have shown that it is feasible to assess dead fuel moisture content by 

means of remote sensing data. Air temperature and relative humidity have 

been estimated with MSG-SEVIRI data in a meteorological station in Spain. 

The main source of error is due to unexplained variance in the retrieval of 

the vapour pressure. Future research will be focused on improving the estimation 

of water vapour with precipitable water content. A more robust relationship 

will be addressed by increasing the number of meteorological data 

trying to cover the whole Spanish territory. 

N MAE RMSE a b R U bias U slope U error 

e a (kPa) 59 0.17 0.22 0.32 0.76 0.61 0.25 0.04 0.71 T 

(ºC) 364 2.51 3.37 1.84 0.85 0.89 0.20 0.08 0.72 

RH (%) 364 9.09 12.59 12.22 0.75 0.68 0.12 0.08 0.80 

10h (%) 364 2.12 3.28 3.85 0.64 0.62 0.08 0.15 0.77 

Table 1. Error measurements of retrieved parameters. N, number of elements; MAE, mean 

absolute error; RMSE, root mean square error; a and b, intercept and slope of the regression 

between observed versus predicted; r, Pearson correlation between observed and predicted; 

U bias , proportion of RMSE associated with mean differences between observed and predicted 

values; U slope , proportion of RMSE associated with deviations from the 1:1 line; U error , proportion 

of RMSE associated with unexplained variance. 

References 

Aguado, I., Chuvieco, E., Borén, R., Nieto, H., 2007. Estimation of dead fuel 

moisture content from meteorological data in Mediterranean areas. 

Applications in fire danger assessment. International Journal of Wildland 

Fire, 16, 390-397. 

Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: 

Guidelines for computing crop water requirements (p. 300). Rome, 

Italy: FAO. 

Bolsenga, S.J., 1965. The Relationship Between Total Atmospheric Water 

Vapor and Surface Dew Point on a Mean Daily and Hourly Basis. Journal 

of Applied Meteorology, 4, 430-432. 

Bradshaw, B.S., Deeming, J.E., 1983. The 1978 National Fire Danger Rating 

System. Technical documentation (p. 41). Ogden, Utah: USDA. 

Dimitrakopoulos, A., Papaioannou, K.K., 2001. Flammability assessment of 

Mediterranean forest fuels. Fire Technology, 37, 143-152. 

Goward, S.N., Waring, R.H., Dye, D.G., Yang, J.L., 1994. Ecological Remote- 

Sensing at OTTER: Satellite Macroscale Observations. Ecological 

Applications, 4, 322-343. 

Rothermel, R.C., 1972. A Mathematical Model for Predicting Fire Spread in 

Wildland Fuels. Ogden, Utah: USDA, Forest Service. 

Schwarz, F.K., 1968. Comments on “Note on the Relationship between Total

Estimation of national fire danger rating system 10 hour timelag fuel moisture content with MSG-SEVIRI data 137 

Precipitable Water and Surface Dew Point”. Journal of Applied 

Meteorology, 7, 509-510. 

Simard, A.J., 1968. The moisture content of forest fuels - A review of the 

basic concepts (p. 47). Ottawa, Ontario: Forest Fire Research Institute. 

Smith, W.L., 1966. Note on the Relationship Between Total Precipitable 

Water and Surface Dew Point. Journal of Applied Meteorology, 5, 726- 

727. 

Sobrino, J.A., Romaguera, M., 2004. Land surface temperature retrieval 

from MSG1-SEVIRI data. Remote Sensing of Environment, 92, 247-254. 

Sobrino, J.A., Romaguera, M., 2008. Water-vapour retrieval from Meteosat 

8/SEVIRI observations. International Journal of Remote Sensing, 29, 

741-754. 

Viney, N.R., 1991. A Review of Fine Fuel Moisture Modelling (pp. 215-234). 

Viney, N.R., Catchpole, E.A., 1991. Estimating Fuel Moisture Response Times 

From Field Observations. International Journal of Wildland Fire, 1, 211- 

214.

GLOBAL MONITORING OF THE ENVIRONMENT AND SECURITY: 

A COMPARISON OF THE BURNED SCAR MAPPING SERVICES OF 

THE RISK-EOS PROJECT 

Abstract: The RISK-EOS project of the European Space Agency started in 

2003 under the framework of the Global Monitoring for Environment and 

Security (GMES) initiative, with the objective to establish a network of 

European service providers for the provision of geo-information services in 

support to the risk management of meteorological hazards. The Fire component 

of RISK-EOS project features as the main element, the Burn Scar 

Mapping (BSM) service, which provides seasonal mapping of forests and 

semi-natural burned areas at high spatial resolution (minimum mapping 

unit of 1 to 5 ha). The RISK-EOS BSM service builds on the achievements 

of ITALSCAR, a demonstration project for the yearly mapping of burned 

areas in Italy, using LANDSAT TM. The major objectives of the BSM service 

are to provide post-crisis information on the vegetated areas affected by 

wildfires to assess the damages and provide a baseline for recovery and 

restoration planning. A good knowledge of the land use/cover changes 

helps the forestry authorities and the fireguards to better assess the risk of 

fire ignition and fire spread in order to better allocate efforts for fire prevention 

and suppression. The BSM service has been provided by different 

providers in Portugal (Critical Software and ADISA), Spain (TELESPAZIO and 

INSA), France (TELESPAZIO and Infoterra France), Italy (TELESPAZIO) and 

Greece (NOA ISARS) and has been harmonised across countries. This paper 

briefly describes the methodologies applied by each service chain and compares 

the classification results on a specific burn scar which occurred in 

Portugal during the 2007 fire season. 

1 - Summary 

M. Paganini 1 , O. Arino 1 

A. Priolo 2 , G. Florsch 3 , Y. Desmazières 4 

C. Kontoes 5 , I. Keramitsoglou 5 , R. Armas 6 , A. Sá 7 

1 ESA ESRIN, Frascati, Italy, Marc.Paganini@esa.int; Olivier.arino@esa.int 

2 Telespazio, Palermo, Italy, Agata.Priolo@telespazio.com 

3 Infoterra France, Toulouse, France, geraldine.florsch@infoterra-global.com 

4 Astrium Satellites, Toulouse, France, yves.desmazieres@astrium.eads.net 

5 NOA, Athens, Greece, kontoes@space.noa.gr; ikeram@space.noa.gr 

6 Critical Software, Coimbra, Portugal, rgoncalves@criticalsotware.com 

7 Technical University of Lisbon, Lisbon, Portugal, anasa@isa.utl.pt 

In the scope of the RISK-EOS project, some Burn Scar Mapping (BSM) services 

have been delivered in five different countries in Southern Europe: 

Portugal, Spain, France, Italy and Greece, using four different service 

chains: Telespazio, Infoterra France, ADISA and NOA. 

139

140 


2 - BSM service chains methodologies 

The following chapters present the main characteristics of the RISK-EOS 

BSM service chains. 

2.1 - Telespazio Service Chain 

The BSM chain applied by TPZ can be summarized as follows: 

• Pre-processing with geometric and radiometric correction, and cloud 

mask generation; 

• Core Processing for the automatic generation of preliminary BSM polygons 

with: 

- Highlightness of the spectral characteristics of burned areas using 

vegetation indexes (NDII, BAI and GEMI) and the hue component 

of the IHS transformation (bands 7-4-3); 

- Identification of the burned core pixel based on a multi-threshold 

analysis between the pre and post-fire images, and inter-annual 

and intra-annual comparisons; 

- Automatic delineation of the burned area perimeter, starting with 

the identification of pixels correlated to the core pixels of the 

burned area. This process is based on the application of two parallel 

algorithms: regions growing and watershed classification. 

• Post-processing for final refinement of the burned area perimeter with 

photo interpretation and support of ground truth data when available. 

2.2 - Infoterra France Service Chain 

The BSM chain applied ITF is based on the following steps: 

• Pre-processing with atmospheric and radiometric corrections; 

• Identification of cloud masks; 

• Extraction of the biophysical parameters (GLCV - Green cover fraction; 

CSH - Canopy shade factor; SB- Soil brightness; DCL -Cloud reflectance); 

• Development of a decision tree with the biophysical parameters in order 

to identify seed pixels; 

• Application of a region growing algorithm in order to have a fist version 

of BSM product; 

• Post-processing of the first version of BSM by visual interpretation and 

manual edition. 

2.3 - ADISA Service Chain 

The BSM chain applied by ADISA is a multi-temporal approach based on differences 

between a pre and a post-fire dataset using a classification tree

Global monitoring of the environment and security: a comparison of the burned scar mapping services of the RISK-EOS project 141 

algorithm: 

• Pre-processing in order to get geometrically and atmospherically corrected 

images; 

• Masking out clouds and cloud shadows (automatic process followed by 

on-screen edition); 

• Use of the spectral Landsat TM bands and normalized difference band 

combinations (spectral indices: NDVI, GNDVI and VI7); 

• Development of a supervised burned area classification tree; 

• Application of the set of rules to the input variables; 

• Map filtering to reduce noise and remove small unburned patches inside 

burned areas; 

• Vector edition and application of a pre-defined minimum mapping unit 

threshold. 

2.4 - NOA Service Chain 

The NOA processing chain is a fixed thresholding approach that relies on 

automatic processing of spectral indices (NBR, NDVI, multi date NDVI and 

ALBEDO), as well as radiometric change vector analysis: 

• Satellite data radiometric normalisation, geo-referencing and mosaicking. 

Cloud, water and shadow mask generation; 

• Calculation of radiometric change vectors and generation of change/ 

no-change pixel masks; 

• Derivation of uni- or multi-temporal spectral indices and definition of 

the appropriate index thresholds (sensor and area specific); 

• The thresholded spectral indices are then combined in order to achieve 

a first separation between burnt and unburned areas. To resolve any 

ambiguities, the burn scars are compared against the derived change 

pixel map (output of the radiometric vector change analysis); 

• Removal of pixel noise using a median filter, and elimination of objects 

smaller than the specified minimum mapping unit (MMU); 

• Generation of GIS compatible burn scar polygons and enhancement of 

their thematic value. 

3 - Comparison of the different methodological approaches 

The four service chains have similar approaches, applying vegetation indexes 

and band combinations in a multi-temporal analysis, followed by a 

supervised classification performed with a decision tree. In order to share 

the information and experiences, a workshop was performed within the 

scope of RISK-EOS project. An analysis of the different service chains 

approaches was performed using as a reference a forest fire that occur in 

the centrer of Portugal during the 2007.

142 


3.1 - Common Problems 

After analyzing the different approaches and results, several common issues 

were found: 

Difficulties in mixed land use types separation (mixing of forests and seminatural 

lands with clear cuts, permanent crops, agriculture, etc.) given the 

coarser spatial resolution of the satellite data used (30m). 

Decision of Burnt/Not Burnt boundary (ancillary data need); 

Commission errors due to non-distinguishable spectral data; 

Availability of the proper images regarding the acquisition dates. Images 

should be clean of clouds and atmospheric contamination and they should 

be acquired at the end of the fire season; 

Differences between smoothed boundaries (generalization) and raster looking-like 

boundaries; 

Availability of the proper “ground truth” data for classifier training and reliable 

accuracy assessment. 

3.2 - Results 

The results obtained by the different methodologies were very similar in 

terms of burn scar shape, with small difference appearances due to differences 

in the post-processing of the four service chains, such as boundary 

smoothness or removal of small unburned islands. Table 1 presents the main 

characteristics of the different BSM service chains and the results obtained 

during the RISK-EOS project. 

Service Chain Input Data Validation Data Detection Efficiency Rate 

TPZ Landsat 5 TM 

SPOT 4 

IRS-P6 

ITF Landsat 5/7 TM 

SPOT 2,4, 5 

F-2, K-2 

ADISA Landsat 5 TM 

SPOT 4 XS 

NOA Landsat 5 TM 

SPOT 4 XS 

F-2 P&XS 

AIB Fire logs and/or GPS surveys in 

Italy. 

Promethee surface in France. 

Fire logs or GPS perimeters in Spain. 

GPS contours 

Promethee database 

GPS surveys and data acquired by 

cameras installed in surveillance 

towers 

Existing burn scar drawings on 

1:50.000-scale analogue maps 

GPS surveys of specific reference 

fires 

Table 1 - Main characteristics of the different BSM service chains. 

Between 77,25 - 90,0 % 

for fires ≥ 2ha 

98% for fires > 1 ha with 

SPOT 

84,5 % for fires ≥ 2ha 

Between 77 - 93 % for 

fires ≥ 1ha

Global monitoring of the environment and security: a comparison of the burned scar mapping services of the RISK-EOS project 143 


All service providers have used similar methodological approaches. Due to 

specific requests of the users, different post-processing techniques were 

applied and thus the final results were slightly different. The same problems 

were encountered by all service providers and were mainly resulting from 

the image acquisition dates and ground truth data applied for adjustments 

in the post-processing phase and validation. Due to its spectral and spatial 

resolutions, Landsat TM has been the sensor most widely used to map burnt 

areas. However, the high risk of reaching the end of its life in a short period 

of time causes the very high dependency on this sensor. Because of this 

risk, the BSM service providers have adapted their production chain and are 

now using SPOT, Formosat, IRS and other satellite sensor images which 

include near infrared and red spectral bands. However, these sensors have 

limitations concerning the needed spectral information and the full extended 

European coverage (Formosat), and do not seem to be the most suitable 

satellite sources for this type of service. The ESA Sentinel 2 (to be launched 

in 2012) with its large swath (300km), its high frequency revisit and its 

good spectral and spatial resolutions (13 optical channels and 10m resolution) 

will solve most of the identified limitations. This comparison study 

provided concrete evidence that the four methods offer advanced burnt 

area mapping in terms of cost and accuracy, compared to conventional field 

methods and/or aerial photo-interpretation. The accuracy results and the 

overall experience gained through the RISK-EOS project suggest that the 

satellite-based mapping methods replace the conventional methods at an 

accuracy level far exceeding the existing mapping standards established by 

Forestry Services in many Mediterranean countries. 

References 

Armas, R., Desmazières, Y., 2008. Meeting Minutes of the Burn Scar 

Mapping <strong>Workshop</strong>, Risk-EOS Project, Lisbon. 

Di Federico A., Priolo A., 2008. Service Portfolio Specification, Risk-EOS 

Project. 

Kontoes C.C., 2008, Operational Land Cover Change Detection Using 

Change-Vector Analysis, International Journal of Remote Sensing, Vol. 

29, No. 16, pp.4757-4779, DOI:10.1080/01431160801961367. 

Kontoes C.C., Poilvé H., Florsch G., Keramitsoglou I., Paralikidis S., 2009, A 

Comparative Analysis of a Fixed Thresholding vs. a Classification Tree 

Approach for Operational Burn Scar Detection and Mapping, submitted 

for publication in the International Journal of Applied Earth Observation 

and Geoinformation, in press, DOI: 10.1016/j.jag.2009.04.001 

Paganini, M., Arino O., 2003. ITALSCAR, a Regional Burned Forest Mapping 

demonstration project in Italy IEEE Transactions on Geoscience and 

Remote Sensing: 1290-1292. 

RISK-EOS website, 2008, http://www.riskeos.com .

EFFECTS OF FIRE ON SURFACE ENERGY FLUXES IN A CENTRAL SPAIN 

MEDITERRANEAN FOREST. GROUND MEASUREMENTS AND SATELLITE 

MONITORING 

J.M. Sánchez 1 

1 Applied Physics Department, Un Castilla-La Mancha, Albacete, Spain 

juanmanuel.sanchez@uclm.es 

E. Rubio 1,2 , F. R. López-Serrano 3 , V. Caselles 4 & M.M. Bisquert 4 

2 Regional Development Institute, Un. Castilla-La Mancha, Albacete, Spain 

3 School of Agronomy Engineering, Un. Castilla-La Mancha, Albacete, Spain 

4 Earth Physics and Thermodynamics Department, Un. Valencia, Burjassot, Spain 

Abstract: Forest fires are one of the main agents involved in the change of 

structure and function of ecosystems. In this work we used a set of 5 

Landsat 5 Thematic Mapper (TM) images, of the years 2007-2008, covering 

an area of mediterranean forest and shrubs, affected by a fire in the summer 

of 2001. Two control areas (non-burned) were established, representative 

of the pre-fire conditions in the burned areas. The simplified twosource 

model STSEB was applied to elaborate instantaneous energy flux 

maps, at the time of the satellite overpass. A Bowen station placed in the 

study site permitted a previous validation of the results. Regarding the 

energy fluxes the most remarkable is the increase of more than 150 W m -2 

in sensible heat flux at instantaneous scale, and 40 W m -2 at daily scale, 

and the decrease of more than 250 W m -2 at instantaneous scale, and 60 W 

m -2 (2.1 mm/day) at daily scale, in actual evapotranspiration, observed in 

the forested area. In the shrubs area, the fire effect is almost negligible 

after 6 years, since the vegetation regenerates. 


Forest fires are highly destructive for nature, affecting the landscape, the 

natural cicle of the vegetation, and the structure and functioning of ecosystems. 

Beyond that, they also provoke changes in the local and regional 

meteorology, and particularly in the surface energy fluxes regimen. There is 

an increasing concern among the scientific community about the effect of 

forest fires on climate change at this point (Randerson et al., 2006). 

Remote sensing techniques allow us to estimate surface energy fluxes over 

large areas. In the present work we use the Simplified Two-Source Energy 

Balance model (STSEB), developed by Sánchez et al., (2008a), applied to 

high resolution imagery obtained from Landsat 5 Thematic-Mapper (TM). 

The objective of this study is to quantify the effect of a forest fire in terms 

of net radiation, and soil, sensible, and latent heat fluxes in two different 

ecosystems, mature pine forests and shrublands. 

145

146 


2 - Study site and materials 

The study site is a forest area, with some inserted crop fields, located in 

Almodóvar del Pinar, Cuenca (39º 40´N, 1º 50´W, 950 m above sea level). 

Climate is mediterranean, with warm and dry summers, and cool winters. 

The dominant tree species is Pinus pinaster Ait., but many other species 

coexist. In the summer of 2001, a fire affected a total of 172 ha, of which 

113 ha were covered by pines and 59 ha by shrubs (Fig. 1). After the fire, 

the species Quercus ilex L. was occupying the burned area as a consequence 

of a natural regeneration process. Four 125x125 m test sites, two inside and 

two outside the burned area perimeter, were selected for this study, as samples 

of both pine areas and shrublands. Test sites outside the fire perimeter 

were called control sites (-c). Environmental conditions in these control 

sites mimic those in the two test sites inside the burned perimeter in case 

the fire had never happened. A meteorological tower was placed in the 

Forested-c area. Also, a Bowen station was set up in the Forested site in 

september 2007. For this work we have used a set of 5 Landsat 5-TM scenes 

(19 July 2007, 4 August 2007, 28 September 2007, 2 May 2008, 21 July 

2008). 

Figure 1 - Study site: (a) False color composition (7,5,3) from the L7-ETM+ image for the 8 

June 2001 (pre- fire), (b) Idem for the 26 July 2001 (post- fire), (c) Land use map before the 

fire, (d) Idem after the fire.


Effects of fire on surface energy fluxes in a Central Spain Mediterranean forest 147 

The model used is based on the Energy Balance Equation: 

R n = G + H + LE (1) 

where R n (W m -2 ) is the net radiation flux, G (W m -2 ) is the soil heat flux, 

H (W m -2 ) is the sensible heat flux, and LE is the latent heat flux or evapotranspiration. 

LE can be obtained as a residual of equation (1) if H, G, and 

R n are previously known. Detailed description of the guidelines to estimate 

every single surface energy flux at both instantaneous and daily scales, as 

well as the processing and treatment of the satellite images, can be seen 

in Sánchez et al. (2008b). 

4 - Results 

4.1 - Comparison with observed fluxes 

Merging the satellite information with values of air temperature, wind 

speed and global radiation, the data set we need is completed. Figure 2 

shows, as an example, H and LE maps generated from the 28 September 

2007 image. Three of the five scenes in the present study are concurrent 

with ground flux measurements. A good agreement is shown between predicted 

and observed fluxes. To sum up, relative errors of 6, 12, 20, and 30% 

are obtained for R n ,G, H, and LE, respectively. 

(a) (b) 

Figure 2 - Maps from L5-TM image for the 28 September 2007: (a) H (W m -2 ), (b) LE (W m -2 ).

148 


4.2 - Analysis of the fire effect 

Average values of the fluxes for each one of the test sites and dates were 

obtained. Note that the fire event occurred in 2001, and 6 years is enough 

time for the shrublands to recover its original stage (prior the fire); however, 

it is a very short time period for the forested area. After those 6 years, 

the effect of the fire on the energy flux regimen has almost disappeared in 

the shrub sites, while it is still significant in the forested sites. Figure 3 

shows the plots with the average values (for the 5 study dates) of the differences 

in terms of energy fluxes between calculated values for the forested 

and shrub test sites and their respective control sites. Sensible heat flux 

is higher in the burned area, in average 155±11 W m -2 at the time of the 

satellite overpass, and 43±14 W m -2 at daily scale, whereas latent heat flux 

is lower, 257±13 W m -2 at the time of the satellite overpass, and 62±12 W 

m -2 (2.2±0.4 mm/day) at daily scale. Therefore, even though the effect of 

the fire on the total net radiation is not very important, it is significant the 

increase in the Bowen ratio (H/LE), and the drastic decrease in the evapotranspiration 

in forested areas. 

Figure 3 - Average values of the differences between burned and control sites: (a) at instantaneous 

scale, (b) at daily scale (note that G can be neglected).


Effects of fire on surface energy fluxes in a Central Spain Mediterranean forest 149 

In this work we focus on a mediterranean forest area affected by a fire in 

the summer of 2001, located in a central Spain region. The effect of the 

fire on the energy flux regimen is analysed, using high resolution satellite 

imagery, over two different ecosystems, a pines area and a shrublands area. 

Maps of the different fluxes are created for each one of the 5 Landsat 5-TM 

images. Validation with ground measurements shows relative errors of 6, 

12, 20, and 30% for R n ,G, H, and LE, respectively. The effect of the fire in 

the shrubland test site is negligible after 6 years. However, in the forested 

test site, an increase in H over the 150 W m -2 , and a decrease in LE over 

250 W m -2 , still remain around midday. 

References 

Sánchez, J.M., Kustas, W.P., Caselles, V., Anderson, M.C., 2008a. Modelling 

surface energy fluxes over maize using a two-source patch model and 

radiometric soil and canopy temperature observations. Remote Sensing 

of Environment, 112:1130-1143. 

Sánchez, J.M., Scavone, G., Caselles, V., Valor, E., Copertino, V.A., Telesca, 

V., 2008b. Monitoring daily evapotranspiration at a regional scale from 

Landsat-TM and ETM+ data: Application to the Basilicata region. Journal 

of Hydrology, 351: 58-70. 

Randerson, J.T., Liu, H., Flanner, M.G., Chambers, S.D., Jin, Y., Hess, P.G., 

Pfister, G., Mack, M.C., Treseder, K.K., Welp, L.P., Chapin, F.S., Harden, 

J.W., Goulden, M.L., Lyons, E., Neff, J.C., Schuur, E.A.G., Zender, C.S., 

2006. The impact of Boreal Forest Fire on Climate Warming. Science, 17: 

1130-1132. 


This work was financed by the Spanish Science and Innovation Ministry 

(Project CGL2007-64666/CLI and Juan de la Cierva contract of J.M. 

Sánchez), and the JCCM (project ECOFLUX II, Ref: PCC08-0109).

DAILY MONITORING OF PRE-FIRE VEGETATION CONDITIONS USING 

SATELLITE MODIS DATA: THE EXPERIENCE OF FIRE-SAT IN 

THE BASILICATA REGION 

Abstract: This paper presents the results obtained in the context of FIRE- 

SAT project focused on the use of satellite data for pre-operational monitoring 

of fire susceptibility in the Basilicata Region. 

The use of satellite data was manyfold, to obtain: (i) fuel property (type 

and loading) map, (ii) fuel moisture estimation, (iii) fire danger/susceptibility 

indices using both fuel properties and fuel moisture. Results obtained 

during the first year project (2008) suggest that the MODIS-based model 

identified areas at higher fire susceptibility. In particular, the integration 

of the fuel type/model with daily fuel moisture and Greenness maps into a 

single, integrated model allow us to properly monitor spatial and temporal 

variations of fire susceptibility. 


A. Lanorte 1 , R. Lasaponara 1 , R. Coluzzi 1 , 

G. Basile 2 , G. Loperte 2 , F. Antonucci 2 



2 Regione Basilicata - Dip. Infrastrutture, Italy 

In the recent years, the Basilicata Region has been characterized by an 

increasing incidence of fire disturbance which also tends to affect protected 

(Regional and national parks) and natural vegetated areas. FIRE-SAT 

project has been funded by the Civil Protection of the Basilicata Region in 

order to set up a low cost methodology for fire danger/risk monitoring 

based on Earth Observation techniques. 

To this aim, NASA Moderate Resolution Imaging Spectroradiometer (MODIS) 

data were used. The spectral capability coupled with the daily availability 

makes MODIS products especially suitable for estimating fuel moisture variations 

and assessing fire danger. 

Landsat TM images were also used for mapping fuel types and loading. Fuel 

moisture, Fuel properties, and Fire danger maps obtained for the investigated 

area were compared with forest fire catalogue of 2008 year. 

151

152 


2 - Method 

The following flow chart shows the process used for estimating vegetation 

fire susceptibility assessment. Both MODIS and Landsat TM images were 

used. In particular, MODIS data were used to obtain variations in vegetation 

Greenness and moisture content, while Landsat TM data to obtain fuel 

types and loading. TM were processed using supervised classification techniques 

and spectral analysis methodologies performed (as in Lanorte and 

Lasaponara, 2007) at sub-pixel level to map: (i) Vegetation type (ii) Fuel 

type (Prometheus system), (iii) Fuel model (NFFL system), (iv) Fuel load. 

Figure 1 - Earth Observation Processing chain development.

Daily monitoring of pre-fire vegetation conditions using satellite MODIS data: the experience of FIRE-SAT in the Basilicata Region 153 

MODIS data have been used to obtain both (i) Greenness and (ii) moisture 

index. 

(i) Greenness is a quite popular fire danger index, developed by Burgan, et 

al. (1998). The basis for calculating RG is historical NDVI data that defines 

the maximum and minimum NDVI values observed for each pixel. Thus RG 

indicates how green each pixel currently is in relation to the range of historical 

NDVI observations for it. RG values are scaled from 0 to 100, with 

low values indicating the vegetation is at or near its minimum greenness. 

Specifically the algorithm is: 

RG = (ND0 - NDmn)/(NDmx - Ndmn) * 100 

where 

ND0 = highest observed NDVI value for the considered 

composite period which 8 days 

NDmn = historical minimum NDVI value for a given pixel 

NDmx = historical maximum NDVI value for a given pixel 

The purpose of using relative greenness in the fire danger estimation is to partition 

the live fuel load between the live and dead vegetation fuel classes. 

(ii) Among the wide range of vegetation indices, specifically devised to 

estimate vegetation water content, we adopted the MSI. This choice was 

driven by the results from statistical analyses that we performed on a significant 

time series. Such results pointed out the all the available satellitebased 

moisture index exhibited high correlation values. 

MSI = R 1600/ R 820 

Where R 1600 and R 820 denote the MODIS Reflectance as acquired in the 

spectral bands 1600nm and 820 nm. 

The danger classification on live fuel can be estimated by dividing the 

range of the MSI maps into different classes. Finally, The fire danger index 

(FDI), related to vegetation state, was obtained by combining the danger 

classes obtained from RG and those from MSI following an approach similar 

to that adopted in Lasaponara (2005) for combining NDVI and 

Temperature. High fire danger, as classified by FDI, was deduced by a combination 

of high dryness and low RG values, and low fire danger was 

deduced by a combination of low dryness and high RG values. 


The analysis was performed in the Basilicata (9,992 km 2 ) Region for the 

2008 year. Figures 2 show some fire danger maps obtained for the summer 

season. Currently, the fire susceptibility maps are provided daily during the 

fire season (summer season) and weekly for the rest of the year. 

Results obtained during the first year of the project (2008) shows that more 

tha 85% of fires occurred in the areas classified as high and very high dan-

154 


ger. The 15% percentage of fires which occurred in areas classified as moderate 

or low danger generally took place in forest areas and this was mainly 

due to the fact that the understory and dead fuel are masked by the 

canopy. The satisfactory results obtained for the study area suggests that 

the MODIS-based model identified the main fire danger zone. In particular, 

the integration of the fuel type/model map, with daily fuel moisture and 

Greenness maps into a single index allows us to properly capture the spatial 

and temporal variation of fire susceptibility. 

Figure 1 - Fire danger maps obtained for the 2008 summer season (on the left for 15 June 

2008 and on the right for the 15 July 2008) September). Palette indicates low to high danger 

level from green to red, white denotes cloud area or bare soil after the grain harvesting. 

References 

Lanorte, A., Lasaponara, R., 2007. Fuel type characterization based on 

coarse resolution MODIS satellite data. Forest@ 4(2), 235-243. 

Lasaponara, R., 2005. Inter-comparison of AVHRR-based fire susceptibility 

indicators for the Mediterranean ecosystems of southern Italy. 

International Journal of Remote Sensing, 26(5), 853-870.

THE GLOBAL MODIS BURNED AREA PRODUCT: VALIDATION RESULTS 

L. Boschetti 1 

1 Department of Geography, University of Maryland, USA 

luigi.boschetti@hermes.geog.umd.edu 

D.P. Roy 2 & C.O. Justice 3 

2 Geographic Information Science Center of Excellence, South Dakota State University, USA 

david.roy@sdstate.edu 

3 Department of Geography, University of Maryland, USA 

justice@hermes.geog.umd.edu 

Abstract: Earth-observing satellite systems provide the potential for an 

accurate and timely mapping of burned areas. Remote sensing algorithms 

developed to map burned areas are difficult to implement reliably over large 

areas, and globally, however, because of variations in both the surface state 

and those imposed by the remote sensing. The availability of robustly calibrated, 

atmospherically corrected, cloud-screened, geolocated global data 

provided by the Moderate Resolution Imaging Spectroradiometer (MODIS) 

allows for major advances in satellite mapping of burned area. 

This paper presents the global MODIS burned area product which is part of 

the NASA Collection 5 MODIS land product suite. The algorithm is applied 

to MODIS-Terra and MODIS-Aqua land surface reflectance time series. It has 

been implemented in the MODIS land production system as part of the standard 

MODIS land product suite to systematically map burned areas globally, 

and it is available for the 8+ year MODIS observation record. Validation 

is the term used to refer to the process of assessing satellite product accuracy 

by comparison with independent reference data. A validation protocol 

to assess the accuracy of moderate resolution burned area products has 

been developed using multi-date high resolution satellite data, and applied 

for continental validation of the product. 

1 - The MODIS Burned Area Product 

Burned areas are characterized by deposits of charcoal and ash, removal of 

vegetation, and alteration of the vegetation structure. The MODIS algorithm 

used to map burned areas takes advantage of these spectral, temporal, and 

structural changes using a change detection approach (Roy et al., 2005). It 

detects the approximate date of burning at 500 m by locating the occurrence 

of rapid changes in daily surface reflectance time series data, mapping 

the spatial extent of recent fires and not of fires that occurred in previous 

seasons or years. 

155

156 


MODIS reflectances sensed within a temporal window of a fixed number of 

days are used to predict the reflectance on a subsequent day. Rather than 

attempting to minimize the directional information present in wide fieldof-view 

satellite data by compositing, or by the use of spectral indices, this 

information is used to model the directional dependence of reflectance. 

This provides a semi-physically based method to predict change in 

reflectance from the previous state. A statistical measure is used to determine 

if the difference between the predicted and observed reflectance is a 

significant change of interest. The algorithm is repeated independently for 

each pixel, moving through the reflectance time series in daily steps. A 

temporal constraint is used to differentiate between temporary changes, 

such as shadows, that are spectrally similar to more persistent fire induced 

changes. The identification of the date of burning is constrained by the frequency 

and occurrence of missing observations and to reflect this, the algorithm 

is run to report the burn date with an 8 day precision. Further algorithm 

details are provided in Roy et al. (2005) and Roy et al. (2008), and 

the product is available to the user community (WWW1). 

2 - The burned area validation protocol 

The potential research, policy and management applications of satellite 

products place a high priority on providing statements about their accuracy 

(Morisette et al., 2006). Inter-comparison of products made with different 

satellite data and/or algorithms provide an indication of gross differences 

and possibly insights into the reasons for the differences. However 

validation with independent reference data is needed to determine accuracy 

(Justice et al., 2000). Validation is the term used here, and more generally, 

to refer to the process of assessing satellite product accuracy by comparison 

with independent reference data (Strahler et al., 2006). 

An validation protocol for the validation of moderate resolution burned area 

products (Boschetti et al., 2009) has been developed as a joint initiative 

of the Committee on Earth Observation Satellites (CEOS) Land Product 

Validation (LPV) Subgroup (WWW2) and GOFC GOLD (Global Observation of 

Forest and Land Cover Dynamics) Fire (WWW3). 

2.1 - Temporal requirements for validation datasets 

Given that burned areas are a non-permanent land cover change, it is necessary 

to define the temporal interval described by the validation reference 

data. For example, in areas where forests burn, fire affected areas may 

remain observable in satellite data for years, while in grass/shrubland systems 

burned areas may disappear within a single fire season. The length of 

time that the spectral signature of burned areas is detectable in satellite 

data after a fire depends on the physical evolution of the post-burn sur-

The global MODIS burned area product: validation results 157 

face, (vegetation re-growth, dissipation of ash and charcoal by wind and 

rain) and on the spectral bands available for the analysis (Eva and Lambin, 

1996; Trigg and Flasse, 2000). 

It is always preferable to use two TM acquisitions and then map the area 

that burned between the acquisition dates. In this way, fires that occurred 

before the first acquisition date are not mistakenly mapped as having 

burned between the two acquisition dates. Further, using two acquisitions 

provides several interpretative advantages over single date data for mapping 

burned areas. These include a reduction in the likelihood of spectral 

confusion with spectrally similar static land cover types (e.g. water bodies, 

dark soil), and the option to interpret the data by mapping relative changes 

rather than using single image classification approaches (Roy et al., 2005). 

Figure 1 - regression between proportion of area burned as detected by the MCD45 product (y 

axis) and by the Landsat reference data (x axis) over seven sites in Europe and seven sites in 

Australia, using 5 x 5 km cells. 

3 - Validation results 

The MCD45 product run globally for the first time as part of the fourth general 

reprocessing of the MODIS products suite (Collection5). The version of 

the algorithm used for collection 5 was intentionally a conservative one, 

and in several instances the comparison with Landsat data highlights the 

presence of ommission errors. A modified, less conservative version of the 

algorithm will run as Collection 5.1 during the Summer of 2009, and this 

new version will replace the existing Collection5. Figure 1 shows preliminary 

results of the validation of a prototype of collection 5.1 over Europe 

and Australia, adopting the validation protocol for the reference data and 

using the same regression over 5 x 5 km cells adopted for the validation 

over Southern Africa of Collection5 (Roy and Boschetti, 2009).

158 



Full papers presenting the results of the MCD45 validation in Europe and 

Australia are currently in preparation, together with the international collaborators 

involved in the interpretation of the Landsat images and in the 

selection of the sites: Pietro Alessandro Brivio, Daniela Stroppiana, Jan 

Kucera, Jesus San Miguel, Patricia Oliva, Andrew Edwards, Grant Allan and 

Belinda Heath. 

References 

Boschetti, L., Roy, D.P. and Justice, C.O., 2009, International Global Burned 

Area Satellite Product Validation Protocol Part I – production and standardization 

of validation reference data, available online at 

http://lpvs.gsfc.nasa.gov 

H. Eva, and E. Lambin, “Remote sensing of biomass burning in tropical 

regions: sampling issues and multisensor approach,” Remote Sens. 

Environ.t, vol. 64, pp. 292-315, 1998. 

Morisette, J.T., F. Baret, S. Liang, (2006). Special issue on Global Land 

Product Validation, IEEE TGARS, 44(7) 1695-1697. 

Roy, D.P., Jin, Y., Lewis, P. E. and Justice, C. O, (2005). Prototyping a global 

algorithm for systematic fire affected area mapping using MODIS time 

series data. Remote Sensing of Environment, 97:137-162. 

Roy, D.P., Boschetti, L., Justice, C.O. and Ju, J. 2008. The Collection 5 

MODIS Burned Area Product - Global Evaluation by Comparison with the 

MODIS Active Fire Product, Remote Sensing of Environment, 112:3690- 

3707 

Strahler, A., Boschetti, L., Foody, G., Friedl, M., Hansen, M., Harold, M., 

Mayaux, P., Morisette, J., Stehman, S., Wodcock, C., 2006. Global 

Landcover Validation: Recommendations for Evaluation and Accuracy 

Assessment of Global Landcover Maps, Luxembourg, Office for Official 

Publication of the European Communities, EUR 22156 EN, 58p. 

Roy, D.P. and Boschetti, L., 2009, Southern Africa Validation of the MODIS, 

L3JRC and GlobCarbon Burned Area Products, IEEE transactions on 

Geoscience and Remote Sensing, 47(4), 1032-1044, 

S. Trigg, and Flasse, S., “Characterizing the spectral - temporal response of 

burned savannah using in situ spectroradiometry and infrared thermometry,” 

Int. J. Remote Sens., vol. 21, pp. 3161-3168, 2000. 

WWW1: http://modis-fire.umd.edu 

WWW2: http://lpvs.gsfc.nasa.gov/ 

WWW3: http://gofc-fire.umd.edu

III FIRE DETECTION AND FIRE 

MONITORING

Abstract: This study shows the correlation between large fire emissions and 

trace gases estimated by atmospheric sensors. The zone analysed is the 

Iberian Peninsula, Spain and Portugal, during the summer season. 

Concerning atmospheric sensors, data from MOPITT (Measurements of 

Pollution in the Troposphere) onboard Terra satellite, are analysed in order 

to measure CO emissions by fires. The CO is a very important trace gas produced 

by forest fire emissions and its role in the cycle of atmospheric carbon 

is very relevant. This study is focused on two main topics: the assessment 

of dispersion of CO emissions caused by large fires and the study of 

fires series and its CO emissions. 


ANALYSIS OF CO EMISSIONS, BY FOREST FIRES, 

IN THE IBERIAN PENINSULA 

A. Calle, J-L. Casanova, J. Sanz & P. Salvador 

Remote Sensing Laboratory of University of Valladolid (LATUV), 

Valladolid, Spain 

abel@latuv.uva.es 

Carbon monoxide (CO) is a trace gas located in the atmosphere mostly as 

the result of anthropogenic activities. CO plays a significant role in the carbon 

cycle and it substantially affects the budgets of OH radicals and Ozone 

(O 3 ) that are present in the atmosphere. The anthropogenic activities linked 

to CO release into the atmosphere can be divided into two well-defined 

groups: on the one hand, urban pollutant emissions from vehicles and other 

industrial processes; on the other, from fires and global biomass burning 

emissions. The MOPITT (Measurements of Pollution in the Troposphere), 

onboard the Terra spacecraft, has proved to be the most operative sensor 

for the continuous estimation of CO [1], [2]. On the other hand, scientists 

from the National Centre for Atmospheric Research (NCAR), financed by 

NASA, have spread data and results concerning the global distribution of 

CO based on MOPITT measurements (http://www.acd.ucar.edu/) which have 

revealed both the seasonal dynamics of CO throughout the planet and direct 

correlations between the increase in the CO total column measured by 

MOPITT and large fires. Concerning the validation of MOPITT data, see [3]. 

161

162 

III - FIRE DETECTION AND FIRE MONITORING 


The Fast Fourier Transform (FFT) is a method applied to time series data 

that may present certain periodicity. Its application and analysis aims at 

the attainment of two objectives. In the first place, to describe the temporal 

behaviour of a magnitude whose data present short-term strong fluctuations 

which are probably caused by “noise”. In the second place, the 

Fourier Transform allows us to decompose the behaviour of a variable in different 

terms with different weighting factors in the reproduction of the 

series, which can be analyzed independently in order to draw conclusions 

which are explained from the hidden effects caused by the variability of the 

original data. The Fourier transform harmonics are calculated according to: 

The Factor F A (n) is the Fourier Transform for frequency n, and there are N 

frequencies corresponding to the harmonics of the series. F A (n) is calculated 

from this already known time series. If there is a discrete variable A 

which takes N values equi-spaced in time series, such variable can be represented 

(the Inverse Fourier Transform) as: 

3 - Results 

Before establishing correlations with fire data, it was necessary to find a 

yearly evolution curve of CO total column for years in the series 2005-08. 

This analysis is a synoptic approximation to the CO behaviour in the Iberian 

Peninsula. For this, a central geographical point in the Iberian Peninsula 

was selected and a radial distance of 600 km was established to determine 

CO column average values. After applying the FFT to the data, the spectral 

energy of all 1460 harmonics in the series was analyzed in order to retrieve 

the groups that would later provide the inverse Fourier transform. Thus, two 

groups were chosen for this case comprising harmonics [1-10] and [1450- 

1459] respectively. The adopted criteria, to explain this selection, were: i) 

selection of biggest spectral energy of harmonics and ii) selection of harmonics 

modulating the oscillation amplitude of values. The reference curve 

found is finally calculated through: 

where COS is the CO total column value for this synoptic curve, k represents 

the day of the series (k = 1 is 1st January 2005 and k = 1460 is 31st

Analysis of CO emissions, by forest fires, in the Iberian Peninsula 163 

December 2008), N is the total number of data, n is the ordinal of the harmonic 

used and F real (n) and F imag (n) are the two coefficients of harmonic 

determined through the FFT. Figure 1 shows in grey colour the original data 

of the series used; the continuous line superposed represents Inverse FFT. 

By means of the FFT, we have also determined the harmonics of the data 

series for each year and region. Thus, the curve which establishes the reference 

level is determined through the inverse Fourier transform and by 

using the group of harmonics that retrieve the general behaviour of data. 

Once the adjustment between both curves was carried out, we superimposed 

the inverse Fourier transform using further groups of harmonics to 

retrieve concrete behaviours. Figure 2 shows all the curves involved in the 

process described as well as the temporal situation of the fires. The figure 

corresponds to the real case of the region of Castilla y León, year 2005. The 

region’s annual curve is labelled as “yearly curve smoothed” and it has been 

reproduced through the harmonic groups: [1-3] and [360-364]; the peninsular 

synoptic curve of reference was adjusted during the spring season, 

and it appears as “adjusted synoptic curve”, eq. (4); finally, the curve that 

reproduces the local variations of the CO total column appears as “yearly 

curve reproducing data behaviour” and it has been constructed with harmonic 

groups [1-25] and [350-364]. The table 1 shows the CO exceeded 

percentage, which analytical expression is as following: 

References 

Deeter, M.N., Emmons,L.K., Francis, G.L., Edwards, D.P., Gille, J.C., Warner, 

J.X., Khattatov, B., Ziskin, D., Lamarque, J.-F., Ho, S.-P., Yudin, V., 

Attie, J.-L., Packman, D., Chen, J., Mao, D. and Drummond, J.R., 2003. 

Operational Carbon Monoxide Retrieval Algorithm and Selected Results 

for the MOPITT Instrument. Journal of Geophysical Research, 108(D14), 

4399, doi:10.1029/2002JD003186 

Deeter, M.N., 2009. MOPITT (Measurements of Pollution in the Troposphere) 

Provisional Version 4 Product User’s Guide. MOPITT Algorithm 

Development Team. Atmospheric Chemistry Division. National Center for 

Atmospheric Research. Boulder, CO 80307. Last Revised March 31, 2009 

Emmons, L.K., Deeter, M.N., Gille, J.C., Edwards, D.P., Attie, J.-L., Warner, 

J., et al., 2004. Validation of Measurements of Pollution in the 

Troposphere (MOPITT) CO retrievals with aircraft in situ profiles. Journal 

of Geophysical Research, 109(D3), D03309.

164 


Figure 1 - Seasonal evolution of the CO total column, in the Iberian Peninsula. In the background 

of figure (grey) are original data from Level processing 2 and the continuous line 

(black) is the Inverse Fourier Transform, by means of two groups of harmonics. 

Figure 2 - Representation of yearly evolution curves of CO: adjusted synoptic curve (S), yearly 

smoothed curve and yearly curve reproducing data behavior, corresponding to region (R) 

Castilla y León and year (Y) 2005.

Analysis of CO emissions, by forest fires, in the Iberian Peninsula 165 

R, Y Relationship: Interval of AR,Y Exceeded percentage, 

AR,Y vs Nf: R 2 (1018 mol·day/cm2) E(%) 

Galicia, 2005 0.84 [0.2-1.7] from 2% to 8% 

Galicia, 2006 —- 2.5 8.6% 

Castilla&León, 2005 0.75 [2.0-5.8] from 13% to 15% 

Castilla&León, 2006 —- [0.8-2.9] from 3.3% to 15.6% 

Castilla&León, 2007 0.41 [0.5-1.7] from 3% to 4% 

Table 1 - Results of AR,Y calculations and R2 values, for the most fire occurrence affected 

regions in Spain. Even when R2 was not significant, AR,Y was positive. Final column shows 

the magnitude E(%).

REAL-TIME MONITORING OF THE TRANSMISSION SYSTEM: 

WATCHING OUT FOR FIRES 

Abstract: Fires beneath the transmission lines on Eskom’s transmission system 

account for some 20% of all line faults. Transient faults on the transmission 

system are of extremely short duration measured in milliseconds, 

but their effect and cost to the utility and its customers can be considerable. 

Studies in South Africa studies show that the cost is highly variable 

with the estimated cost of damage experienced by customers varying 

between ZAR 5,000 and ZAR150, 000 per voltage dip. Eskom have now 

introduced an Active Fire Information System (AFIS) to monitor fires under 

the transmission lines in real-time and convey this information to field 

staff. The system also improves the management of fires by providing the 

opportunity to collate historical data and identify burn scar areas on an ongoing 

basis. 


Quality of supply on the Transmission system is an important matter. 

Studies show that great economic benefits may be reaped where voltage 

dips on Transmission lines are curtailed. Fires under Transmission lines of 

the Eskom Transmission system accounts for about 20% of all line faults. 

The management of this source of line faults requires of the asset manager 

to not only monitor fires in real-time and convey this information to field 

staff, but also to take cognisance of trends and map the history of fires and 

burn scar areas on an ongoing basis. This paper discusses the Advanced Fire 

Information System (AFIS) utilised by Eskom in the management of fires 

under Transmission lines. 

2 - Eskom and the space age 

P. Frost 

CSIR, Meraka Institute RSRU, Pretoria, South Africa 

pfrost@csir.co.za 

H. Vosloo, A. Momberg & I.T. Josephine 

Eskom, Johannesburg, South Africa 

hfvosloo@eskom.co.za 

The launching of the Aqua and Terra satellites with the Moderate Resolution 

167

168 


Imaging Spectroradiometer (MODIS) by NASA in 1999 and 2002 provided 

the world with a tool to be used inter alia in fire tracking. 

In early 2004 Eskom and the Council of Scientific and Industrial Research 

(CSIR), launched a research project to demonstrate the ability to track 

active fires by using polar orbiting satellites. MODIS being medium-resolution 

scanner enables four updates daily with a 1km 2 resolution [1]. The 

detection of grass fires as small as 0.25ha is possible with this data [2]. 

Eskom required information on fires every 15 minutes and whilst the MODIS 

data was sufficiently high in spatial resolution, the temporal resolution was 

less than satisfactory. Consequently the CSIR proposed the use of the 

Spinning Enhanced Visible and Infrared Imager (SEVIRI) sensor on board 

the Meteosat Second Generation (MSG) satellite. This satellite is in geostationary 

orbit above the equator and the SEVIRI sensor transmits data every 

15 minutes. MSG data is observed at a spatial resolution of 5km 2 [3]. Once 

the processing of the hot spots have been completed the information is 

published on a website, and e-mail alerts are sent. Due to the fact that the 

Eskom field personnel are normally out on patrols, the idea emerged to send 

text message warning of fires to their mobile phones. In the case of power 

lines, only the fires within 2,5km of the line are reported. Where possible, 

the National Control Centre could temporarily switch out the lines under 

threat and the field staff can activate fire suppression teams where available. 

Field staff also report to the control centre on the conditions at the 

site of the fire. This system was the first of its kind in the world where an 

electrical utility applied remote sensing together with cell phone technology 

in the monitoring of fires under power lines. 

3 - Vegetation management in 2009 

In addition to the real-time use of the system, the use of historic data 

accumulated by the system is proving to be most helpful in the management 

of the Transmission system. History on fires close to transmission 

lines is now available since 2003 and this information is used to track past 

performance as well as planning vegetation management strategies for the 

future. In Table 1 the number of fires detected close to power lines gives 

an indication of the extent of the fire season and is used to evaluate fire 

prevention and reduction strategies. The planning of vegetation management 

has three main objectives. They are: 

• Ensure safe electrical clearances; 

• Ensure access to the line for inspection and maintenance; 

• Reduction of fuel loads to reduce the effects of fires.

Real-time monitoring of the transmission system: watching out for fires 169 

Figure 1 - An example of three text message alerts of a fire received from the Aqua satellite. 

The first one is of a fire 2.07km west of tower 45 on the No2 Ankerlig - Aurora line. The observation 

time was 14:45 South African Standard time. 

FIRES OCCURING WITHIN 2.5KM OF ESKOM’S POWER LINE 

YEAR No. of fires 

2003 3987 

2004 3905 

2005 6147 

2006 5013 

2007 5135 

2008 4998 

Table 1 - The number of fires occurring within the buffer from 2003-2008. 

4 - Conclusion 

The AFIS project has shown great potential and has demonstrated the ability 

of technologies such as satellite remote sensing to aid industries in the 

public and private sector. The combination of research, public and private 

organisational partnership have shown to work very well, and will provide 

a strong link for further work in this domain.

170 


The statistics have shown that AFIS can provide information that could be 

used in the reduction of fire line faults by early identification of fires close 

to transmission lines. The ability to stop even 1 flashover from occurring 

could save Eskom hundreds of thousands or rands. It is extremely difficult 

to put a price tag on a line fault. In some cases there might be very little 

impact of a line fault but cases have been reported where companies lost 

huge amounts of money due to the malfunctioning of machinery as a result 

of a fire flashover. 

References 

Anon, 2009. http://modis.gsfc.nasa.gov/about/specifications.php, Accessed, 

6 April 2009. 

Frost, P. and Vosloo, H.F., 2006. Providing Satellite-Based Early Warnings of 

Fires to Reduce Fire Flashovers on South Africa’s Transmission Lines. 

Proceedings of the Bushfire conference 2006 - Brisbane 6-9 June 2006. 

Anon, 2009. _http://www.eumetsat.int/HOME/Main/What_We_Do/Satellites/Meteosat_Second_Generation/ 

index.htm, Accessed, 6 April 2009.

NOAA’S OPERATIONAL FIRE AND SMOKE DETECTION PROGRAM 

M. Ruminski 1 , P. Davidson 2 , R. Draxler 2 , S. Kondragunta 2 , 

J. Simko 2 , J. Zeng 3 , P. Li 4 

1 Satellite Analysis Branch, NOAA/NESDIS, Camp Springs, USA, mark.ruminski@noaa.gov 

2 NOAA, Silver Spring, USA, paula.davidson@noaa.gov; roland.draxler@noaa.gov; 

shobha.kondragunta@noaa.gov; john.simko@noaa.gov 

3 Earth Resources Technology, Camp Springs, USA, Jian.zeng@noaa.gov 

4 Perot Systems, Camp Springs, USA, po.li@noaa.gov 

Abstract: Biomass fire occurs in nearly every ecosystem and geographic 

region of the world. It can be managed, as with prescribed or agricultural 

burns, or rage out of control in the case of wildfires. Accompanying the 

fires are emissions which can be limited and localized or which can literally 

span the globe and impact regional atmospheric conditions. Knowledge 

of the number, location, duration and size of the fires and estimates of 

their emissions and subsequent trajectories has become increasingly important 

for public health, property loss, transportation, etc. This paper briefly 

describes NOAA’s operational fire and smoke analysis product and smoke 

forecasting system. 

1 - Fire Analysis Methodology 

NOAA utilizes 7 satellites with multispectral imagery for optimal operational 

fire detection. The WildFire Automated Biomass Burning Algorithm 

(WFABBA) is employed at 30 minute intervals using GOES-East and GOES- 

West imagery. Details on the algorithm can be found in Prins and Menzel, 

1992. NOAA polar orbiting data from NOAA-15/17/18 is currently used in 

the Fire Identification, Mapping and Monitoring Algorithm (FIMMA) based 

on the scheme described in Li et al. (2000) and subsequently developed 

and updated at NESDIS. The algorithm for MODIS Terra/Aqua imagery is 

described in Justice et al. (2002) and Giglio et al. (2003). All of the algorithms 

generate fire detection locations which are synthesized into an 

interactive visualization system - the Hazard Mapping System (HMS) 

(Ruminski et al., 2008) - which integrates algorithm output with all of the 

underlying satellite imagery as well as numerous ancillary data layers. Data 

layers include previously identified locations of persistent thermal anomalies, 

power plant locations, land types, stable light regions, etc. which aid 

the analyst in identifying possible false detects. Analysts quality control 

the automated fire detections by deleting those detections that are deemed 

171

172 


to be false detects and adding fires that the algorithms have not detected. 

False detects can be due to a number of conditions including urban heat 

islands, solar specular reflection off water surfaces, highly reflective clouds, 

relatively brighter vegetated land embedded within darker areas, instrument 

noise, etc. Fires not detected by the algorithms can be attributed to 

partial obscuration by clouds, overhead vegetation canopy, detected temperature 

that is not sufficiently hot or hotter than the surrounding area, 

etc. Validation results (Schroeder et al., 2008) using 30m ASTER data found 

that fires added by analysts reduced the omission error rate compared to 

WFABBA and MODIS detections. However, the commission error rate was not 

reduced by automated detections that were deleted, with actual fires being 

erroneously removed. Validation using ground based reports from Florida, 

Montana, Idaho and Manitoba supported the finding of reduced omission 

errors in the HMS compared to the automated only product (WFABBA, 

MODIS and FIMMA) with approximately twice as many detections, although 

the overall detection rate was only 25-30%. The low detection rate is 

ascribed to the small size (less than 2 ha) of many of the fires as well as 

prohibitive cloud cover. HMS products may be accessed at 

www.osdpd.noaa.gov/ml/land/hms.html. 

2 - Smoke Detection and Model Transport 

Smoke detection is exclusively performed with visible imagery, primarily 

utilizing animated GOES data, although polar imagery is occasionally used. 

Detection is optimized over the contiguous US with GOES-West in the morning 

and GOES-East in the evening due to favorable solar zenith and satellite 

viewing angles. Analysts graphically depict the smoke extent by manually 

drawing polygons. An estimate of the vertically integrated smoke concentration 

is assigned to each polygon. Three ranges of values (in µm/m 3 ) 

are available for the analyst to assign to the polygon. The automated GOES 

Aerosol and Smoke Product (GASP) (Knapp et al., 2005) which generates 

Aerosol Optical Depth (AOD) is a tool to aid analysts in this determination. 

The smoke depicted may be attached to actively burning fires or may be 

several days old and have drifted hundreds or thousands of km from the 

source. 

Many fires do not produce emissions that are detectable. Clouds obscure 

some smoke emissions while other fires may have minimal emissions due to 

the short duration of the fire, limited biomass or the engineering of the fire 

in the case of agricultural/prescibe burns. For fires producing smoke that is 

detected the analyst provides an estimate of the initiation and duration of 

emissions and a coarse estimate of the fire size. The size is ideally estimated 

using higher (1km) resolution polar data with the fire close to the 

suborbital track. This is not always possible, especially in dynamic wildfire 

situations in the mid and lower latitudes. In these cases the analyst pro-

Noaa’s operational fire and smoke detection program 173 

vides the input information based on visual observation of the amount of 

smoke being produced and experience. 

This information is used as input for NOAA’s operational air quality forecast 

capability (Rolph et al., 2009) viewed at www.nws.noaa.gov/aq/ Emission 

information from the HMS is used by the HYSPLIT model (Draxler and Hess, 

1998) in conjunction with the North American Mesoscale (NAM) meteorological 

model to generate smoke dispersion and trajectories for the ensuing 

48 hours for the contiguous US. Smoke emission rates are obtained from the 

US Forest Service BlueSky (www.airfire.org/bluesky) emission algorithm, 

which includes a fuel type database and consumption and emissions models. 

An automated satellite-based smoke detection and tracking technique has 

recently been developed for verifying the smoke predictions. A source 

apportionment technique matches identified fires with maps of AOD from 

GASP and tracks identified smoke at a 30 minute interval. Initial validation 

results comparing the smoke optical depth from this product to the Ozone 

Mapping Instrument (OMI) total optical depth for absorbing aerosol indicate 

agreement within 0.2 AOD. 

References 

Draxler, R.R. and Hess, G.D., 1998. An Overview of the HYSPLIT_4 Modelling 

System for Trajectories, Dispersion, and Deposition. Australian 

Meteorological Magazine, 47, 295-308. 

Giglio, L., Descloitres, J., Justice, C.O., Kaufman Y.J., 2003. An Enhanced 

Contextual Fire Detection Algorithm for MODIS. Remote Sens. Environ, 

87, 273-282. 

Justice, C.O., Giglio, L., Korontzi, S., Owens, J., Morisette, J., Roy, D., 

Descloitres, D.J., Alleaume, S., Petitcolin, F., Kaufman, Y., 2002. The 

MODIS Fire Products. Remote Sens. Environ, 83, 244-262. 

Knapp, K.E., Frouin, R., Kondragunta, S., Prados, A., I., 2005. Towards 

Aerosol Optical Depth Retrievals Over Land from GOES Visible Radiances: 

Determining Surface Reflectance, Int. J. Rem. Sens., 26, 4097-4116. 

Li, Z., Nadon, S., Cihlar., J., 2000. Satellite Detection of Canadian Boreal 

Forest Fires: Development and Application of an Algorithm, Int. J. Rem. 

Sens., 21, 3057-3069. 

Prins, E.M., Menzel, W.P., 1992. Geostationary Satellite Detection of Biomass 

Burning in South America, Int. J. Remote Sens., 13, 2783-2799. 

Rolph, G.D., Draxler, R.R., Stein, A.F., Taylor, A, Ruminski, M.G., 

Kondragunta, S., Zeng, J., Huang, H., Manikin, G., McQueen, J.T., 

Davidson, P.M., 2009. Description and Verification of the NOAA Smoke 

Forecasting System: The 2007 Fire Season, Weather and Forecasting, 24, 

361-378.

174 


Ruminski, M. and Hanna J., 2008. Validation of Remotely Sensed Fire 

Detections Using Ground and Aircraft Reports, American Geophysical 

Union Fall Conference, San Francisco, California. 

Ruminski, M., Simko, J., Kibler, J., Kondragunta, S., Draxler, R., Davidson, 

P., Li, P., 2008. Use of Multiple Satellite Sensors in NOAA’s Operational 

Near Real-time Fire and Smoke Detection and Characterization Program, 

SPIE Optics and Photonics Conference, San Diego, California. 

Schroeder, W., Ruminski, M., Csiszar, I., Giglio, L., Prins, E., Schmidt, C., 

Morisette, J., 2008. Validation Analysis of an Operational Fire Monitoring 

Product: The Hazard Mapping System, Int. J. Rem. Sens., 29, 6059-6066. 

Graphic of HMS analysis with 

fires in red and smoke in gray. 

Forecast of smoke concentration.

SYSTEM FOR EARLY FOREST FIRE DETECTION: FIREWATCH 

Abstract: The project involves construction and operation of a fire detection 

system, called FIREWATCH, which could exceed the limits of current 

systems for monitoring large areas. It has described how the system works 

and the data obtained during its operation in 2005 in a region of Germany 

are reported. 

1 - Firewatch system 

T. Berna, F. Manassero 

ETG Risorse e Tecnologia, 

Montiglio Monferrato (AT), Italy 

infoetg@etgrisorse.com 

In this section we provide a description of the characteristics of the system 

without entering in the details of engineering. 

Firewatch system is schematically illustrated in figure 1. 

175

176 


1.1 - Features and operation 

The system is based on digital cameras in black and white, with CCD sensor. 

The use of black and white permits, compared with a colour camera, to 

have, for the same sensor, a resolution 4 times higher compared with a 

colour camera. The image picked by the camera is processed in real time in 

order to detect the presence of the fire and only in this case is sent with 

reporting to the operational center. This is crucial because the operator is 

not alerted continuously, but should only validate the alarm, alerting, as 

various protocols, the various fire brigades. This mode allows you to intervene 

timely, since the detection of smoking occurs even when there was 

still development of flame. In this way, with a simple and efficient management, 

it is possible to intervene when the fire is still small. The time 

that passes from smoke generation and sending the alarm to central is 

between 4 and 8 minutes. Another important feature is that Firewatch can 

work even at night, thanks to optical part that light the picture; the performances 

are comparable with those of day: the ability of detection is at 

a distance of 10 km, a cloud of smoke 10 x 10 m. Firewatch can precisely 

detect the smoke due to fire wood, as it is characterized by a combustion 

process that emits smoke with certain characteristics, allowing its detection 

for the particular brightness and contrast. Finally, the scope of the system: 

a single camera works very well with a radius of 15 km, that can cover 

an area of approximately 70,000 ha. 

2 - Forest fires in Italy 

The existence of a fire problem in Italy is confirmed by data of 2007, the 

year in which 226’000 hectares had burned, of which forest 116’000. In our 

country, which has a limited extension and where the percentage considered 

forest, in accordance with current standards, is 34.7% the fire becomes 

a source of damages, with incalculable consequences. 

Currently, the trend of growth of the forest in Italy is positive. It thus 

becomes important to careful management of forests in particularly in its 

defence against fire, not only in forest terms, but the general commitment 

to the environment. In 2007, to August 31 have been released into the 

atmosphere 7 million and a half of tonnes of CO2 due to forest fires, which 

is the amount of CO2 that would have been produced by the consumption 

of 14 GWh, considering the Italian energy mix. 2007 was a year particularly 

burdensome, with area burned in a proportion higher than the number of 

fires in other years because the surface was that of 1993, while the number 

of fires is lower.

3 - Result in Germany 

System for early forest fire detection: firewatch 177 

In Germany, in the area of Brandenburg, Firewatch system was adopted and 

evaluated by the German Ministry for Forests. In these areas there was 

already a structure based on observation towers with operator warning, 

which was gradually replaced by the system Firewatch. 

Data were collected in 2005, when all facilities were completed and operational, 

and compared with the averages for the period 1975 to 1996 and 

relate to a wooded area of 1,100,000 ha. In Germany the central office is 

managed by the Forest Guard, which constantly monitor the images with its 

alarms reported by the system, just one of the records is validated by the 

operators, firemen are alerted, sending the image detected by camera, the 

coordinates and all the useful data. 

Figure 2 - Comparison 1975-1996 and 2005. 

Figure 3 - Comparison 1975-1996 and 2005.

178 


The data in Figure 2 show that 2005 was characterized by a number of fires 

smaller than the average for the period 1975-1996, but Figure 3 reveals 

how the damaged area has suffered a decrease in greater proportion than 

the number of fires. The latter, however, is comparable to the Italian statistics, 

as well as between number of fires on forest land to which the data 

is a factor of 10 if we have approximately 10,000,000 ha in Italy, data from 

the German Ministry is cover approximately 1,000,000 hectares and the 

number of fires varies in proportion, then the German and Italian statistics 

are comparable. The number of fires is not a controllable variable, which 

depends on a multitude of factors; Firewatch is a powerful tool for what 

concerns the reduction of area damaged, as it allows prompt intervention 

of a few minutes after the beginning of the occurrence a possible development 

of fire. 


The results of Figure 3 were obtained with 130 installations over an area of 

1,100,000 ha in the region of Brandenburg. This territory is characterized, 

in addition to the interface areas on the outskirts of town, by a multitude 

of small houses in the surrounding woods, whose presence necessarily creates 

danger. 

The variety and complexity of the Italian territory, where there are interface 

areas and 22.2% of the area in a forest falls in SIC and ZPS, makes it 

all more relevant the importance of a systematic approach to the problem. 

Emblematic is the case of fires in Quiliano and Alassio in Liguria, which 

have grown in 01/01/2007 and where the lack of early warning from a man 

and a particularly critical of wind, caused that after three hours from the 

beginning of a fire the smoke reached Capo Corso and Elba. 

If we optimistically in the same reduction that has been in the data of 

Figure 3, in 2007 we had a tenth of hectares burned, which is approximately 

11,600 ha, thus below the minimum annual values recorded from 1990 to 

2007 in Italy. 

Actually in Italy the system is under assessment in the National Park of 

Abruzzo, Lazio and Molise (see photo) and it will be used also in California 

(see photo: CEBIT 2009).

References 

System for early forest fire detection: firewatch 179 

Berna, T., Manassero, F., 2008. Sistema di rilevamento incendi boschivi. 

Congresso Nazionale di Selvicoltura, Taormina. 

Das Lebensministerium, 2005. Waldzustandsbericht 2005. Freistaat Sachsen, 

Staatministerium fur Umwelt und Landwirtschaft. 

Domenichini, P., et al., 2005. Manuale per l’operatore antincendio boschivo. 

Regione Liguria, Dipartimento Agricoltura e Protezione civile, Genova. 

Land Brandenburg, 2007. Titelthema: Waldbrand 2007. Ministerium fur 

Landliche Entwicklung, Umwelt und Verbraucherschutz. 

Milazzo, A., 2008. Gli incendi boschivi 2008, Corpo Forestale dello Stato, 

2008.

EARLY WARNING SYSTEM FOR FIRES IN MEXICO AND CENTRAL AMERICA 

G. López Saldaña 

Remote Sensing Department, CONABIO, Mexico City, Ave. Liga Periférico - Insurgentes Sur 4903, 

Tlalpan 14010, Mexico D.F., Mexico 

Gerardo.Lopez@conabio.gob.mx 

I. Cruz López, R. Ressl 

Conabio, Mexico City, Mexico 

icruz@conabio.gob.mx; rressl@conabio.gob.mx 

Abstract: Forest fires have severe repercussions in Mexico: only in 2008, 

more than 200,000 ha were affected, and 25,000 ha were woody vegetation. 

The local and global impact this represents motivated Conabio to conduct 

a study called “Fires in Mexico - an analysis of their threat to biodiversity,” 

and to implement in 1999 an operational program for wildfire 

detection using satellite images. 

Information about environmental variables before and after fire occurrence 

is necessary to support decision-makers; therefore, data about vegetation 

conditions based on NDVI anomaly is provided to show areas where fire is 

more likely to spread. In addition, a fire risk index based on vegetation and 

soil moisture is now developed as a pilot project, and the post-fire analysis 

is carried out mapping burnt areas using a daily-basis identification 

approach. 

The Early Warning System for Fires has been providing active fire locations 

since 1999 and will provide valuable information in near-real time for all 

stages about fire events for Mexico and Central America. 


Between 1998 and June 2009, the annual average area affected by fires in 

Mexico was approximately 210,295 ha, with a maximum in 1998 of almost 

495,000 ha. The average affected area per fire was 24.83 ha for the same 

period (CONAFOR, 2009). 

Traditionally, fire protection and early fire alerts in Mexico are performed in 

a conventional manner by tower observations, which are by far insufficient 

to cover the entire 22.75 million ha currently under protection. In the last 

years, airborne and satellite observations have also been used, the latter 

mainly provided by the Direct Readout (DR) station of the National 

Commission for the Knowledge and Use of the Biodiversity (CONABIO). The 

aim of this paper is to show the variety of products generated in near-real 

181

182 


time with the Early Warning System for Fires operated by CONABIO for 

Mexico and Central America. 

2 - Data processing and product generation 

The operational fire monitoring system at CONABIO has continuously been 

developed and enhanced since 1999. The main component is the near-real 

time detection and characterization of fires, including fast information supply 

to the user by a processing chain. CONABIO receives up to eight MODIS 

satellite overpasses and up to ten satellite passes of NOAA-AVHRR on a 

daily basis. All passes are processed and provided as Level-1B data in nearreal 

time to the scientific community and any interested user. 

2.1 - Early-warning products 

Information about environmental variables, especially before fire occurrence, 

is crucial to support fire-related decisions. 

In order to evaluate if current vegetation conditions represent a threat, it 

is necessary to include several variables related to fire ignition, fire propagation, 

fire vulnerability (Chuvieco et al., 2003b). Fuel moisture content 

(FMC), defined as the proportion of water over dry mass, is the most 

extended measurement of fire propagation potential (Trowbridge and Feller, 

1988; Viegas et al., 1992); since FMC has a strong correlation with the 

Normalized Difference Vegetation Index (NDVI) for grasslands and shrublands 

(Yebra, 2008), NDVI was used to derive a Fire Propagation Index (FPI) 

based exclusively on vegetation conditions. 

FPI is calculated for a 10-day period with a normalized difference of the 

present NDVI and a synthetic NDVI created by a harmonic analysis of a 4year 

time series (de Badts et al., 2005); therefore, if the present vegetation 

condition is below historical conditions, it can be assumed that the vegetation 

is under stress and the possibility of fire propagation is higher. The 

result is a map showing the areas where a wildfire could propagate once it 

has been started. 

Since vegetation conditions are strongly correlated to meteorological phenomena, 

it was necessary to include some variables related, for instance, to 

precipitation; and since vegetation takes humidity from the surrounding 

environment and loses this humidity due to temperature, these three variables 

were taken into account from different sources to model dead litter 

moisture content: 1) duration of precipitation from TRMM (Tropical Rainfall 

Measuring Mission), and from MODIS standard products: 2) land surface 

temperature (LST) from MOD11 and 3) relative humidity from MOD07. 

The model used equations from the US Forest Service risk model to estimate 

moisture flux between dead litter over forested areas.

2.2 - Active fire identification 

Early warning system for fires in Mexico and Central America 183 

Immediately after each satellite overpass, Level-0 data is preprocessed to 

Level-1B, including geolocation, calibration, and bowtie correction. In the 

case of MODIS, a suite of standard products is derived automatically including 

MOD14. The extracted fire information is projected to a Lambert 

Conformal Conic projection and exported to a generic binary format for further 

processing. Then, the fires are attributed using additional GIS information 

such as geographic coordinates, states and districts, type of vegetation, 

slope, and information on the proximity to NPAs. Contiguous fires 

are grouped and defined as a fire complex. Besides tabular information, the 

following suite of visual products is created: 1) a quick look to provide rapid 

clarification for satellite coverage and cloud cover; 2) information about 

cloud cover and invalid data percentages; 3) Enhanced Vegetation Index 

(EVI) with superimposed fires; 4) 10-day FPI composite; 5) shaded relief of 

a DEM (200 m) with superimposed fires. 

2.3 - Burnt area identification 

Many burnt area products use a set of thresholds together with a time series 

analysis to detect rapid and dramatic land cover changes. 

For instance, the on-going Global Burnt Areas 2000-2007 (L3JRC; Tansey et 

al., 2007) or the MODIS burnt area product (MCD45A1) (Roy et al., 2002, 

2005). All products are available on a monthly or annual basis. In contrast, 

this study uses two spectral indices for daily assessment, which have been 

deemed useful for burnt area detection (Walz et al., 2007). For this study, 

daily surface reflectance data of the MODIS-Aqua instrument with 500 m 

spatial resolution (MYD09GA) have been employed. In order to utilize only 

high quality data pixels, the quality assurance data set (QA) provided with 

the MODIS granules has been analyzed. Only pixels which fulfilled the 

strong specifications of quality and with a sensor zenith angle lower than 

45° were preserved. 

Two indices have been used for burnt area detection in this study. The 

NDVI; Eq. (1) indicates vegetation with higher photosynthetic activity and 

has been used for burnt area mapping in boreal forests and the Normalized 

Burn Ratio (NBR; Eq. (2)), indicates burnt areas by low values (Walz et al., 

2007). While the reflectance in the near infrared and therefore also in the 

NDVI decreases due to the loss of biomass, the reflectance in the shortwave 

infrared increases because of the change in water content and soil 

exposure. 

NDVI = ρNIR – ρRED / ρNIR + ρRED eq (1) 

NBR = ρNIR – ρSWIR / ρNIR + ρSWIR eq (2)

184 


This methodology has been assessed in other studies (Ressl et al., 2009) 

where showing the MODIS burnt area identification has yielded a conservative 

estimate, i.e. pixels identified as burnt area have likely been burnt in 

reality. The quantitative comparison between ASTER and MODIS burnt area 

considering the proportional spatial coverage has clearly indicated the 

scale issues of burnt area mapping. However, if more than 80% of the 

MODIS pixel is covered by burnt area derived from ASTER, the pixel is likely 

to be detected by the MODIS burnt area algorithm presented above. In 

contrast to most other burnt area detection techniques, this simpler algorithm 

solely operates on daily datasets without contextual or time series 

analysis. Therefore, the approach for burnt area mapping presented in this 

study has the clear advantage of a rapid assessment. Further information 

about the system is available at: (http://www.conabio.gob.mx/conocimiento/puntos_calor/doctos/ 

puntos_calor.html). 

References 

Chuvieco, E., Allgöwer, B., Salas, F.J., 2003. Integration of physical and 

human factors in fire danger assessment. In: Chuvieco, E. (Ed.), Wildland 

Fire Danger Estimation and Mapping: The Role of Remote Sensing Data. 

World Scientific Publishing, Singapore, pp. 197-218. 

CONAFOR, 2009. Reporte semanal de incendios forestales 2009 - Datos acumulados 

del 01 de Enero al 04 de Junio de 2009. http://www. conafor. 

gob.mx/portal/docs/subsecciones/incendios_forestales/reporte_semanal.pdf 

de Badts, E., López, G., Wickel, B., Cruz, I., and Jiménez, R., 2005. A fire 

risk propagation map based on NDVI anomalies. In de la Riva, J., Pérez- 

Cabello, F. & Chuvieco, E., (Eds.), Proceedings of the 5th Internacional 

workshop on remote sensing and GIS applications to forest fire management: 

Fire effects assessment, Universidad de Zaragoza (pp. 113- 

118). 

Ressl R., López G., Cruz I., Colditz R.R., Schmidt M., Ressl S., Jimenez R. 

Operational active fire mapping and burnt area identification applicable 

to Mexican Nature Protection Areas using MODIS and NOAA-AVHRR direct 

readout data. Remote Sensing of Environment 113 (2009), pp. 1113- 

1126 

Trowbridge, R., Feller, M.C., 1988. Relationships between the moisture content 

of fine woody fuels in lodgepole pine slash and the fine fuel moisture 

code of the canadian forest fire weather index system. Can. J. Forest 

Res. 18, 128-131. 

Viegas, D.X., Viegas, T.P., Ferreira, A.D., 1992. Moisture content of fine forest 

fuels and fire occurrence in central Portugal. Int. J. Wildland Fire 2 

(2), 69-85. 

Yebra M., Chuvieco E., Riaño D., 2008. Estimation of live fuel moisture content 

from MODIS images for fire risk assessment. Agricultural and forest

Early warning system for fires in Mexico and Central America 185 

meteorology 148 (2008), pp. 523-536. 

Walz, Y., Maier, S.W., Dech, S.W., Conrad, C., and Colditz, R.R., 2007. 

Classification of burn severity using Moderate Resolution Imaging 

Spectroradiometer (MODIS): A case study in the jarrah-marri forest of 

southwest Western Australia. Journal of Geophysical Research, 112, 

G02002. doi:10.1029/2005JG000118.

ADVANCING THE USE OF MULTI-RESOLUTION REMOTE SENSING DATA 

TO DETECT AND CHARACTERIZE BIOMASS BURNING 

W. Schroeder 

University of Maryland, Earth System Science Interdisciplinary Center, 

College Park MD, USA, wilfrid.schroeder@noaa.gov 

I. Csiszar 1 , L. Giglio 2 & C. Justice 3 

1 NOAA/NESDIS/STAR, Camp Springs MD, USA, ivan.csiszar@noaa.gov 

2 SSAI, Lanham MD, USA, louis_giglio@ssaihq.com 

3 University of Maryland, College Park MD, USA, justice@hermes.geog.umd.edu 

Abstract: In this study, we sought to detect and characterize biomass burning 

in Amazonia using 1-km MODIS Thermal Anomalies and the 4-km GOES 

imager WF-ABBA data. We applied field measurements, airborne data, and 

higher spatial resolution remote sensing data from CBERS, ASTER, and 

Landsat (TM and ETM+) to assess sub-pixel processes governing MODIS and 

GOES fire product performance. The results from the analyses above were 

used to create a set of judicious criteria aimed to integrate the active fire 

data from MODIS and GOES imager for Amazonia. The resulting product differs 

from the simple sum of the individual input fire products as it compensates 

for potential sources of commission and omission errors in the 

data. 


Satellite fire data provide key information for the scientific community as 

well as for fire managers. In this study we describe how multiple fire data 

sets are being combined to detect and characterize vegetation fires with a 

focus in Amazonia. Using ground, airborne and higher resolution spaceborne 

data we investigate sub-pixel fires routinely mapped by coarser spatial 

resolution instruments that currently serve regional and global biomass 

burning applications. Through a thorough product assessment we complement 

the use of a polar orbiting fire data set with a geostationary one creating 

a regional map of fire activity for Amazonia in which commission and 

omission errors are minimized. 

2 - Data and methods 

Several data sets were utilized in this study including: (i) field measurements 

(fire temperature and duration derived from prescribed burns along 

187

188 


a gradient of fire types); (ii) Airborne Hyperspectral Sensor (AHS) data 

(opportunistic imaging of several vegetation fires at ~1.5m resolution); 

(iii) China-Brazil Earth Resources Satellite (CBERS) (burned area mapping 

at 20m resolution with 26 day repeat cycle); (iv) Advanced Spaceborne 

Thermal Emission and Reflection Radiometer and Landsat Thematic Mapper 

(TM and ETM+) (active fire detection and burned area mapping at 30m resolution 

with 16 day repeat cycle); (v) Moderate Resolution Imaging 

Spectroradiometer (MODIS) onboard Terra and Aqua satellites (1km active 

fire detection and characterization from the Thermal Anomalies (TA) product 

– 12hour interval (can be reduced depending on use); and (vi) 

Geostationary Operational Environmental Satellite (GOES) imager (4km 

active fire detection and characterization from the WildFire Automated 

Biomass Burning Algorithm (WF-ABBA) – 30min interval). 

2.1 - Fire Product Assessment 

Our assessment of the TA and WF-ABBA fire products built on the study of 

Morisette et al. (2005). We used coincident and near coincident higher spatial 

resolution data to map sub-pixel fires within the MODIS and GOES imager 

footprints (see Giglio et al., 2008; Csiszar et al., 2006; Schroeder et al., 

2008c). The effects of short-term variations in fire behavior were analyzed 

using Landsat ETM+ and ASTER data acquired 30min apart (see Csiszar and 

Schroeder, 2008). We used the Vegetation Continuous Fields (VCF) product 

to stratify our results in terms of the percentage tree cover found in and 

around the fire pixel area. 

2.2 - Consideration of Cloud Obscuration Omission Errors 

Opaque clouds can greatly reduce the ability to detect fires using spaceborne 

instruments due to severe attenuation of the spectral signal emitted 

by either flaming or smoldering phases of biomass combustion. In order 

account for the resulting omission errors, we applied a probabilistic 

approach to the WF-ABBA product over Brazilian Amazonia using precipitation 

estimates, a cloud mask, and land use data, all derived from the GOES 

imager data (for details see Schroeder et al., 2008b). 

2.3 - Fire Characterization 

Fire characterization is required in order to understand the processes leading 

to and resulting from biomass burning. Currently, fire size and temperature 

estimates are calculated by the WF-ABBA product, whereas both WF- 

ABBA and TA provide estimates of Fire Radiative Power. In order to assess 

those parameters we used coincident data from ASTER and Landsat ETM+.

Advancing the use of multi-resolution remote sensing data to detect and characterize biomass burning 189 

Fire masks derived from those two instruments were used along with airborne 

data and field measurements to either directly or indirectly evaluate 

the MODIS and GOES imager fire characterization data on a pixel basis 

(Schroeder et al., submitted). 

3 - Results 

Estimates of clear sky omission errors were generated for WF-ABBA and the 

TA product based on active fire pixel summary statistics derived from 30m 

ASTER and Landsat ETM+ data. Fire unrelated commission errors were estimated 

to be ~2% for the two products above. Commission associated with 

recently burned pixels varied as a function of percentage tree cover, being 

larger (~15-35%) for densely forested areas. 

WF-ABBA omission errors due to cloud obscuration in Amazonia were equivalent 

to 11%. Reduced cloud coverage during periods of peak fire activity 

(i.e., early afternoon hours during the dry season months) minimized the 

omission errors. Assessment of the technique indicated a 5% agreement 

between predicted x observed omission error values based on monthly statistics. 

Figure 2 - Annual (2005) fire pixel density maps for the original TA/Terra (A), TA/Aqua(B), and 

WF-ABBA (C), and the integrated product (D) for Brazilian Amazonia. The scales represent the 

average number of days with detections calculated for individual 40km cells.

190 


Among the fire characterization parameters provided by WF-ABBA and TA, 

the FRP estimates showed better sensitivity to varying fire conditions. The 

interplay among sub-pixel fires, background characterization, and the sensor 

characteristics (in particular the point spread function), had an important 

effect on the accuracy of the estimates produced. Pixel based estimates 

derived from different products remain poorly correlated and subjected 

to random errors. 

Using the results above, we designed a set of judicious criteria to integrate 

the active fire data from MODIS Terra and Aqua and the GOES imager (see 

Schroeder, 2008a) (Figure 2). The resulting product incorporates corrections 

aimed to minimize errors of commission and omission to maximize the 

complementarity between products. 


In this study we discuss briefly how multiple data sets of rather different 

characteristics are used to advance our understanding of sub-pixel processes 

governing vegetation fire activity routinely mapped by two major hemispheric-to-global 

fire products. The results derived from the group of analyses 

described above are being used to further develop the fire algorithms 

in use, to reduce errors and to improve their performance over a broad 

range of conditions. 

References 

Csiszar, I., Morisette, J.T. and Giglio, L., 2006. Validation of active fire 

detection from moderate-resolution satellite sensors: the MODIS example 

in Northern Eurasia. IEEE Trans. Geo. Rem. Sens., 44(7), 1757-1764. 

Csizar, I., Schroeder, W., 2008. Short-term observations of the temporal 

development of active fires from consecutive same-day ETM+ and ASTER 

imagery in the Amazon: Implications for active fire product validation. 

IEEE J. Sel. Topics in App. Earth Obs. and Rem. Sens., 10.1109/JSTARS. 

2008.2011377. 

Giglio, L., Csiszar, I., Restas, A., Morisette, J.T., Schroeder, W., Morton, D., 

Justice, C.O., 2008. Active fire detection and characterization with the 

Advanced Spaceborne Thermal Emission and Reflection Radiometer 

(ASTER), Rem. Sens. Environ., 2008, doi:10.1016/j.rse.2008.03.003. 

Morisette, J.T., Giglio, L., Csiszar, I., Justice, C.O., 2005. Validation of the 

MODIS active fire product over Southern Africa with ASTER data. Int. J. 

Rem. Sens., 26(19), 4239-4264. 

Schroeder, W., 2008a. Towards an integrated system for vegetation fire monitoring. 

Univ. of Maryland, URL: http://hdl.handle.net/1903/8168. 

Schroeder, W., Csiszar, I. and Morisette, J., 2008b. Quantifying the impact 

of cloud obscuration on remote sensing of active fires in the Brazilian

Advancing the use of multi-resolution remote sensing data to detect and characterize biomass burning 191 

Amazon. Rem. Sens. Env., doi:10.1016/j.rse.2007.05.004. 

Schroeder, W., Prins, E., Giglio, L., Csiszar, I., Schmidt, C., Morisette, J., 

Morton, D., 2008c. Validation of GOES and MODIS active fire detection 

products using ASTER and ETM+ data. Rem. Sens. Env., 

doi:10.1016/j.rse.2008.01.005.

SIGRI - AN INTEGRATED SYSTEM FOR DETECTING, MONITORING, 

CHARACTERIZING FOREST FIRES AND ASSESSING DAMAGE 

BY LEO-GEO DATA 

F. Ferrucci 1 , R. Rongo 1 , A. Guarino 1 , G. Fortunato 1 , 

G. Laneve 2 , E. Cadau 2 , B. Hirn 3 , 

C. Di Bartola 3 , L. Iavarone 4 , R. Loizzo 5 

1 Dipartimento di Scienze della Terra - Università della Calabria, Rende (CS), Italy 

2 Centro di Ricerca Progetto San Marco - Sapienza Università di Roma, Roma Italy 

3 IES Consulting,-Intelligence for Environment and Security, Roma, Italy 

4 Società Aerospaziale Mediterranea S.r.l, Pozzuoli (NA), Italy 

5 Agenzia Spaziale Italiana - Centro di Geodesia Spaziale, Matera, Italy 

Abstract: Forest fires are in the focus of the SIGRI project (Integrated 

System for Fire Risk Management) funded by the Italian Space Agency 

(ASI). The project, started late in November 2008. It is due for completion 

by end-2011. 

The EO part of the project is centred on (1) SAR borne observation in the 

X, C and the L-bands, from ASI and ESA platforms Cosmo-Skymed and 

Envisat; (2) on TIR/MIR/SWIR/NIR - and Red, where appropriate - observation 

by opto-electronic payloads operating at all spatial resolutions from 

2006 onwards (SEVIRI, MODIS, HRVIR, HRG, TM, ASTER, LISS-III) and (3) 

upon SAR very high resolution (Cosmo SkyMed) and V-NIR observation by 

new commercial or dual-use satellites. 

The system, of which the appointed user is the Italian Department of Civil 

Protection (DPC), is expected to deal at once with law enforcement (burn 

scar mapping), preparedness (risk mapping and urban interface fire contingency 

planning) and operational issues (fire detection and propagation prediction). 

It will be demonstrated in three operational theaters (northern 

Italy - Liguria, southern Italy - Calabria, and the island of Sardinia), all 

characterized by high frequency of occurrence of fires, but greatly differing 

in terms of fires style. 


Forest fires are the main threat to forests in the Mediterranean Region, as 

in Italy, where they kept steadily above ca. 7500 fires and ca. 40.000 

hectares per year in the last decade. The management of this phenomenon 

by providing forecasts, performing detections and assessing damages, 

based on the use of advanced technologies (satellite data) is the objective 

of the SIGRI (Integrated System for Fire Risk Management) project. In particular, 

the objective of SIGRI is the development of products which can 

useful to the firefighting activities along all the phases which can be dis- 

193

194 


tinguished in the fire contrasting activity: forecast, monitoring/detection, 

counteract/propagation prediction, damage assessment/recover. 

SIGRI builds upon four main axes: 

a. to define, make and end-to-end demonstrate a functional architecture, 

accounting for all types of operational scenarios of fire-related emergencies 

at local, regional and national level, respectively; 

b. to customize and implement into the system the whole set of unsupervised, 

fire-oriented EO techniques already consolidated at the start of 

final negotiations; 

c. to explore, test, and implement into the system if appropriate, innovative 

unsupervised techniques for fire detection and burn scar mapping; 

d. to provide the overall system with relevant, quantitative decision support 

functionalities, as the near-real-time straightforward modeling of 

fires in controlled environment under known boundary conditions. 

Even if many project have been funded in the last years aiming at increasing 

the exploitation of satellite images in the management of forest fires 

(AFIS, PREVIEW, FIRE-M3, SENTINEL, etc.) no one has a so ambitious objective 

of developing a system capable to support firefighting in real time, 

mostly based on information provided by satellite images. In fact, most of 

these applications are not devoted to the early detection or monitoring of 

fires, and therefore, they are not suitable to support counteracting operations 

and event management. In fact, the main purpose of fire-detection 

applications (except in very rare cases) is that of carrying out a statistical 

study of the events and their possible environmental impact in terms of 

burnt area and variation of the optical characteristics of the atmosphere 

due to burning products (global-scale climate change). Fires occurring in 

the Mediterranean area are rarely significant in terms of burning products 

released in the atmosphere. Nevertheless, they have a dramatic impact on 

the extension of vegetated areas in regions with relatively scarce vegetation 

and on human lives and infrastructure. 

This paper is devoted to introduce the SIGRI project, that is a three years 

lasting project funded by the Italian Space Agency (ASI). Of this “pilot 

project”, initiated officially on November 2008, the main objectives and 

characteristics will be described. 

2 - SIGRI Project generalities (data, methods) 

The SIGRI pilot project will be developed in the mainframe of the project 

“Civil Protection from forest fires”, as a consequence it should take into 

account the institutional requirements, as: the normative aspects in forest 

fires matter, the distribution of responsibilities and competence of the 

authorities involved in the following activities: planning and management 

of the land, dangerousness forecast and risk assessment, prompt fire detection, 

monitoring and management of the fire event, damage assessment.

SIGRI - An Integrated System for Detecting, Monitoring, Characterizing Forest Fires and Assessing damage by LEO-GEO Data 195 

Mission/ Sensor Product 

Cosmo-SkyMed 

IKONOS-2 

QUICKBIRD 

GeoEye-1 

KOMPSAT 2 

Pleiades (when available) 

SPOT-4 / 5 / 

IRS-P6-LISS3 

Landsat 5 / TM 

EO-1 / HYPERION 

or PRISMA 

TERRA e AQUA / MODIS 

Burned areas maps at very high spatial 

resolution 

• Land use maps 

• Vegetation indices 

• Burned areas map 

• Vegetation regeneration maps 

• Dynamics vulnerability maps 

Dynamics vulnerability maps 

• Hot Spot maps 

• Vegetation indices 

• Dynamics vulnerability maps 

MSG SEVIRI Hot Spots maps 

COSMO-SkyMed 

ERS-SAR 

ENVISAT/ASAR 

SAOCOM/SIASGE (when available) 

Burned areas maps 

Dynamics vulnerability maps 

Table 1 - List of satellite sensors possibly used (or of interest) and the products expected by 

the SIGRI project. 

The principal user (reference user) of such a system would be the Italian 

Dept. of the Civil Protection (DPC). Therefore, the demonstrative system 

implemented during SIGRI will be structured in a way to be easily interfaced 

to the DPC infrastructures network and functional centers. 

Nevertheless, the system would be able to product information useful for 

supporting different user types having the role of responding, operationally, 

to the forest fire management according with the guideline and operational 

addresses indicated by DPC. 

The system should provide products based on the use of EO data useful for 

being applied for managing the forest fire risk, for detecting fires, and mapping 

burned areas. Products Specifications will respond to user requirements 

from Italian DPC.

196 


The system should provide products based on the use of EO data useful for 

being applied for managing the forest fire risk, for detecting fires, and mapping 

burned areas. Products Specifications will respond to user requirements 

from Italian DPC. 

Fighting forest fires requires the availability of information and data to be 

used as a support during all the year to the authorities in charge for the 

contrasting the forest fires. For this reason, three operational modalities 

which respond to the requirements of the different actors involved in the 

management of risk associated with the forest fires, have been identified 

that is, the prevention, the extinguishment, land management and damages 

assessment. Consequently, the demonstrative system of SIGRI would be 

operable according with the following modalities: 

Mode STRATEGIC. This operational mode finds application out the fire season 

(SSI); the products generated will support the activities of planning 

and management of the territory for contrasting the fire events. 

Mode TACTICAL. It is applied during the fire season; the generated products 

are characterized by the high frequency of the updating and will provide 

support to the activity of detection, management and monitoring of the 

burning events. 

Mode LEGISLATIVE. It finds application out of the fire season, its products 

regard mainly the development of an archive (cadastral) of the burned 

areas at the conclusion of the time period during with a fire occurrence 

is probable. 

The development of the products to be provided by the project will be 

based, according with the requirements expressed by the DPC, on satellite 

sensor already available but taking into account EO satellite missions 

planned for the near future, too. 

Fig. 1 - Definition of the test area for the demonstration of the SIGRI system. 

2.1 - Demonstration phase 

The demonstration of the products generated in the mainframe of the SIGRI 

project will be carried out on test area selected in the following regions:


Calabria, Sardinia and Liguria. The demonstration activities, in which ASI 

and DPC will be involved, aim at demonstrating the service functionality 

and the effective functioning of the end-to-end system. During the demonstration 

phase the products will be generated for each one of the test area. 

The areas, as shown in Fig. 1, correspond to: Imperia province (Liguria), 

Locride province (Calabria) and Cagliari/Iglesias province (Sardinia). For 

these areas all the products foreseen for the specific system release will be 

provided. 

3 - SIGRI Research activities 

This paragraph is devoted to show aspects of the project mainly deserving 

a research activity. 

3.1 - Fire Detection 

From the point of view of the fire detection, several studies have clearly 

assessed the capability of suitable algorithms to detect fires of very small 

size, compared with the satellite image pixel size, using the brightness 

temperature measured in the MIR and TIR spectral channels. However the 

limited temporal revisit frequency of low earth orbit (LEO) satellites has 

prevented, up to now, the possibility of using satellite observations as a 

support to the real time counteraction of fire events. For this reason, given 

the improved characteristics of the SEVIRI sensor, notwithstanding its limited 

spatial resolution, it is interesting to explore the actual applicability 

of the MSG geostationary satellite, that is able to guarantee a 15 min. 

images temporal resolution. 

The innovation, with respect to other forest fires detection methods based 

on geostationary or low orbit satellites, represented by the SFIDE algorithm, 

consists in the attempt to exploit the quasi-continuous Earth observation 

that SEVIRI provides to set up a wildfire automatic early detection 

system. 

The research activity, in this case, aims: 

- at improving the sensitivity limit of the fire detection algorithm maintaining 

a suitable rate of false alarms. This will be obtained by better 

defining the surface characteristics (emissivity, land cover), improving 

the cloud mask algorithm, and introducing a much accurate description 

of the atmospheric effects. 

- at increasing the reliability of the fire parameters (size, temperature, 

FRP) and at the development of new products (burned biomass, etc.).

198 


3.2 - Burn Scar mapping in SIGRI 

The most effective, passive remote-sensing methods for detecting and mapping 

burn scars in vegetated areas, rely upon the observation of nearinfrared 

(NIR) and short-wavelength infrared (SWIR) bands, with wavelengths 

comprised between 0.8 and 2.3 µm. A method to separate 

reflectance variation due to vegetation damages from changes due to other 

factors influencing the at-satellite reflectance, is that of identifying pseudo-invariant 

features to be used as reference targets in different scenes. 

Such invariants behave as Permanent Reflectors (PRs) ideally in three or 

more infrared bands, and allow (a) improving the robustness of the code 

developed and fine tuned from 2002, nicknamed MYME2. 

The research activities, in this case, aim at improving the already available 

algorithm for detecting burned areas by exploiting temporal series of radar 

images (phase and amplitude). In particular, the possibility to obtain maps 

compliant with the requirements of the Italian rules, by testing different 

techniques, will be assessed 

3.3 - Fire risk maps 

Regarding the development of a daily map of fire risk/vulnerability index, 

among the indices already developed/proposed (FPI, FWI, etc.) worldwide 

the one under consideration should be characterized by the fact that it has 

to be based, at least partially, on satellite images. As consequence such an 

index would be based on the FPI (Fire Probability Index) recognized as an 

index able to identify those areas at risk of fire. The addition of further 

information like the presence of infrastructure, or areas of relevant importance 

will lead to define a fire vulnerability index. Based on: an ameliorate 

description of the vegetation type (fuel type), by using hyperspectral and 

radar images, and a satellite based estimate of the vegetation water content 

an improvement of the forecast based on the fire probability index 

would be obtained 


The present paper aims at presenting the objectives and expected products 

of the three-years-long pilot project called SIGRI, recently funded by 

Italian Space Agency. This project represents one of the 7 initiatives funded 

by ASI intended to enhance the utilization of satellite images for planning, 

managing and contrasting natural or anthropic hazardous events like 

wild fires, landslides, floods, sea and air pollution. A description of the 

project has been presented. Results of the research activity and of the operational 

application of the system will be presented in the next future, as 

they will be obtained.


References 

Di Bartola C., Hirn B.R., Ferrucci, F., 2005. MYME2: a multi-payload integrated 

procedure for the automated, high-resolution remote sensing of 

Burn Scars. IEEE-IGARSS 2005, Geoscience and Remote Sensing 

Symposium, Seoul (S. Korea) 25-29 July 2005, paper no. 20238 

Gambardella, A., Ferrucci, F., Marzoli M., 2002. Il satellite e la legge-quadro 

in materia di incendi boschivi. Antincendio; 16, 35-39. 

Hirn, B.R., Ferrucci, F., 2008. Southern Italy Burn Scar Mapping by MYME2 

Procedure using IRS-P6 LISS3. IEEE-IGARSS 2008, Geoscience and 

Remote Sensing Symposium, Boston (MA), 7-11 July 2008; 4, 746-749 

Hirn, B.R., Di Bartola C., Ferrucci F., 2007. Improvement and validation of 

MODIS performance in automated detection and extent estimate of wildfires. 

IEEE-IGARSS 2007, Geoscience and Remote Sensing Symposium, 

Barcelona (Spain) 23-28 July 2007, 3004-3007 

Laneve, G., Castronuovo, M.M., Cadau, E.G., 2006a. Continuous Monitoring 

of Forest Fires in the Mediterranean Area Using MSG. IEEE TGRS 44(10): 

2761-2768. 

Laneve, G., Cadau, E.G., 2006b. Assessment of the fire detection limit using 

SEVIRI/MSG sensor, Geoscience and Remote Sensing Symposium, 

IGARSS06, 4157-4160.

IV BURNED LAND MAPPING, 

FIRE SEVERITY DETERMINATION, 

AND VEGETATION RECOVERY 

ASSESSMENT

IMPROVEMENT OF DNBR BURNT AREAS DETECTION PROCEDURE 

BY PHYSICAL CONSIDERATIONS BASED ON NDVI INDEX 

R. Carlà & L. Santurri 

National Research Council - Institute of Applied Physics “N. Carrara” (IFAC-CNR), 

Sesto Fiorentino (FI), Italy 

satellit@ifac.cnr.it; l.santurri@ifac.cnr.it 

L. Bonora & C. Conese 

National Research Council - Institute of Biometerology (IBIMET-CNR), 

Sesto Fiorentino (FI), Italy 

l.bonora@ibimet.cnr.it; c.conese@ibimet.cnr.it 

Abstract: The detection and mapping of burned areas have been long studied 

and several algorithms have been developed, such as that used in the 

dNBR (differential Normalized Burn Ratio) method. This work aims at evaluating 

the performance of the dNBR as regards to both the positively 

detected burnt areas and the false alarms, when applied in a Mediterranean 

region. The dependence of the dNBR method efficiency from the adopted 

threshold is assessed, and the performances in terms of omission and commission 

errors are evaluated in mapping and detection tasks. A new 

enhanced version of the dNBR method based on the introduction of four 

simple environmental conditional tests is then presented and the related 

results are comparatively reported. 


The Tuscany is a typical italian mediterranean region, affected each year by 

many fires of small dimensions. Several experimental tests reported in the 

scientific literature demonstrate the usefulness of satellite data for the 

observation of vegetated areas affected by fire (Chuvieco & Congalton, 

1988; Koutsis and Karteris, 1998; Pereira and Setzer, 1993; Epting et al., 

2005), but up to now very little attention has been paid to the recognition 

and analysis over large territories of areas affected by small fires of very 

few hectares (Martin et al., 2006). 

For the most part, the algorithms aimed at detecting burnt areas from satellite 

data are based on multitemporal analysis. Among the others, the 

Normalized Burnt Ratio (NBR) index and its differential form (dNBR), that 

is the NBR index temporally differenced between before and after the fire 

season, have been widely tested on vast territories and have demonstrated 

their efficiency on large fire events (Key et al., 1999). In order to obtain a 

binary map of burnt / not burnt pixels by the dNBR method, a spatially 

invariant or locally adaptive threshold needs to be defined. This work aims 

203

204 

IV - BURNED LAND MAPPING, FIRE SEVERITY DETERMINATION, AND VEGETATION RECOVERY ASSESSMENT 

to evaluate the performance of the dNBR method as regards to the positively 

detected burnt areas and the related false alarms, when applied at 

regional scale to detect and map fires of small dimensions in the 

Mediterranean region. The dNBR method has been tested on two Landsat- 

TM images covering the whole Tuscany region (Italy), collected respectively 

before and after the fire season of summer 2000 (characterized by 18 

medium to small fire events occurred in considered area). A modified version 

of the dNBR method obtained by the analysis of four simple environmental 

conditional tests is then presented and the related results are comparatively 

reported. 


The differential NBR index (dNBR) is the difference among the NBR index 

evaluated before (NBR b ) and after (NBR a ) the fire season, where the NBR 

index is defined (if Landsat TM sensor data are considered) as: 

NBR = (B 4 - B 7 )/(B 4 + B 7 ) 

being B i the surface spectral reflectances as measured in bands 4 and 7 

after a suitable calibration (Thome et al., 1997). The procedure to detect 

burnt area based on dNBR index consists in classifying as burned the pixels 

whose dNBR exceeds a given threshold K i . A suitable threshold needs 

therefore to be defined according to a desired trade-off among the number 

of pixels correctly classified as burned (hereafter true positive) and the 

number of unburned pixels wrongly classified as burned (hereafter false positive). 

The analysis of this trade-off is the main focus of this work, in which 

the performance of the dNBR method has been evaluated by varying the 

threshold value. The obtained results are than compared with those of four 

modified dNBR method, all relying on a second processing step (to be 

applied after the dNBR one) based on physical considerations coming from 

the literature and resulting in the following thresholding: 

a) Pixels of the image acquired after the fire characterized by a value of 

B5 – B7 higher than 0.08 are considered unburned. 

b) Pixels of the image acquired after the fire with a value of B7 lower than 

0.06 are considered unburned. 

c) Pixels of the image acquired before the fire with a NDVI lower than 0.2 

are considered unburnt (pixel must be vegetated before the fire). 

d) Pixel with the value of the NDVI index in the image acquired before the 

fire lower than that of the image acquired after are considered unburned 

(vegetation after the fire must be diminished with respect to the image 

acquired before the fire).

Improvement of dNBR burnt areas detection procedure by physical considerations based on NDVI index 205 

3 - Results 

The performances of the dNBR method have been assessed by evaluating 

the number of true positives, true negatives and false positives (Story 

Congalton, 1986) by varying the threshold value. 

Figure 1 - The producer accuracy with the threshold. 

The results reported in percentage in Fig. 1 show that the performances of 

the dNBR method are not completely satisfactory, from “fire mapping” point 

of view, that is in highlighting all the burned pixels. As a matter of fact it 

is not possible to find a good trade-off between the required high number 

of true positives and a low number of false positives. The dNBR method has 

been then analysed in terms of “fire detection”. The fire detection task 

needs that only one pixel (at least one!) for each burned area is flagged as 

burned, and thus requires a lower number of true positives. It has been 

found that the best (higher) possible threshold to have yet all the burn 

scars detected by at least a true positive is equal to 0.10. By using this 

value, a percentage of 81.5% (2048 of 2513) of the burnt pixel are correctly 

classified, but these true positives are countered by more than five millions 

(5,378,754) false positives, a number trivially to high to consider the dNBR 

method useful for fire area detection at regional scale. In order to increase 

the performances, the aforementioned four additional criteria (a – d) have 

been introduced and tested. The related performances are reported in Table 

1 together with those of the original method. It can be noticed how the 

proposed upgrades increase the efficiency of the dNBR methods; nevertheless, 

the performances are still not satisfactory, and other additional criteria 

are under evaluation in order to obtain a more efficient detection procedure.

206 


True positive False Positive 

dNBR 2,124 (100 %) 5,378,754 (100%) 

dNBR + Upgrade a) 2058, (96.89 %) 3,485,823 (64.8%) 

dNBR + Upgrade b) 2,124 (100 %) 5,314,345 (98.8%) 

dNBR + Upgrade c) 2,101 (98.91 %) 4,811,680 (89.46%) 

dNBR + Upgrade d) 2,082 (98.00 %) 4,705,166 (87.48%) 

dNBR + a),b),c) e d) 2,019 (95 %) 2,831,343 (52.64%) 

Table 1 - The performances of the upgraded dNBR method. 


The dNBR method performances in the mapping and detecting burnt area 

have been assessed, in terms of true and false positives. The method has 

been found not so much suitable for the unsupervised burnt area detection 

and mapping in a regional context, because of the high number of false 

positives. An upgrade of this method based on some physical considerations 

has been proposed and assessed. The resulting performances show a 

consistent reduction of the false positives to the detriment of a slight 

reduction of the true positives. 

Reference 

Chuvieco, E. and Congalton, R.G., 1988. Mapping and inventory of forest 

fires from digital processing of TM data, Geocarto International, vol. 4, 

pp. 41-53 

Epting, J., Verbyla, D., Sorbel, B., 2005. Evaluation of remotely sensed 

indices for assessing burn severity in interior Alaska using Landsat TM and 

ETM+, Remote Sensing of Evnvironment, vol. 96, pp. 328-339. 

Key, C.H. and Benson, N.C., 1999. The Normalized Burn Ratio (NBR): a 

Landsat TM radiometric measure of burn severity, www.nrmsc.usgs.gov/ 

research/ndbr.htm. 

Koutsias, N. and Karteris, M., 1998. Logistic regression modelling of multitemporal 

Thematic Mapper data for burned area mapping, International 

Journal of Remote Sensing, vol. 19, pp. 3499-3514. 

Martin, M.P., Gomez, I., Chuvieco, E., 2006. Burnt area index (baim) for 

burned area discrimination at regional scale using Modis data, V Int. 

Conf. On Forest Fire Research, Figuerira da Foz, Portugal, 27-30 nov. 

2006, (D.X. Viegas Ed.). 

Pereira, M.C. and Setzer, A.W., 1993. Spectral characteristics of fire scars in 

Landsat-5 TM images of Amazonia, International Journal of Remote 

Sensing, vol. 14, pp. 2061-2078. 

Story, M., Congalton, R.G., 1986. Accuracy Assessment: A User’s Perspective,

Improvement of dNBR burnt areas detection procedure by physical considerations based on NDVI index 207 

Photogrammetric Engineering and Remote sensing, Vol. 52, No. 3, March 

1986, pp. 397-399. 

Thome K., Markham B., Barker J., Slater P., Biggar S. Radiometric calibration 

of Landsat. Photogrammetric engineering and remote sensing 1997, 

vol. 63, no7, pp. 853-858.

Abstract: The aim of this paper is to show the results of a comparative 

analysis of some of the most commonly used spectral indexes in burnt land 

mapping applications. The objective is to verify its operative consistency 

between ASTER and MODIS data derived indexes. The analyzed indexes were 

the SVI, NDVI, TVI and SAVI for ASTER and BAIM and NBR for MODIS data. 

This test has been focused on establishing the discrimination ability of 

each index between the recently burned zones and other land covers using 

a post-fire image from Gran Canaria (Canary Islands-Spain) 2007 fire. The 

results have been compared with the burnt area perimeter provided by a 

SPOT multitemporal image study. 


BURNT AREA INDEX USING MODIS AND ASTER DATA 

A. Alonso-Benito, P.A. Hernandez-Leal, A. Gonzalez-Calvo, 

M. Arbelo, A. Barreto & L. Nunez-Casillas 

Grupo de Observacion de la Tierra y la Atmosfera (GOTA), Avda. Astrofisico Francisco Sanchez s/n, 

Facultad de Fisicas, Universidad de La Laguna, S/C Tenerife, Islas Canarias (España) 

asaloben@gmail.com 

Satellite data constitute a useful tool for a rapid assessment, usually at no 

cost, of the area affected by a wildfire, and the pixel burnt severity estimation. 

Many authors have written about the appropriateness of different 

satellite sensors in the mapping of burned areas ie. AVHRR and MODIS 

(Chuvieco et al., 2005) and ASTER (Nikolakopoulus, 2003). During the summer 

of 2007, a large fire took place in Gran Canaria Island (Spain) affecting 

nearly 15,000 ha. This region, in the south of the island, has been 

taken as the test site. 

2 - Data and methodology 

With the aim of testing the proposed indexes for burnt pixels identification, 

an ASTER Level 1B image (March 6, 2008) has been used. All ASTER evaluated 

indexes use VNIR and SWIR bands of the sensor, which have been corrected 

for reflectance values R TOA as described in (ERSDAC, 2007). On the 

other hand, a Terra-MODIS Level 1B image from 15 May 2008 has been used 

209

210 


to derive BAIM and NBR indexes. 

The Simple Vegetation Index (SVI), the Normalized Difference Vegetation 

Index (NDVI), the Transformed Vegetation Index (TVI) and the Soil Adjusted 

Vegetation Index (SAVI) have been widely used for fire studies 

(Nikolakopoulus, 2003) and have been selected in this case for ASTER burnt 

pixels discrimination. 

On the other hand, the Burnt Area Index for MODIS (BAIM) is an adaptation 

of the BAI (Burnt Area Index) originally developed to be used with 

NOAA-AVHRR, data as described in Martín et al. (2005). As convergence 

values the percentiles 5 and 95 from reflectance values on NIR and SWIR 

bands have been respectively used (González-Alonso et al., 2007), being 

ρc IRC = 0,032 y ρc SWIR = 0,215. The Normalized Burnt Ratio (NBR) is a similar 

ratio to the one described in the NDVI formulation, but using NIR and 

SWIR bands (Key and Benson, 1999). 

1.1 - Classification method 

The algorithm Support Vector Machines (SVM) falls within the supervised 

classification (Vapmik, 1995). After analyzing all kernels, the Radial Basis 

Function (RBF) equation has been selected. We have come to this conclusion 

after studying the Kappa ratios and percentages of success achieved 

with each kernel, noting that the group of locals achieved values higher 

than 91% with the sigmoid and RBF equations, while the results were 

somewhat worst with global kernels, in which distant points have great 

influence and the studied area presents a very diffuse boundary between 

burned and unburned soil. 

Using SVM algorithm we have analyzed the suitability of each ASTER index 

to distinguish the burned area from the rest of the unburned island (Figure 

1).

Burnt area index using MODIA and ASTER data 211 

Figure 1 - (a) Results of the SVM algorithm application inside the fire perimeter for Gran 

Canaria ASTER image. In grey, unburned areas; in white, reservoirs; and in black, burnt areas. 

(b) Results of the SVM algorithm application to Gran Canaria MODIS image. In grey, unburned 

areas; and in black, burnt areas. 

According to Chuvieco et al. (2005), the global reliability of the algorithm 

and the confidence interval of the real reliability are obtained with the 

omission and commission values yielded. Similarly, considering a 95% reliability, 

the degree of reliability of the burnt area mapping, the sampling 

error and the confidence interval are worse for MODIS indexes than those 

obtained for ASTER. 


Using SVM algorithm, we have analyzed the suitability of each index to distinguish 

the burned area from the rest of the unburned island (Figure 1). 

Within burnt area perimeter different areas have been selected as training 

areas for SVM algorithm application and different Kernels have been applied 

to the SVM algorithm although the RBF gave better results. The category 

water has been chosen on the basis of existing reservoirs in the south of 

the island.

212 


Among the indexes used in this test for Gran Canaria Island the SVI is the 

one that yields better results in the application to ASTER data categorization 

of pixels affected by a fire, this index presents minor errors of commission 

for classified burnt pixels, and the NDVI is the one that presents 

minor errors of omission. 

INDEX 

ACCURACY 

KAPPA 

COEFFICIENT 

SAMPLING 

ERROR 

CONFIDENCE 

INTERVAL (F) 

RGB 92,05% 0,8385 26,10% 92,0 ± 0,5117 

NDVI 86,39% 0,7180 33,10% 86,4 ± 0,6486 

ASTER SAVI 85,59% 0,7014 33,88% 85,6 ± 0,6641 

SVI 86,50% 0,7213 32,98% 86,5 ± 0,6463 

TVI 86,41% 0,7188 33,99% 85,5 ± 0,6663 

RGB 95,72% 0,8813 43,19% 95,7 ± 0,8466 

MODIS NBR 77,48% 0,4861 164,59% 77,5 ± 3,2259 

BAIM 77,59% 0,5257 159,53% 77,6 ± 3,1268 

NDVI SAVI SVI TVI 

ASTER Comis. Omis. Comis. Omis Comis. Omis. Comis. Omis. 

Burnt 16,14 4,76 16,08 4,65 17,01 5,41 15,98 4,82 

Volcanic 4,6 9,7 5,12 16,94 4,66 9,76 4,65 9,68 

Unburned 8,65 27,16 10,05 27,51 9,64 25,75 8,84 26,87 

NBR BAIM 

MODIS Comis. Omis. Comis. Omis 

Burnt 0 27,41 0 28,92 

Unburned 55,77 0 49,84 0 

Table 1 - Results of the SVM algorithm application to Gran Canaria ASTER and MODIS images. 

Also, results of sampling error, comission and omission errors and confidence interval. 

In relation to the MODIS sensor data study the NBR presents worse percentages 

of accuracy (Table 1) compared to the BAIM values. Nevertheless 

commission and omission errors for these indexes are too high comparing 

to the ASTER ones, probably as an effect of the different spatial resolution 

for these two types of satellite data. The sampling errors and confidence 

intervals are worse for MODIS indexes than those obtained for ASTER, probably 

because of the vegetation composition inside the fire perimeter and 

the orographic influence in an insular environment in which the spatial resolution 

of 250 and 500 m. of MODIS channels may be not enough.


Burnt area index using MODIA and ASTER data 213 

This paper shows the estimation of burnt area in a large wildfire that took 

place during summer 2007 in Gran Canaria island (Spain), limiting the 

number of classes to the minimum, in order to check if a rapid assessment 

of the affected area with only one ASTER or MODIS post-fire image is possible. 

The orographic complex effects in an island like this, make the ASTER 

derived indexes more suitable due to their better spatial resolution. 

Nevertheless, high errors of commission and omission are detected in some 

cases due to the special vegetation characteristics of this test area. A 

future work under development will also analyze the error due to the pixel 

resolution in terms of a classification method based on object-oriented and 

neural networks. 

Acknowledgments 

This work has been supported by the “Ministerio de Educacion y Ciencia” 

(Spain), under Project CGL2007-66888-C02-01/CLI. 

References 

Chuvieco, E., Ventura, G., Martín, M.P., Gómez, I., 2005. Assessment of multitemporal 

techniques of MODIS and AVHRR images for burned land mapping. 

Remote Sensing of Environment 94: 450-462. 

ERSDAC, 2007. ASTER User’s Guide. Earth Remote Sensing Data Analysis 

Center. http://www.science.aster.ersdac.or.jp/en/documents/users_ 

guide/ 

González-Alonso, F., Merino de Miguel, S., Cuevas Gonzalo, J.M., 2007. Un 

nuevo algoritmo para la cartografía de áreas quemadas a partir de información 

NIR, SWIR y TIR. Revista de Teledetección. ISSN: 1133-0953. 

2007. 28: 97-105. 

Key, C., Benson, M., 1999. The Normalized Burned Ratio, a Landsat TM radiometric 

index of burn severity incorporating multi-temporal differencing. 

U.S. Department of the Interior Northern Rocky Mountain Science 

Center. 

Martín, M.P., Gómez, I., Chuvieco, E., 2005. Performance of a burned-area 

index (BAIM) for mapping Mediterranean burned scars from MODIS data. 

Proceedings of the 5th International <strong>Workshop</strong> on Remote Sensing and 

GIS applications to Forest Fire Management: Fire Effects Assessment. 

(Zaragoza), GOFC-GOLD, EARSeL: 193-198. 

Nikolakopoulus, K.G., 2003. Use of Vegetation Indexes with ASTER VNIR data 

for burned areas detection in Western Peloponnese, Greece. Geoscience 

and Remote Sensing Symposium, IGARSS apos; 03. Proceedings. 2003 

IEEE International Volume 5, Issue, 2003 Page(s): 3287-3289.

214 


Stroppiana, D., Tansey, K., Gregoire, J.M., Pereira, J.M.C., 2003. An algorithm 

for mapping burnt areas in Australia using SPOT-VEGETATION data. 

Geoscience and Remote Sensing, IEEE Transactions on 

Volume 41, Issue 4, April 2003 Page(s): 907-909. 

Vapmik, V., 1995. The nature of statistical learning theory. New York, NY: 

Springer-Verlag.

SATELLITE DERIVED MULTI-YEAR BURNED AREA PERIMETERS WITHIN 

THE NATIONAL PARKS OF ITALY 

P.A. Brivio 1 , B. Petrucci 2 , M. Boschetti 1 , P. Carrara 1 , M. Pepe 1 , 

A. Rampini 1 , D. Stroppiana 1 & P. Zaffaroni 1 

1 CNR-IREA, Institute for Electromagnetic Sensing of the Environment, Milan, Italy, 

brivio.pa@irea.cnr.it; boschetti.m; carrara.p; pepe.m; rampini.a; stroppiana.d; zaffaroni.p 

2 Ministero dell’Ambiente, Direzione Protezione Natura, Roma, Italy, 

petrucci.bruno@minambiente.it 

Abstract: Vegetation fires play a key role in land cover and land use change 

and satellite data are a unique source of information for fire monitoring. In 

Italy, fires severely affect forest resources and monitoring and mapping 

have been recognized as key activities for the prevision, the prevention and 

for fighting against fires. We present burned area maps derived mainly from 

ASTER satellite images over the National Parks for the period 2001 to 2005. 

They are available through an Inspire compliant Web Map Service (WMS) at 

the CNR-IREA geo-portal and the the “Portale Cartografico Nazionale” of the 

Italian Ministry of the Environment. 


Land cover and land use dynamics are object of interest due to their key 

role in the global carbon cycle and their impacts on climate change. In this 

context fire is probably the most important factor driving land cover 

changes that occurs in almost all the ecosystems around the World 

(Thonicke et al., 2001). Mediterranean countries in Europe, among which 

Italy, are particularly affected by vegetation fires during the summer season 

when the weather is hot and dry; yet most of them do not have a geographic 

database of fire events and only recently remote sensing technology 

has been explored for integrating traditional field monitoring (Paganini 

et al., 2003; Piccoli and Cattoi, 2007). We present the outcome of a project 

funded by the Italian Ministry of the Environment, Direzione Protezione 

Natura, that aimed at producing a geo database of the burned perimeters 

in the Italian National Parks for the period 2001-2005 from satellite 

images. 

215

216 



In the framework of the project we used mainly images acquired by the 

NASA-Terra Advanced Spaceborne Thermal Emission (ASTER) sensor that 

provides measurements of the surface reflectance in the visible (VIS), near 

infrared (NIR) and shortwave infrared (SWIR) regions of the spectrum. Since 

the ASTER coverage was not deemed sufficient for the project’s requirements 

over the Aspromonte, Cilento, Gargano and Pollino National Parks in 

2005, we acquired SPOT 2, 4 and 5 images in the framework of the OASIS 

(Optimising Access to Spot Infrastructure for Science) Programme. The 

database is therefore composed by about 500 ASTER scenes and 20 SPOT 

images. Only satellite images that contained burned surfaces were 

processed to derive burned area maps with one of the following methods: 

photo-interpretation, maximum likelihood supervised classification and 

multi-index multi-threshold approach (Brivio et al., 2009). The multi-index 

multi-threshold approach was applied approximately to half of the satellite 

images selected for classification; for this reason we focused the validation 

exercise only on burned area maps derived with this technique from ASTER 

images, that constitute the core of our satellite database. Accuracy measures 

(Congalton, 1991) were derived by comparison with a reference database 

built by photo-interpretation of about 20 ASTER scenes. 


The major result is a geo database of burned area maps that fills the gaps 

of the 2001-2005 historical archive of fire information for protected areas 

in Italy. 

Year n. images n. burned polygons Area [ha] 

2001 65 140 1179 

2002 92 49 228 

2003 140 177 1172 

2004 90 157 1441 

2005 74 129 1018 

Total [ha] 461 652 5038 

Table 1 - Number of satellite images (ASTER and SPOT) used for mapping burned perimeters, 

number of detected burned polygons and total burned area between 2001 and 2005. 

We estimated that more than 5000 ha were affected by fires in the study 

period and the distribution of burned polygons and burned surface among 

the years is given in the table. About 75% of the total burned area was 

within the boundaries of parks located in Southern Italy (i.e. Cilento, 

Pollino, Aspromonte and Gargano). It is likely that the much higher tem-

Satellite derived multi-year burned area perimeters within the national parks of Italy 217 

peratures and drier conditions make the Southern regions of the country 

more susceptible to fire ignition and spread although the two main causes 

of vegetation fires are arson and careless. 

The figure shows burned perimeters identified in the available satellite 

images for each year between 2001 and 2005 in the Aspromonte and 

Cilento Parks. 

Burned area maps for period 2001-2005 over Aspromonte (left) and Cilento (right). 

On the background is the distribution of forest and agriculture/pasture land 

cover classes derived from the Corine Land Cover map. Fire monitoring and 

mapping in National Parks is important since they are established to protect 

natural (forest) resources. 

Burned area in the Aspromonte and Cilento Parks for the vegetated land cover classes. 

Burning can significantly vary from year to year as in the case of 

Aspromonte where we estimates 6.5 and 17.8 hectares of burned surface in 

2002 and 2004, respectively. Also the distribution of the total area among 

the land cover classes can vary. In 2001 in Aspromnte more than 50% of 

the total burned area was mapped in forest (121.88 ha) whereas in 2005 

the most affected land covers were grassland and shrubland (134.36 ha, 

49% of the total). In the case of Cilento the smallest amount of burned sur-

218 


face is detected in 2002 (36.07 ha) only in the crop (74%) and grass/shrub 

(26%) classes. In this park burned area is also detected in pastureland 

(data not shown in the histograms) although it is only 3% of the five-year 

total (1157.55 ha). This type of information can be very useful for the 

assessment of land cover changes driven by fires especially when they 

affect natural resources such as protected forests. 

Accuracy assessment showed that on average the overall accuracy of ASTER 

scene classification is better than 90% and omission and commission errors 

are about 50% and 15%; low commission errors highlight that classification 

is conservative with few false alarms. These figures quantify the average 

scene accuracy and not the accuracy of the annual burned area maps 

obtained by summing burned perimeters. 

Since our results can be of interest for scientific groups working on fire 

monitoring, we distribute burned area maps through a Web Map Server 

(WMS) (URL: http://geoportal.irea.cnr.it:8080/geoportal/local_it.jsp) created 

at CNR-IREA premises for purposes of the European Project IDE- 

Univers (www.ideunivers.eu). These burned areas perimeters are also available 

through the “Portale Cartografico Nazionale” (URL: http://www.pcn. 

minambiente.it/PCN/progetto_incendi.htm) of the Italian Ministry of the 

Environment. 


In this work we built a geographic database of burned area perimeters for 

Italian National Parks derived from ASTER and SPOT satellite images 

between 2001 and 2005. Accuracy measures highlighted the conservative 

character of the classifications with a low rate of false alarms (average 

scene commission error is less than 15%). 

References 

Brivio, P.A., Petrucci, B., Boschetti, M., Carrara, P., Pepe, M., Rampini, A., 

Stroppiana, D., Zaffaroni, P., 2009. A multi-year geographic database of 

fire affected area derives from satellite images in the National Parks of 

Italy. Italian Journal of Remote Sensing, Vol 41, n. 2. 

Congalton, R.G., 1991. A review of assessing the accuracy of classifications 

of remotely sensed data. Remote Sensing of Environment, 37, 35-46. 

Paganini, M., Arino, O., Benvenuti, M., Cristaldi, M., Bordin, M., Coretti, C., 

Musone, A., 2003. ITALSCAR, a regional burned forest mapping demonstration 

project in Italy. International Geoscience and Remote Sensing 

Symposium, 2003. IGARSS ‘03. Proc. 2003 IEEE Volume 2, 21-25 July 

2003: 1290-1292. 

Piccoli D., Cattoi M., 2004. I rilievi delle aree percorse dal fuoco effettuati 

dal Corpo Forestale dello Stato per le attività di polizia giudiziaria.

Satellite derived multi-year burned area perimeters within the national parks of Italy 219 

Presentazione del seminario “Telerilevamento e spazializzazione nel 

monitoraggio forestale e ambientale per la difesa degli ecosistemi” 

Università degli Studi del Molise, Pesche (IS), 24 January 2007. 

Thonicke, K., Venevsky, S., Sitch, S., Cramer, W., 2001. The role of fire disturbance 

for global vegetation dynamics: coupling fire into a Dynamic 

Global Vegetation Model. Global Ecology & Biogeography, 10, 661-677.

BURN SEVERITY AND BURNING EFFICIENCY ESTIMATION USING 

SIMULATION MODELS AND GEOCBI 

A. De Santis 1 , G.P. Asner 2 , P.J. Vaughan 3 , D. Knapp 2 

1 Ingenieria y Servicios Aeroespaciales, SA (INSA), Systems & Earth Observation Department, 

Madrid, Spain, angela.de.santis@insa.org 

2 Stanford University, Department of Global Ecology, Stanford (CA) 

gpa@stanford.edu; deknapp@stanford.edu 

3 Laboratorio de Espectro-radiometria y Teledetección Ambiental. CCHS-CSIC 

Madrid, Spain, patrick.vaughan@cchs.csic.es 

Abstract: Uncertainties in burning efficiency (BE) estimates can lead to 

high errors in fire emission quantification, due to the spatial variability of 

fuel consumption within the burned area. Burn severity (BS) can be used 

to improve the BE assessment. Therefore, in this study, a burn severity map 

of two large fires in California was obtained by inverting a simulation model 

and using a post-fire Landsat TM image. Estimated BS were validated 

against field measurements, obtaining a high correlation (r 2 =0.85) and low 

errors (Root Mean Square Error, RMSE = 0.14). The BS map obtained was 

then used to adjust BE reference values per vegetation type found in the 

area before the fire. The adjusted burning efficiency (BEadj) was compared 

to the burned biomass, which was estimated by subtracting NDVI from preand 

post-fire images. Results showed a high correlation for hardwoods 

(r 2 =0.72) and grasslands (r 2 =0.69) and medium correlation (r 2 ~0.5) for 

conifers and shrubs. In general for all vegetation types, BEadj perform better 

(r 2 =0.4 - 0.72) than literature BE (r 2

222 


system of fire statistics. However, only a few examples of burning efficiency 

(BE) quantification can be found in the literature. Most regional to global 

emission estimates assume an average BE, either (i) for each major vegetation 

or fuel type, or (ii) over the entire burned area, considering in both 

cases all vegetation as completely burnt, which can lead to uncertainties 

ranging from 23% to 46%. A different approach consists of estimating the 

degree of biomass consumption in terms of burn severity (BS). Therefore, 

this paper reports on work done to extend and improve the method, and to 

test and apply it in a new region. The first goal was to test the simulation 

model proposed by De Santis et al. (2009) to estimate BS in a new environment 

and at a different scale, in this case two very large forest fires in 

California. The second aim was to apply the BS estimated from satellite 

imagery, in an attempt to adjust the BE values found in the literature. 


Figure 1 - Methodological workflow followed in this study. 

The study area corresponds to two very large forest fires (summer 2008) 

located in the Big Sur region (Monterey, California, USA): the Indians Fire 

(31,590 ha) and the Basin Complex Fire (65,942 ha). The study was comprised 

of two consecutive phases (figure 1). First, BS was estimated using 

a post-fire Landsat TM ortho-image and the inversion of the simulation 

model proposed by De Santis et al. (2009). This approach was validated 

using field data (40 plots), in terms of GeoCBI (De Santis and Chuvieco, 

2009). In the second phase, the resulting BS map was used to scale liter-

Burn severity and burning efficiency estimation using simulation models and geoCBI 223 

ature-based BE values per vegetation type found in the pre-fire scenario 

(table 1). Vegetation types were extracted from a Californian vegetation 

map and grouped in four main classes: grass, shrubs, conifers and hardwoods. 

The percentage of burned biomass, extracted from a pre- and a postfire 

Landsat TM images in terms of differences in pre-and post- NDVI, was 

correlated against literature BE (BEref, commonly used for fire emission 

estimation, table 2) and also against the adjusted BE (BEadj, using BS), in 

order to compare both efficiency estimations. 

BS 

GRASS 

BEadj PER VEGETATION TYPE (CWHR) 

SHRUB CONIFER HARDWOOD 

LOW 0.83 0.71 0.25 0.25 

MODERATE 0.90 0.84 0.47 0.40 

HIGH 0.98 0.95 0.65 0.56 

Table 1 - Adjusted burning efficiency values (BEadj) using the burn severity (BS) information. 

VEGETATION TYPE BEref REFERENCES 

GRASS 0.98 

SHRUB 0.95 

CONIFER 0.48 

HARDWOOD 0.45 

2006 IPCC Guidelines for 

National Greenhouse Gas 

Inventories 

Table 2 - BE values extracted from the literature (BEref) and used in the validation process. 

3 - Results 

dNDVI 

BE INDEX 

BEref 

BEadj 

GRASS SHRUB CONIFER 

y=-7E-16x + 0.98 

R2 = _ _ _ 

y = 0.58x + 0.71 

R2 = 0.69 

y = -4E-15x + 0.95 

R 2

224 


high correlation (R 2 =0.845) between observed and simulated GeoCBI and 

an almost 1:1 linear fitting (slope ~1 and constant ~0) as well as a very 

low dispersion of the values. The total RMSE was also very low (RMSE= 

0.14). In addition, the accuracy was quite homogeneous throughout the 

range of GeoCBI values observed in the field (1.68 to 3.00). For the validation 

of BE estimation, the results of the linear fitting between dNDVI and 

both BEref and BEadj (table 3) showed that the correlation coefficients 

were considerably higher in the case of BEadj, ranging from 0.45 (shrubs) 

to 0.72 (hardwoods). In the case of BE, the correlation was practically inexistent 

(table 3) for all vegetation types. 


The simulation model selected for this study had already been successfully 

applied in other Mediterranean fires in Spain and Portugal (De Santis et al., 

2009). Validation results showed that the accuracy of the model inversion 

was considerably homogeneous in all ranges of GeoCBI values, and the 

RMSE = 0.14 was, in fact, slightly lower than the ones obtained in Spain 

and Portugal (RMSE = 0.18-0.21). In addition, the model was tested over a 

wider range of GeoCBI values. The results confirmed the consistency of the 

model, which was independent of specific site characteristics, at least within 

Mediterranean/temperate environments. The BE validation confirmed 

that the use of GeoCBI as an adjustment factor improved BE estimation 

considerably for all vegetation types analyzed. From a broader perspective, 

this study must be regarded as a first attempt to improve BE estimation. 

The fundamental improvements shown in BE estimation still lies on the 

high accuracy of BS estimation. Therefore, there is room for improvement, 

especially concerning alternative ways of optimizing the use of BS to adjust 

BE estimation. 

References 

Barbosa, P.M., Cardoso Pereira, J.M., and Gregoire, J.-M., 1998. 

Compositing criteria for burned area assessment using multitemporal 

low resolution satellite data. Remote Sens. Environ. 65, 38-49. 

Crutzen, P. and Andreae, M., 1990. Biomass burning in the tropics: Impact 

on atmospheric chemistry and biogeochemical cycles, Science, 250, 

1669-1678. 

De Santis, A., Chuvieco, E. & Vaughan, P.J., 2009. Short-term assessment 

of burn severity using the inversion of PROSPECT and GeoSail models. 

Remote Sensing of Environment, 113 (1), 126-136. 

De Santis, A. & Chuvieco, E., 2009. GeoCBI: a modified version of the 

Composite Burn Index for the initial assessment of the short-term burn 

severity from remotely sensed data. Remote Sensing of Environment, 113

Burn severity and burning efficiency estimation using simulation models and geoCBI 225 

(3), 554-562. 

IPCC2006 Guidelines for National Greenhouse Gas Inventories. Volume 4: 

Agriculture, Forestry and Other Land Use. Harald Aalde (Norway), Patrick 

Gonzalez (USA), Michael Gytarsky (Russian Federation), Thelma Krug 

(Brazil), Werner A. Kurz (Canada), Rodel D. Lasco (Philippines), Daniel 

L. Martino (Uruguay), Brian G. McConkey (Canada), Stephen Ogle (USA), 

Keith Paustian (USA), John Raison (Australia), N.H. Ravindranath 

(India), Dieter Schoene (FAO).

MODIS REFLECTIVE AND ACTIVE FIRE DATA FOR BURN MAPPING 

IN COLOMBIA 

S. Merino-de-Miguel 1 

1 EUIT Forestal, Universidad Politécnica de Madrid, Madrid, Spain, silvia.merino@upm.es 

F. González-Alonso 2 , M. Huesca 2 , D. Armenteras 3 , S. Opazo 4 

2 Remote Sensing Laboratory, CIFOR - INIA, Madrid, Spain, alonso@inia.es 

3 Departamento de Biología, Universidad Nacional de Colombia, Bogotá, Colombia, 

darmenterasp@unal.edu.co 

4 Escuela de Ciencia y Tecnología Agropecuarias, Universidad de Magallanes, Punta Arenas, Chile 

sergio.opazo@umag.cl 

Abstract: Satellite-based strategies for burn mapping may rely on two types 

of remotely sensed data: reflective images and active fires detections. The 

present work uses both in a synergistic way. In particular, burn mapping is 

carried out using MCD43B4 (MODIS Terra+Aqua Nadir BRDF- Adjusted 

Reflectance 16-Day L3 Global 1km SIN Grid V005) post-fire datasets and 

MOD14A2 and MYD14A2 hotspot products (MODIS Terra and Aqua Thermal 

Anomalies 1km). Developed methodology was applied to Colombia in 2004, 

an area not covered by MODIS ground antennas. Burned area as resulted 

from this work was compared to two burn area products: L3JRC (Terrestrial 

Ecosystem Monitoring Global Burnt Area Product) and MCD45A1 (MODIS 

Terra+Aqua Burned Area Monthly Global 500m SIN Grid V005). Reached 

results showed that this method would be of great interest at regional to 

national scales, since it was proved to be quick, accurate and cost-effective. 


Wildfires are a major source of concern not only for environmental reasons, 

but also those including economy, society and human safety in many parts 

of the world. Wildfire effects at a local scale alter the ecosystem functionality 

and generate an important landscape impact (Pyne et al., 1996). In 

addition, wildfires release a significant amount of greenhouse gases, particulates 

and aerosol emissions into the atmosphere, which significantly 

increases the anthropogenic CO 2 emissions (Levine, 1991). 

The use of remote sensing data provides temporal and spatial coverage of 

biomass burning without costly and intense fieldwork. The resulting information 

is suitable for its integration into a GIS which allows the storage 

and processing of large volumes of spatial data. Remote sensing techniques 

have been used for a wide variety of tasks including burned area estimation 

(Chuvieco et al., 2008). At regional to global scales, the detection of 

227

228 


burned areas by means of satellite data has been traditionally carried out 

by the Advanced Very High Resolution Radiometer (AVHRR) (Kaufman et al., 

1990; Chuvieco et al., 2008). However, the MODIS sensor is opening a new 

era in the remote sensing of burned areas (Justice et al., 2002; Roy et al., 

2005). MODIS is a sensor housed on Terra and Aqua NASA satellites with 

more than 30 narrow bands at wavelengths from the visible to the thermal 

infrared and at variable spatial resolutions (250 to 1000m). 

Among the different methods for burn mapping by means of reflectance 

satellite data, the use of spectral indices is one of the most widespread. In 

addition to the latter, several studies have showed the utility of active fire 

detections. In this case, fire thermal energy, as measured by mid-infrared 

channels, is used to identify active fires and scar mapping is developed 

based for example in the total number of active fires (Pozo et al., 1997). 

Although the temporal and spatial patterns of biomass burning cannot be 

estimated reliably from active fire data (Roy et al., 2005), this difficulty 

can be solved through the combination of active fire information together 

with spectral indices, as proposed in this work. 

The work presented here assesses the estimation and mapping of burned 

areas in Colombia in 2004 using a method that integrates a spectral index 

with active fire data, both as derived from MODIS data. In a second step, 

the resulting 1km spatial resolution MODIS-based scar map was validated 

using two satellite-derived burn area products. 

2 - Study area and materials 

The approach is applied here to H10V08 tile in Colombia (figure 1) where 

hundreds to thousands of wildfires occurred during the winter months 

(January to March) each year. The area covered by the MODIS image is 1200 

by 1200km size. First analyses were carried out over a sub-scene of 300 by 

300km size (figure 1). 

Three types of data were used for this work: (i) MCD43B4-2004025 (MODIS 

Terra + Aqua Nadir BRDF-adjusted reflectance 16-day L3 Global 1km; 25 th 

January to 9 th February 2004), (ii) MOD14A2 and MYD14A2 (MODIS Terra 

and Aqua Thermal Anomalies 1km; 1 st January to 9 th February 2004) and 

(iii) ancillary maps and information. For validation purposes, two datasets 

were used: (i) L3JRC-2004 (Global, Daily, SPOT VGT-derived Burned Area 

Product) and (ii) MCD45A1-2005001 (MODIS Terra + Aqua Burned Area 

Monthly L3 Global 500m; 1 st to 31 st January 2004).

3 - Methodology and results 

MODIS reflective and active fire data for burn mapping in Colombia 229 

Figure 1 - Study area: Colombia (in green), H10V08 tile (in red) and test area (in black). 

All burned area estimation and mapping followed three steps: (i) 

Normalized Burn Ratio (NBR) calculation using the MODIS reflectance product, 

(ii) NBR threshold establishment using MODIS active fire information 

and (iii) validation. Image and data processing was carried out using ENVI 

4.2 and ArcGIS 9.2 software packages. 

The NBR (Key and Benson, 1999) is a spectral index that integrates nearinfrared 

(NIR) and shortwave-infrared (SWIR) bands, both of which register 

the strongest response, albeit in opposite ways, to burning (Lentile et al., 

2006; Roldán-Zamarrón et al., 2006). The NBR is estimated using the following 

equation: 

NBR = (ρ swir – ρ nir )/(ρ swir + ρ nir ) (1) 

NBR threshold value was established based on the best spatial correlation 

between burn area as derived from the NBR image itself and MODIS active 

fire locations. On the one hand, the NBR image was threshold using several 

values ranging from 0.00 to 0.10, what resulted in a set of binary images. 

On the other hand, active fire locations point format file were converted 

into a binary image of 1km pixel size. At this point it is worth pointing out 

that we only accounted for active fire locations occurred between the 1 st 

of January and the 9 th of February with a higher than 60 confidence value. 

The two sets of data were then compared using confusion matrices in order 

to find the threshold value that better correlate the two types of data 

(post-fire reflective and active fire data). The latter was tested using the

230 


Kappa coefficient. The highest Kappa coefficient was for a NBR threshold 

value of 0.06. 

Burn area map as resulted from this methodology was validated using two 

burn area products: L3JRC and MCD45A1. The overall accuracy of our product 

was 92.52 and 97.14%, respectively. Burned area within the test region 

was of 2065km 2 (our estimation), 2182km 2 (L3JRC) and 1568km 2 

(MCD45A1). Figure 2 shows an example. 

Figure 2 - Left - Our estimation delineated in red; center: L3JRC (black); right: MCD45A1 

(black). 

References 

Chuvieco, E., Englefield, P., Trishchenko, A.P., Luo, Y., 2008. Generation of 

long time series of burn area maps of the boreal forest from NOAA- 

AVHRR composite data. Remote Sens Environ 112, 2381-2396. 

Justice, C.O., Giglio, L., Korontzi, S., Owens, J., Morisette, J.T., Roy, D.P., 

Descloitres, J., Alleaume, S., Petitcolin, F., Kaufman, Y., 2002. The 

MODIS fire products. Remote Sens Environ 83, 244-262. 

Kaufman, Y., Tucker, C., Fung, I., 1990. Remote sensing of biomass burning 

in the tropics. J Geophys Res 95, 9927-9939. 

Key, C.H., Benson, N.C., 1999. The Normalized Burn Ratio (NBR): A Landsat 

TM Radiometric Measure of Burn severity. USDA (Bozeman, Mont). 

(Available at http://nrmsc.usgs.gov/research/ndbr.htm). 

Lentile, L.B., Holden, Z.A., Smith, A.M.S., Falkowski, M.J., Hudak, A.T., 

Morgan, P., Lewis, S.A., Gessler, P.E., Benson, N.C., 2006. Remote sensing 

techniques to assess active fire characteristics and post-fire effects. 

Int J Wildland Fire 15, 319-345. 

Levine, J.S., 1991. Introduction. In: Levine, J.S. (ed.), Global Biomass 

Burning: Atmospheric, Climatic and Biospheric Implications. MIT Press, 

Cambridge, USA. 

Pozo, D., Olmo, F.J., Alados Arboledas L., 1997. Fire detection and growth 

monitoring using a multi-temporal technique on AVHRR mid-infrared 

and thermal channels. Remote Sens Environ 60, 111-120. 

Pyne, S.J., Andrews, P.L., Laven, R.D., 1996. Introduction to Wildland Fire.

MODIS reflective and active fire data for burn mapping in Colombia 231 

Final Report MMF practices - 3015, Canada. John Wiley & Sons, New York, 

USA. 

Roldán-Zamarrón, A., Merino-de-Miguel, S., González-Alonso, F., García- 

Gigorro, S., Cuevas, J.M., 2006. Minas de Riotinto (South Spain) forest 

fire: Burned area assessment and fire severity mapping using Landsat 5- 

TM, Envisat-MERIS and Terra-MODIS postfire images. J Geophys Res 111 

(G04S11). 

Roy, D.P., Jin, Y., Lewis, P.E., Justice, C.O., 2005. Prototyping a global algorithm 

for systematic fire-affected area mapping using MODIS time series 

data. Remote Sens Environ 97, 137-162.

A NEW ALGORITHM FOR THE ATSR WORLD FIRE ATLAS 

S. Casadio 

SERCO c/o ESA/ESRIN, Frascati (RM), Italy 

Stefano.Casadio@esa.int 

O. Arino 

ESA/ESRIN, Frascati (RM), Italy 

Olivier.Arino@esa.int 

Abstract: A new fire detection algorithm ALGO3 has been developed and 

tested (based on basic threshold on the ATSR NIR 1.6 µm channel) in order 

to fully exploit the ATSR time series back to 1991 by using the ATSR-1 

measurements. The algorithm has been prototyped and tested under different 

conditions, and results are discussed with respect to the ALGO1 and 

ALGO2 products currently available from the ESA ATSR World Fire Atlas web 

sites. Pros and cons of using the results of ALGO3 instead of (or in synergy 

with) ALGO1 and ALGO2 are discussed and some specific cases are 

analysed in detail. 

1 - ATSR-WFA hot spot detection algorithms 

During his long history the ATSR World Fire Atlas demonstrated its usefulness 

in many research areas. The basic ALGO1 and ALGO2 approaches have 

been recognised to be extremely efficient and related results satisfactorily 

accurate (Arino, 2005 and 2007). ALGO1 and ALGO2 consist in the detection 

of ATSR 3.7 µm night-time brightness temperatures exceeding 312 and 

308 K respectively. The radiometric stability of the ATSR instrument series 

ensures the consistency of the detection capability for long time periods. 

The ATSR WFA products are freely available to public at the following site: 

http://dup.esrin.esa.int/ionia/wfa/ index.asp. 

The ATSR time coverage spans from 1991 to now but the above described 

approach was limited by the absence of the 3.7µm band for ATSR-1 due to 

an instrumental problem occurred in early 1992. In order to complement 

the actual algorithms and to take benefit from the ATSR-1 mission data 

(1991-1996) we developed a new retrieval scheme, called ALGO3, based on 

the analysis of ATSRs’ 1.6 µm band reflectance. The processing chain was 

completely revised, making the processing more efficient and expandable. 

The introduction of more complex fire detection algorithms is being considered 

for future developments. The analysis reported in this paper focus- 

233

234 


es on some of the preliminary results of the reprocessed ATSR-1, ATSR-2 and 

AATSR TOA products relative to the period August 1991 - February 2008, i.e. 

around 16 full years of data, and in particular on the gas flaring monitoring 

for a number of selected areas worldwide. 

One of the ALGO1 and ALGO2 drawbacks is the dependence of the 3.7 µm 

brightness temperature values from the background temperature, i.e. from 

the temperature of the non burning area. In fact, being the total radiance 

that reaches the satellite instrument the combination of the contributions 

from both burning and non burning fractions of the ground pixel, a fixed 

threshold for the hot spot detection cannot account for seasonal variations 

of the contributing background. ALGO3 overcomes this problem considering 

the 1.6 µm band behaviour during night-time observations. Alternative 

approaches, e.g. contextual algorithms, are under development. At NIR 

wavelengths the contribution the of background to night-time radiation is 

negligible (far below the noise level of the ATSR detectors) while it can be 

demonstrated that a useful signal is detected for active fires even for very 

small fire fractions. The ALGO3 method is based on the detection of 1.6 µm 

band reflectance values larger than a fixed threshold (= 0.1) which is twice 

the detector noise level. The following considerations hold: 

1. The solar proton and electron flux produces a quantity of spurious spots 

in the region interested by the South Atlantic Anomaly (SAA) (Cabrera, 

2005), and for high latitudes 

2. Outside SAA the spot density is very similar to that of usual global fire 

maps (not shown here) 

Observation 1 could lead to the conclusion that ALGO3 cannot be used for 

global fire monitoring because Southern America, a small portion of South 

Africa, and high latitudes are impacted by the solar flux. 

2 - Monitoring of Gas Flaring site using ALGO3 

Outside the SAA zone of influence ALGO3 proves to be more efficient than 

ALGO2 in detecting gas flares from oil-gas industrial sites. A characteristic 

of these sites is that their position is, in general, not changing in time, and 

this is crucial for their individuation. 

The gas flaring detection method adopted in this work consists in individuating 

small areas (roughly 1.2 km wide, slightly larger than the ATSR 

ground pixel size) in which hot spots are found more than twice a year. The 

time series of ALGO2/3 spots are recorded for each of these sites and results 

analysed in terms of flaring frequency, flame temperature and size, and 

background temperature. The North Sea area has been preliminarily selected 

for testing and the related gas flaring locations are shown in left panel 

of figure 1. The map of exploration lease in the same area is reported in the 

right panel for qualitative comparison. In figures 3 the red circles are centred 

on the flaring site and the circle’s radius is proportional to the number 

of detected spots for each site. The red circles are relative to ALGO3

A new algorithm for the ATSR World Fire Atlas 235 

while the blue circles refer to ALGO2 spots. As can be depicted from figure 

1 it was possible to detect more than 1700 gas flaring sites using ALGO3 

products, while the previous thermal signal analysis only allowed the individuation 

of 33 sites. The position of the detected flaring sites corresponds 

exactly to the exploration lease areas shown in the right panel of figure 1. 

The larger number of hot spots for a single station in this area is 215 for 

the entire time period considered. The number of satellite overpasses at 

these latitudes for the area in question (and the time frame considered) is 

roughly 2300. This implies that the (maximum) gas flare detection probability 

is around 10% for ALGO3, while for ALGO2 it is 1 %. If we consider 

all factors influencing the detection, e.g. cloudiness, real flaring frequency, 

and ground pixel sensing time (0.15 seconds per overpass), the ALGO3 

detection probability is surprisingly high. 

Figure 3 - North Sea Gas Flaring sites (left) and exploration lease map (right). 

Italy was considered for a second test. There the number of industries related 

to oil and gas refinery is known to be small. As a matter of fact, very 

few sites were individuated. The most active gas flaring sites are close to 

the cities of Cagliari and Taranto, the latter being characterised by a maximum 

of 278 spots for a single site. Considering that the satellite overpass 

frequency at these latitudes is slightly smaller than for the North Sea case, 

the (maximum) detection probability is above 12 %. It should be noted 

that with the adopted method it is possible to monitor the activity of volcanoes 

(Etna and Vulcano in this case). 

The other areas analysed in our work are The Gulf, Mexico Gulf, the West 

Coast of Africa, Mediterranean Africa, and Arctic Russia, with good results.

236 



A new fire detection algorithm has been developed and tested in the context 

of the ATSR WFA project. This new approach appears to extremely useful 

for global and long term gas flaring detection and monitoring, making 

use of the whole ATSR family data, spanning from 1991 to present. A preliminary 

analysis of ALGO3 data has been performed and results discussed 

in details. At present, the most important limitation of ALGO3 is represented 

by the large number of spurious spots in correspondence of the 

South Atlantic magnetic Anomaly (SAA). Nevertheless, outside the SAA 

influence ALGO3 has proven to be extremely efficient with respect to the 

well established thermal IR algorithms. In table 1 a summary of the numbers 

of detected flaring sites and maximum probability of detection pee site 

is reported for ALGO2 and ALGO3. These numbers clearly demonstrate the 

potential of this new algorithm for gas flaring detection and monitoring on 

global scale. In addition, the monitoring of volcanic activity is also possible 

from ALGO3 product analysis. 

Selected Area ALGO2 % ALGO3 % 

Italy 95 8 260 12 

North Sea 33 1 1732 10 

Arctic Russia 685 8 9898 20 

Mediterranean Africa 1073 14 7381 24 

Western Central Africa 1153 8 8193 20 

The Gulf 2028 13 6769 23 

Mexico Gulf 11 0.6 406 5 

TOTAL 5078 - 34639 - 

Table 1 - Number of Gas Flaring sites for different areas and detection algorithm. 

The complete reprocessing of the ATSR family radiances is on-going and the 

long time series of ALGO1/2/3 products will be made available to public 

after the necessary validation exercise will be successfully completed. 

References 

Arino, O., Plummer S. and Defrenne D., 2005. Fire Disturbance: The Ten 

Years Time Series of The ATSR World Fire Atlas, Proceedings MERIS-AATSR 

Symposium, ESA SP-597 

Arino, O., Plummer S. and Casadio S., 2007. Fire Disturbance: the Twelve 

years time series of the ATSR World Fire Atlas Proceedings of the ENVISAT 

Symposium 2007, ESA SP-636 

Cabrera, J., et al., 2005. Fluxes of energetic protons and electrons measured 

on board the Oersted satellite, Annales Geophysicae, 23, 2975-2982, 

2005, SRef-ID: 1432-0576/ag/2005-23-2975

ASSESSMENT OF POST FIRE VEGETATION RECOVERING IN PORTUGAL 

Abstract: The years of 2003 and 2005 were the two worst in the historical 

record of wildfires in Portugal. In 2003, the country was hit by the most 

devastating sequence of large fires, responsible for a total burnt area of 450 

000 ha, representing about 5% of the Portuguese mainland (Trigo et al., 

2006). However, in 2005, Portugal suffered again strong damages from forest 

fires that affected an area of 300 000 ha of forest and shrub. The years 

of 2003 and 2005 are of particular interest since they are both associated 

to extreme events, namely the heat wave episode of August 2003 and the 

severe drought of 2005. 

Based on monthly values, from 1999 to 2008, of the normalized difference 

vegetation index (NDVI), at the 1km x 1km spatial scale, as obtained from 

the VEGETATION instrument, a methodology is presented that allows identifying 

large burnt scars in Portugal and monitoring vegetation recovery 

throughout the pre and the post fire periods. 


C. Gouveia 

Escola Superior de Tecnologia de Setúbal (EST) 

Instituto Politécnico de Setúbal, Setúbal, Portugal, cmgouveia@fc.ul.pt 


C. DaCamara & R. Trigo 


cdcamara@fc.ul.pt, rmtrigo@fc.ul.pt 

Fire represents one of the main disturbances in Mediterranean ecosystems 

(Whelan, 1995), having a key role in the dynamics and structure of plant 

and animal communities (Gill et al., 1981), that are very inflammable and 

have low moisture contents. Plant communities in these ecosystems have 

high elasticity after fire because the species are able to regenerate by 

means of resprouting from fire resistant structures (Bond and Midgley, 

2001), or by germination of fire protected seeds stored in the soil or in the 

canopy (Lloret, 1998). Succession after fires usually begins with herbaceous 

species, which often recover quickly as a result of the ‘mineral flush’ 

after fires and their ability to colonize open spaces (Christensen, 1994). 

Plant communities frequently evaluate to shrub-dominated populations 

with isolated trees, becoming dense forests in the end stage (Röder et al., 

2008). Frequent fires may prevent plants from reaching sexual maturity 

237

238 


(Keeley, 1986), thus affecting the structure of plant communities and causing 

a decline of long-term ecosystem productivity. The effective manifestation 

of post-fire succession is determined by a combination of local physical 

conditions, climatic factors and pre-fire vegetation (e.g. Lloret and Vilà, 

2003; Whelan, 1995). 

In the above-mentioned context, a considerable number of fire recovery 

studies, based on remote sensing information, have been conducted in 

regions characterized by Mediterranean climates. In particular, spectral 

vegetation indices (Röder et al., 2008), namely NDVI, have proven to be 

especiallty useful to monitor plant regeneration after fire (Díaz-Delgado et 

al., 2003). 

2 - Results 

The response of vegetation was assessed based on fields of Normalised 

Difference Vegetation Index (NDVI) as derived from images acquired by the 

VEGETATION instrument on-board SPOT 4 and SPOT 5. Covering the period 

1999-2006, data consist of the S10 products of the VITO database 

(http://free.vgt.vito.be) at the 1km x 1km spatial scale. 

Large burnt scars in Portugal during the 2003 and 2005 fire seasons were 

identified using k-means clustering applied to 12 monthly NDVI anomalies, 

from September of the year to August of the year after the fire season. As 

shown in Figure 1, cluster analysis allows identifying two (four) clusters in 

2003 (2005). In both years, the centroid associated to systematic negative 

values of monthly anomalies throughout the year (black circles) is worth 

being pointed out since the corresponding spatial distributions (Figure 2) 

correspond to burnt scars (black pixels). 

Figures 3 and 4 present observed NDVI values over areas corresponding to 

four selected burnt scars, two of them respecting to the 2003 fire season 

and the other two to 2005. The locations of the scars are indicated by the 

red boxes in Figure 2. During the late spring preceding the occurrence of 

the fire episodes, namely in May 2003 and 2005, all regions present high 

levels of vegetation activity (left panels of Figures 3 and 4), with values of 

NDVI around 0.7. Nine months after the fire episodes, in May 2004 and 

2006 (middle panels), values of NDVI drop to around 0.35 on average. It 

may be noted, however, that the spatial distributions of NDVI before the 

fire season of 2003 present higher values than the ones before the fire season 

of 2005, in particular over the burnt scar located in the south of 

Portugal (Figure 3, bottom panel), a feature that relates with the severe 

drought that hit the country during 2004/2005. A similar behaviour may be 

observed with the NDVI values after the fire events, which provide an indication 

of fire sevirity. Again, NDVI after the fire season of 2003 tends to 

be larger than NDVI after the fire season of 2005. Finally, relative differences 

between NDVI of May before and after the fire events (left panels of 

Figures 3 and 4) present higher values in the case of the burnt scars locat-

Assessment of post fire vegetation recovering in Portugal 239 

ed in the center of Portugal both fire seasons. The different behavior may 

be related with several factors, namely fire severity, fire recurrence and land 

cover types. For instance, according to Corine Land Cover 2000, the two 

burn scars in 2003 correspond to forested areas (broadleaved in the south 

and coniferous in the center) whereas, in 2005, the selected areas are 

mainly degraded lands. 

References 

Bond, W.J., Midgley, J.J., 2001. Ecology of sprouting in woody plants: the 

persistence niche. Trends Ecol. Evol. 16: 45-51. 

Christensen, N.L., 1994. Fire and soil in Mediterranean shrublands. In J. M. 

Moreno & W. C. Oechel (Eds.), The role of fire in Mediterranean-type 

ecosystems (pp. 79-95). New York: Springer. 

Díaz-Delgado, R., Lloret, F., Pons, X., 2003. Influence of fire severity on 

plant regeneration by means of remote sensing imagery, Int. J. Rem. 

Sensing 24 (8): 1751-1763. 

Gill, A.M., Groves, R.H., Noble, I.R., 1981. Fire and the Australian biota. 

Australian Academy of Science, Canberra, AU. 

Keeley, J.E., 1986. Resilience of Mediterranean shrub communities to fire. In 

Dell, B., Hopkins, A.J.M., Lamont B.B., (Eds.), Resilience in 

Mediterranean type ecosystems. Kluwer Academic Publishers, 95-112. 

Lloret, F. 1998. Fire, canopy cover and seedling dynamics in Mediterranean 

shrubland of northeastern Spain. J. Veg Sci. 9: 417-430. 

Lloret, F., Vilà, M., 2003. Diversity patterns of plant functional types in 

relation to fire regime and previous land use in Mediterranean woodlands. 

J. Veg. Sci. 14: 387-398. 

Röder, A., Hill, J., Duguy, B., Alloza, J.A., Vallejo, R., 2008. Using long time 

series of Landsat data to monitor fire events and post-fire dynamics and 

identify driving factors. A case study in the Ayora region. Rem. Sens. of 

Environment, 112 (1): 259-273. 

Trigo R.M., Pereira, J.M.C., Pereira, M.G., Mota B., Calado, M.T., Da Camara 

C.C., Santo, F.E., 2006. Atmospheric conditions associated with the 

exceptional fire season of 2003 in Portugal. Int. J. of Climatology 26 

(13): 1741-1757. 

Whelan, R.J., 1995. The ecology of fire. Cambridge University Press, 

Cambridge, UK.

240 


Figure 1 - Monthly NDVI anomalies of the centroids of identified clusters, respectively associated 

to burnt pixels (black circles) and non burnt pixels (grey circles) for 2003 (left panel) 

and 2005 (right panel). 

Figure 2 - Annual burned areas in Continental Portugal as identified by means of cluster analysis 

respecting to the fire season of 2003 (right panel) and 2005(right panel). Red boxes represent 

the selected burnt scars.

Assessment of post fire vegetation recovering in Portugal 241 

May 2003 May 2004 

NDVI Difference 

May 2003 May 2004 


0.3 0.4 0.5 0.6 0.7 0 0.5 

Figure 3 - Temporal evolution of the spatial distribution of NDVI and relative differences 

between the two considered months for the selected burnt scars of the 2003 fire season. 

May 2005 May 2005 


May 2005 May 2005 


0.3 0.4 0.5 0.6 0.7 0 0.5 

Figure 4 - As in Figure 3, but respecting to the fire season of 2005.

SOME NOTES ON SPECTRAL PROPERTIES OF BURNT SURFACES 

AT SUB-PIXEL LEVEL USING MULTI-SOURCE SATELLITE DATA 

Abstract: The aim of our study is to explore the spectral properties of 

burned surfaces at sub-pixel level using multi-source resolution satellite 

data. For that purpose, a study case was established in one very destructive 

wildfire occurred in Parnitha, Greece on July 2007. Satellite data at 

multiple spectral and spatial resolutions were acquired shortly after the fire 

from MODIS, LANDSAT, ASTER, and IKONOS satellite sensors. The very high 

resolution imagery of IKONOS satellite was served as the basis for extracting 

the percent of cover of burnt areas, bare land and vegetation by applying 

the maximum likelihood classification algorithm. Furthermore, vegetation 

indices of the post-fire satellite images were computed. Then the percent 

of cover for each type was correlated to surface reflectance values and 

vegetation indices for all satellite images, and linear regression models 

were fitted to characterize those relationships. 


N. Koutsias & M. Pleniou 

Department of Environmental and Natural Resources Management, 

University of Ioannina, G. Seferi 2, Agrinio, Greece, 

nkoutsia@cc.uoi.gr; mpleniou@cc.uoi.gr 

The application of space-borne sensor data for forest fire mapping is an old 

and currently an active research topic in remote sensing studies. Many characteristic 

examples of satellite remote sensing studies on burnt land mapping 

and monitoring can be found in the literature, while some of them 

focus on spectral properties and mapping of burned surfaces at subpixel 

level. It is well justified that there are spectral similarities between burnt 

surfaces and other land cover categories, such as water bodies, urban areas, 

shadows, and mixed land-water and water-vegetation pixels, that result in 

spectral confusion. Some of the problems in discriminating and mapping 

the burned surfaces are related to non-homogenous pixels of different fractions 

of land cover types (Chuvieco and Congalton, 1988; Koutsias and 

Karteris, 2000; Koutsias, Mallinis et al., 2009). 

243

244 



2.1 - Study area and satellite data 

In the summer of 2007, the year with most destructive fires in recent history 

of Greece, a large fire occurred in Parnitha, Greece. This was the study 

area for discussing the spectral properties of burnt surfaces at sub-pixel 

level using multi-source satellite data. For this purpose, a dataset consisting 

by an IKONOS, ASTER, LANDSAT and MODIS imagery was established 

after the fire and constituted the basic source of information. The spatial 

resolution ranges from 1 to 1000 meters, while the spectral resolution of 

the sensors covers the visible, near and mid-infrared, as well as the thermal-infrared 

part of the electromagnetic spectrum. 

a b 

c d 

Figure 1 - A dataset consisting by an IKONOS (a), ASTER (b), LANDSAT (c) and MODIS (d) 

imagery was established after the fire and constituted the basic source of information. 

2.2 - Methods 

Classical image processing algorithms were applied to correct geometrically, 

radiometrically and atmospherically the available satellite images. The 

digital radiometric numbers were converted first to radiance at sensor and 

then to surface reflectance to achieve the necessary compatibility for comparing 

them. The satellite data were orthorectified using a digital elevation 

model created by the 20 meters contour lines extracted from maps of 

1:5000. 

The very high resolution imagery of IKONOS satellite was served as the basis 

for extracting the percent of cover of burnt areas, bare land and vegetation 

by applying the maximum likelihood classification algorithm. Furthermore,

Some notes on spectral properties of burnt surfaces at sub-pixel level using multi-source satellite data 245 

vegetation indices of the post-fire satellite images were computed. Then 

the percent of cover for each type was correlated to surface reflectance values 

and vegetation indices for all satellite images, and regression models 

were fitted to characterize those relationships. These relationships were 

implemented by overlaying a fishnet of 90 meters resolution over the LAND- 

SAT and ASTER imagery and 250 meters for MODIS. 


In the literature, the spectral channel TM4 has been proved to be the most 

sensitive regarding alterations in spectral response of the burned pixels. 

The second spectral channel with the highest contribution to burned area 

discrimination has been reported that it is TM7. A strong decrease in 

reflectance of burned areas occurs in the near-infrared part of the electromagnetic 

spectrum as a result of deterioration of the leaf structure of the 

vegetation layer, which reflects large amounts of the incident solar radiation. 

In addition, a strong increase in reflectance of the burned areas 

occurs in the mid-infrared region as a result of the decrease of the vegetation 

moisture content. The replacement of the vegetation layer with charcoal 

reduces the water content, which absorbs the incident radiation in this 

spectral region, and consequently the burned areas present higher midinfrared 

reflectance values than those of healthy vegetation. 

In the graphs below it is clear that the relationship between percent of 

burned and percent of vegetation with NIR and MIR ground reflectance 

value as well as with NDVI values at grid cells of 90 and 250 meters resolution 

of LANDSAT, ASTER and MODIS data respectively is described by a linear 

model. 

Figure 2 - Relationship between percent of burned and percent of vegetation with NIR and MIR 

ground reflectance and NDVI at grid cells of 90 meters resolution of LANDSAT data.

246 



ground reflectance and NDVI at grid cells of 90 meters resolution of ASTER data. 


ground reflectance and NDVI at grid cells of 250 meters resolution of MODIS data. 

References 

Chuvieco E., Congalton R.G., 1988. Mapping and inventory of forest fires 

from digital processing of TM data. Geocarto International 4, 41-53. 

Koutsias N., Karteris M., 2000. Burned area mapping using logistic regression 

modeling of a single post-fire Landsat-5 Thematic Mapper image. 

International Journal of Remote Sensing 21, 673-687. 

Koutsias N., Mallinis G., Karteris M., 2009. A forward/backward principal 

component analysis of Landsat-7 ETM+ data to enhance the spectral signal 

of burnt surfaces. ISPRS Journal of Photogrammetry & Remote 

Sensing 64, 37-46.

MONITORING POST-FIRE VEGETATION REGENERATION OF THE 2003 

BURNED AREAS IN PORTUGAL USING A TIME-SERIES OF MODIS 

ENHANCED VEGETATION INDEX 

P. Malico 

Department of Forestry, School of Agronomy, Technical University of Lisbon, 

Lisboa, Portugal, patmalico@isa.utl.pt 

J. Kucera 1 , J. San-Miguel Ayanz 1 , B. Mota 2 , J.M.C. Pereira 2 

1 Joint Research Centre, Institute for Environment and Sustainability, Land Management and Natural 

Hazards Unit, Ispra (VA), Italy, jan.kucera@jrc.it; jesus.san-miguel@jrc.it 

2 Department of Forestry, School of Agronomy, Technical University of Lisbon, 

Lisboa, Portugal, bmota@isa.utl.pt; jmcpereira@isa.utl.pt 

Abstract: Our objective is to monitor vegetation recovery after the large 

fires of 2003 in Portugal. We use a time-series of MODIS Terra Enhanced 

Vegetation Index (EVI) data, at 250 m spatial resolution and with 16-day 

image composites (MOD13A1 product). These data were screened with the 

Time Series Generator (TiSeG) software, to detect and replace bad or missing 

data. The analysis begins in early 2000, to characterize pre-fire vegetation 

dynamics, extends until late 2008, and relies on fire perimeters 

mapped from Landsat imagery at 30m spatial resolution to locate the 2003 

fires. Three sample fires were selected for preliminary visual analysis of EVI 

time series within and outside the fire perimeters, and observed differences 

are discussed as a function of pre-fire land cover. 


The Mediterranean climate is characterized by the coincidence of the hot 

season with the dry season and rainy winters (Efe, Cravins et al., 2008). 

This particular aspect of weather conditions determines that is one of the 

world’s major fire-prone biomes where fire controls structure and vegetation 

dynamics (Naveh, 1975; Bond and Keeley, 2005), since plant communities 

have high resilience to fire and many regenerate by resprouting from fire 

resistant structures (Arnan, Rodrigo et al., 2007). The objectives of the 

present study are: to monitor vegetation recovery in areas burned in 

Portugal during the extreme fire season of 2003; to quantify the rate of 

vegetation recovery at spatial scales ranging from pixel to region; to identify 

spatial environmental correlates of vegetation recovery rates, and 

quantify their relative importance as explanatory variables. Wildfires have 

broad impacts, affecting the production of environmental goods and services 

through impacts on biodiversity, soils, water resources and air quality. 

Detailed spatio-temporal monitoring of post-fire vegetation recovery is 

important to help forecast negative environmental impacts, such as land- 

247

248 


slides and floods, and to assist burned area rehabilitation activities. 

Extended post-fire monitoring of vegetation regrowth is also useful for 

characterizing fuel hazard dynamics, an important contributor to wildfire 

risk assessment. Previous studies have addressed various aspects of satellite 

monitoring of post-fire vegetation regrowth in Mediterranean-type 

ecosystems. Viedma et al. (1997) analyzed regrowth pathways and recovery 

rates of different plant communities in the Mediterranean coast of Spain at 

five different dates, between 1984 and 1994, using Landsat Thematic 

Mapper imagery and the normalized difference vegetation index (NDVI). 

Solans Vila and Barbosa (2009) compared different methods to obtain 

quantitative estimates of vegetation recovery in an area burnt in Liguria 

(Italy) in 1998. Bisson et al. (2008) proposed the Vegetation Resilience 

after Fire (VRAF) index, to map the ability of vegetation to recover after 

fire, based on composition of the vegetation community, soil type and 

geology, and terrain slope and aspect. Díaz-Delgado et al. (2002) used the 

NDVI from Landsat imagery to monitor vegetation recovery after successive 

fires in Catalonia (NE Spain) between 1975 and 1993. 


Our study covers the entire area of Continental Portugal (Fig. 1), which has 

the highest fire incidence in Southern Europe, with a mean annual burned 

area of 110,000 ha, corresponding to 1.2% of the study area. The fire season 

of 2003 is the worst on record, with 430,000 ha burned, about 4 times 

the mean annual value and one and a half times larger than the previous 

maximum, recorded in 1985. Between 30 July and 3 August 2003, 80 fires 

burned more than 220,000 hectares, coinciding with an extreme heat wave 

over Europe estimated to have been a 500-yr return interval event. We used 

MODIS Enhanced Vegetation Index (EVI) with 250 m spatial resolution, 16 

days data from NASA MODIS/Terra product named “Vegetation Indices 16- 

Day L3 Global 250m” (MOD13Q1), from 2000 to 2008, to characterize vegetation 

dynamics in the burned areas, starting approximately three and a 

half years prior to fire dates, and extending for over five years after the 

fires. The starting data of February 2000 allows the characterization of 

undisturbed vegetation dynamics in the fire-affected areas. The post-fire 

length of the time-series ought to allow for substantial vegetation recovery, 

up to fuel loads capable of sustaining new fires. The images were 

processed with Modis Reprojection Tool (MRT)in order to mosaic and clip 

the study area. Then, TiSeG (Colditz, Conrad et al., 2008) software was used 

to do a quality assessment to screen out low quality data and to calculate 

time-interpolated values for filling the data gaps created. Interpolation was 

performed on a single pixel basis, using cubic splines. For each of three 

sample fires, located respectively in NE, Central and SW Portugal, buffer 

areas were defined to provide control data for monitoring fire-unaffected 

vegetation dynamics. In order to extract the EVI values from each pixel a

Monitoring post-fire vegetation regeneration of the 2003 burned areas in Portugal using a time-series of MODIS enhanced vegetation index 249 

sample grid was created. Inside the burned areas the grid size is 500 m. The 

values inside the burned areas will provide information related to pre and 

post-fire situation, but it’s also important to have some control plots outside 

the burned areas in order to understand the vegetation dynamic without 

fire. Therefore, a 1000 m sized sample grid was created outside the 

burned areas, but only in the same Corine classes has the ones that exist 

inside. Next we created time-series for the fire and control area at each of 

the three sites using an IDL script to extract the EVI values and plotted EVI 

means and standard deviations. 

Fig. 1 - Portugal, 2003 fire perimeters over 500 ha and respective MODIS-EVI images.

250 


3 - Preliminary results 

The time-series of EVI values for Guarda (2797 ha), Santarém (19466 ha) 

and Faro (25471 ha) fires and respective control areas, are shown in Fig.1. 

All the three fires occurred during the first two weeks of August 2003, as 

revealed by the sharp drop in EVI values for the burned areas (fig. 2 - black 

line). Prior to the fires, the EVI time series for fire and control areas are 

very similar, especially for the Santarém and Guarda fires. At the Faro site, 

pre-fire EVI values of the control area appear consistently lower and more 

strongly seasonal than those within the fire perimeter. The fire induced EVI 

decrease is larger at the Santarém and Faro fires than at Guarda fire. There 

are also clear differences in post-fire EVI response dynamics. At Santarém, 

EVI values at the burned area take over two years to reach values similar 

to those of the control area, while at Faro the recovery is faster (about 

16months). The post-fire vegetation response displays a distinct pattern of 

the Guarda site. Recovery is very fast, such that five months after the fire, 

EVI values are higher within the fire perimeter then in the unburned control 

area. By October 2004, i.e., 14 months after the fire, the fire affected 

area and control EVI time series converge to very similar values. 

Fig. 2 – EVI time series between February 2000 and December 2008 for Guarda (a), Santarém 

(b) and Faro (c) fires.

Monitoring post-fire vegetation regeneration of the 2003 burned areas in Portugal using a time-series of MODIS enhanced vegetation index 251 

4 - To do next 

After this previous analysis of the data we intent to de-seasonalise the data 

using empirical mode decomposition (), an approach which does not rely 

on stationary assumptions. Then, using Spatial data on land cover, topography, 

previous-fire history, soil maps and climate data, we want to adjust 

a model to the trend tendency of vegetation recovery function of these 

variables. 

References 

Arnan, X., Rodrigo, A., et al., 2007. “Post-fire regeneration of Mediterranean 

plant communities at a regional scale is dependent on vegetation 

type and dryness.” Journal of Vegetation Science 18(1): 111-122. 

Bisson, M., Fornaciai, A., Coli, A., Mazzarini F. and Pareschi M.T., 2008. The 

Vegetation Resilience After Fire (VRAF) index: Development, implementation 

and an illustration from central Italy. International Journal of 

Applied Earth Observation and Geoinformation,10: 312-329. 

Bond, W.J. and Keeley, J.E., 2005. “Fire as a global ‘herbivore’: the ecology 

and evolution of flammable ecosystems.” Trends in Ecology and 

Evolution 20(7): 387-394. 

Díaz-Delgado, R., Lloret, F., Pons, X. and Terradas, J., 2002. Satellite evidence 

of decreasing resilience in Mediterranean plant communities after 

recurrent wildfires. Ecology, 83: 2293-2303. 

Colditz, R.R., Conrad, C., et al., 2008. TiSeG: A Flexible Software Tool for 

Time-Series Generation of MODIS Data Utilizing the Quality Assessment 

Science Data Set, IEEE Transactions on Geoscience and Remote Sensing. 

46: 3296-3308. 

Efe, R., Cravins, G., et al., Eds., 2008. Natural Environment and Culture in 

the Mediterranean Region. Newcastle, Cambridge Scholars Publishing. 

García-Haro, F.J., Gilabert, M.A., Meliá, J., 2001. Monitoring fire-affected 

areas using Thematic Mapper data. International Journal of Remote 

Sensing, 22: 533-549. 

Naveh, Z., 1975. “The evolutionary significance of fire in the mediterranean 

region“ Vegetatio 21: 199-208. 

Solans Vila, J.P. and Barbosa P., 2009. Post-fire vegetation regrowth detection 

in the Deiva Marina region (Liguria-Italy) using Landsat TM and 

ETM+ data. Ecological Modelling, in press. 

Viedma, O., Meliá, J., Segarra, D., Garcia-Haro J., 1997. Modeling rates of 

ecosystem recovery after fires by using Landsat TM data. Remote 

Sensing of Environment, 61: 383-398.

PROPERTIES OF X- AND C- BAND REPEAT-PASS INTERFEROMETRIC SAR 

COHERENCE IN MEDITERRANEAN PINE FORESTS AFFECTED BY FIRES 

M. Tanase 

Dpt. of Geography, University of Zaragoza, Zaragoza, Spain, mihai.tanase@tma.ro 

M. Santoro & U. Wegmüller 

Gamma Remote Sensing AG, Gümligen, Switzerland, santoro@gamma-rs.ch 

J. de la Riva & F. Pérez-Cabello 

University of Zaragoza, Spain, delariva@unizar.es; fcabello@unizar.es 

Abstract: Synthetic Aperture Radar (SAR) data has been investigated to 

determine the relationship between burn severity and interferometric 

coherence at two fires. Determination coefficients were used to relate 

coherence to optical sensor based estimates of burn severity expressed by 

differenced Normalized Burn Ratio (dNBR) index. The analysis for a given 

range of local incidence angle showed that coherence increases with the 

increase of burn severity for all datasets. This study indicates that co-polarized 

coherence could be used for burn severity evaluation in the 

Mediterranean environments. 


Repeat pass coherence is primarily related to the temporal stability of the 

scatterers. The higher stability of burnt areas with respect to unburned forest 

reflects in different coherence levels, coherence increasing significantly 

after fires (Liew et al., 1999). The overall goal of the study was to evaluate 

to what extent coherence could be used for burn severity assessment 

taking into consideration the moderate topography of typical Mediterranean 

pine forests. The objectives of the study were (1) to evaluate interferometric 

coherence response in relation to burn severity for X- and C- band SARs, 

and (2) to infer the prediction power of interferometric coherence for burn 

severity estimation. 

2 - Study area and datasets 

The Zuera study area is located in central Ebro Valley, Spain. The climate is 

Mediterranean with continental characteristics and annual average precipitations 

around 500 mm. Most of the area stretches on moderate topography 

(4° to 25° slopes), elevations ranging from 500 m to 750 m. The for- 

253

254 


est, extending on about 14.000 ha of homogeneous, even aged Pinus 

halepensis L., has been affected by recurrent fires during the last decades. 

Two fires -Zuera95 and Zuera08- were studied. 

The SAR dataset consisted of VV polarized images acquired by the European 

Remote Sensing 1 and 2 (ERS-1/2), HH polarized images acquired by 

Environment Satellite Advanced SAR (ASAR) and dual-polarized images (HH 

& HV) acquired by TerraSAR-X (TSX). SAR pre-processing step consisted in 

improvement of orbital state vectors and co-registration of the scenes from 

the same sensor and track. The coherence was estimated from the differential 

interferogram using an estimator with an adaptive window size 

(Wegmüller et al., 1998). ERS and ASAR data were processed using 1*5 

looks in range and azimuth while for TSX inteferograms 8*5 looks were 

used. Post processing consisted of image geocoding of the coherence products 

to UTM 30 coordinate system. The higher resolution TSX scenes were 

geocoded to 10 m pixel spacing. For the C-band sensors the scenes were 

geocoded to 25 m. Two pairs of Landat TM images were also acquired to 

derive optical based estimates of burn severity. 

3 - Methods 

Evaluation of the coherence response over burned areas was carried out 

using descriptive statistics. Determination coefficients (R 2 ) were used to 

evaluate the potential of interferometric coherence for burn severity estimation. 

To minimize the effects of topography on backscattered energy statistic 

analyses were carried out by 5º groups of local incidence angle. Due 

to the limited number of reference plots where burn severity was assessed 

in situ (Tanase et al., 2009a) the study was carried out using a set of pseudo-plots 

based on optical estimates of burn severity. The pseudo-plots were 

generated by averaging pixels of similar dNBR for the same local incidence 

angle rounded to unity. The consistency between field plots assed using 

Composition Burn Index (CBI), and remotely sensed (dNBR) estimates of 

burn severity was previously evaluated for Zuera08 area (Tanase et al., 

2009a). For the lower spatial resolution imagery (ERS and ASAR) pseudoplots 

including 9 pixels were selected while for the higher resolution 

TerraSAR-X data pseudo-plots including 25 pixels seemed appropriate 

(Tanase et al., 2009c). 

4 - Results 

Scatter plot of coherence as a function of burn severity is presented in 

Figure 1 for flat areas. The plot on the left shows 1-day ERS coherence (C- 

VV) for the Zuera95 site and the 35-days ENVISAT ASAR coherence (C-HH) 

for the Zuera08 site. The plot on the right in Figure 1 shows the coherence 

of a TerraSAR-X image pair for HH- and HV-polarization for Zuera08 burn.

Properties of x- and c- band repeat-pass interferometric sar coherence in mediterranean pine forests affected by fires 255 

Average coherence for water and bare surfaces has been included to provide 

a comparison with other land cover classes. 

Figure 1 - Scatter plots of coherence as a function of burn severity in flat areas. 

Figure 2 - Coherence vs. local incidence angle for high burn severity. 

In Figure 2 mean values of coherence are reported with respect to the local 

incidence angle for the pseudo-plots with high burn severity levels (i.e. 

dNBR ≥ 600). Highly burned plots were averaged together by five degree 

intervals of local incidence angle. To infer the utility of intererometric 

coherence for burn severity estimation, liner regression determination coefficients 

(R 2 ) expressing the proportion of burn severity variance predicted 

by coherence are presented in Table 1 for each sensor.

256 


sensor TerraSAR-X ERS ASAR 

dataset 16 Nov 08/19 Dec 2008 28/29 Mar 96 08 Jan/12 Feb 09 

polarization HH HV HH&HV VV HH 

local inc. angle R 2 R 2 R 2 Std. err. R 2 Std. err. R 2 Std. err. 

all pseudo-plots 0.631 0.303 0.692 197.7 0.651 153.5 0.435 269.7 

pseudo-plots grouped by local incidence angle 

10-15 - - - - 0.655 149.9 0.864 132.0 

16-20 - - - - 0.662 152.3 0.819 149.3 

21-25 0.936 0.768 0.936 101.5 0.746 132.5 0.732 173.4 

26-30 0.861 0.551 0.866 126.1 0.741 130.9 0.712 176.7 

31-35 0.678 0.368 0.717 178.8 0.735 135.8 0.648 203.1 

36-40 0.685 0.380 0.702 175.6 0.743 125.3 0.711 200.3 

41-45 0.652 0.381 0.653 197.2 - - - - 

46-50 0.732 0.499 0.732 181.0 - - - - 

51-55 0.727 0.391 0.748 164.9 - - - - 

56-60 0.854 0.594 0.854 140.0 - - - - 

66-65 0.862 0.478 0.865 129.2 - - - - 

Table 1 - Determination coefficients explaining the agreement between burn severity and 

interferometric coherence. 

5 - Discussion 

Figure 1 shows that coherence increases with burn severity at all frequencies 

and polarizations. However, noticeable differences in absolute values 

appear not only between data acquired at different SAR frequencies but 

also for image pairs acquired with the same system at different time interval. 

Coherence increased by about 0.25-0.3 units from unburned to highly 

burnt forests. Regression analysis showed moderate relations between dNBR 

values and interferometric coherence of co-polarized beams when local 

incidence angle was not accounted for. Determination coefficients were 

highest for X-band and one-day C-band data (Table 1). The use of the coand 

cross-polarized X-band coherence in the regression model did not 

improve significantly the results. Analysis of dNBR as a function of coherence 

after grouping by 5 0 local incidence angle intervals showed increased 

values of the determination coefficients for all frequencies and polarizations. 

Coherence was affected by topography, decreasing with the increase 

of the local incidence angle (Figure 2). 


Interferometric data provided important advantages over the use of 

backscattering coefficient for burn severity evaluation: i) lower influence

Properties of x- and c- band repeat-pass interferometric sar coherence in mediterranean pine forests affected by fires 257 

of topography on the coherence estimates ii) higher determination coefficients 

obtained for the single-polarized medium spatial resolution ERS data 

(Tanase et al., 2009b) and iii) no need for multi-polarized datasets. 


The work has been financed by the Spanish Ministry of Science and 

Education and the European Social Fund. ASAR and TerraSAR-X data were 

provided by the ESA and DLR. 

References 

Liew, S.C., Kwoh, L.K., Padmanabhan, K., Lim, O.K., Lim, H., 1999. 

Delineating Land/Forest Fire Burnt Scars with ERS Interferometric 

Synthetic Aperture Radar. Geophysical Research Letters, 26: 2409-2412. 

Tanase, M., Pérez-Cabello, F., de la Riva, J., Santoro, M., 2009a. TerraSAR- 

X data for burn severity evaluation in Mediterranean forests on sloped 

terrain. IEEE Trans. Geosci. Remote Sensing, accepted for publication. 

Tanase, M., de la Riva, J., Pérez-Cabello, F., Santoro, M., 2009b. Backscatter 

properties of X-, C- and L- band SAR in Mediterranean pine forests 

affected by fires. In, Advances in RS and GIS Applications in Forest Fire 

Management Towards an Operational Use of Remote Sensing in Forest Fire 

management. Matera, Italy: EARSeL. 

Tanase, M., Santoro, M., de la Riva, J., Pérez-Cabello, F., 2009c. Backscatter 

properties of multitemporal TerraSAR-X data and the effects of influencing 

factors on burn severity evaluation, in a Meditteranean pine forest. 

In, IEEE International Geosicence and Remote Sensing Symposium, 

IGARSS09. Cape Town, South Africa: IEEE. 

Wegmüller, U., Werner, C., Strozzi, T., 1998. SAR interferometric and differential 

interferometric processing. In, IGARSS`98, Seattle, IEEE: 1106- 

1108

Abstract: SAR data from a test site in Spain has been investigated to determine 

the relationship between forest burn severity and backscatter coefficient. 

The analysis showed that at HH and VV polarizations backscatter 

increases with burn severity. For cross-polarized beams (HV) backscatter 

decreased with burn severity. Backscatter coefficient showed potential for 

burn severity estimation in the Mediterranean environment when information 

from co- and cross-polarized beams is used jointly and the local incidence 

angle is accounted for. 


BACKSCATTER PROPERTIES OF X- AND C-BAND SAR IN 

A MEDITERRANEAN PINE FOREST AFFECTED BY FIRE 

M. Tanase 

Dpt. of Geography, University of Zaragoza, Zaragoza, Spain, mihai.tanase@tma.ro 

J. de la Riva & F. Pérez-Cabello 

University of Zaragoza, Spain, delariva@unizar.es; fcabello@unizar.es 

M. Santoro 

Gamma Remote Sensing AG, Gümligen, Switzerland, santoro@gamma-rs.ch 

The high number of forest fires occurring every year constitutes a major 

degradation factor of Mediterranean ecosystems. One of the most widely 

employed optical-wavelength based spectral indexes for burn severity 

assessment is the differenced Normalized Burn Ratio (dNBR) (Key and 

Benson, 2006). An alternative method for burned severity evaluation is the 

use of synthetic aperture radar (SAR) data which provides information related 

to forest structure, removal of leaves and branches by fire directly influencing 

the backscattered energy. Preliminary studies confirmed the potential 

of SAR data for remote estimation of burn severity (Tanase et al., 

2009). This work is focused on the evaluation of backscatter response from 

burned areas for X- and C- band SARs. The objectives were (1) to evaluate 

and compare backscatter response in relation to burn severity for X- and Cband 

data, and (2) to infer the prediction power of X- and C-band backscatter 

for burn severity estimation. 

259

260 


2 - Study area and datasets 

Zuera study area is characterized by a mediterranean climate with continental 

characteristics. The topography is moderate to steep, elevation ranging 

from 500 m to 700 m above sea level on slopes up to 25°. The dominant 

land cover is forest of Pinus halepensis L. interspersed with agricultural 

fields. The fire was ignited on 6 th of June 2008 after a traffic accident 

and burned around 2200 ha. The SAR dataset consisted of single (VV) and 

dual polarization (HH&HV) images acquired by the Environmental Satellite 

(ENVISAT) Advanced SAR (ASAR) and TerraSAR-X (TSX) sensors. 

3 - Methods 

Pre- and post-fire Landsat TM images were used to estimate burn severity 

by means of dNBR. SAR data were absolutely calibrated, multi-looked and 

geocoded to 25 m pixel using UTM projection. To correct the radiometric 

distortions still present in the geocoded images topographic normalization 

for the varying incidence angle from near to far range and the effective 

pixel area was applied (Ulander, 1996). 

Remotely sensed dNBR index was used as indicator of burn severity levels. 

To minimize the effects of topography on the backscatter the analyses were 

carried out by 5º groups of local incidence angle which provided the optimum 

grouping interval (Tanase et al., 2009). 

The study was carried out using a set of pseudo-plots generated based on 

optical data. The similar trends and the high correlation between dNBR and 

field estimates of burn severity suggested that the use of pseudo-plots 

would result in consistent relations (Tanase et al., 2009). Pseudo-plots were 

generated by averaging pixels of similar dNBRs and identical local incidence 

angles (rounded to unity), by reclassifying the continuous dNBR interval (- 

150 to 1050) into twenty-four classes. To avoid uncertainty related to size 

only pseudo-plots containing nine pixels were allowed. From all generated 

pseudo-plots 900 were randomly selected for the analyses. 

4 - Results 

4.1 - Backscatter properties of burned areas 

Scatter plot of average backscatter with respect to burn severity (dNBR) is 

presented in Figure 1 for pseudo-plots located on flat terrain. Figure 2 presents 

the average backscatter levels for X-band (HH- and HV-polarization). 

To provide a reference, the average backscatter of bare soils outside the fire 

perimeter was computed for each polarization (horizontal lines).

Backscatter properties of x- and c-band sar in a mediterranean pine forest affected by fire 261 

Fig. 1 - Scatter plot of C-band backscatter as 

a function of burn severity ASAR HH, HV and 

VV data from flat areas (21°-25°). 

In Figure 3 mean values of SAR backscatter are reported with respect to the 

local incidence angle for pseudo-plots of highly burned forest (i.e. dNBR 

≥600). Keeping burn severity constant (i.e. less variability due to severity 

levels) illustrated the influence of the local incidence angle on the 

backscatter coefficient for a given frequency and polarization. 

Fig. 2 - Scatter plots of backscatter as a 

function of burn severity in flat areas (Xband). 

4.2 - Backscatter for burn severity evaluation 

Fig. 3 - Backscatter vs. local incidence angle 

highly burned pseudo-plots (TerraSAR-X and 

ASAR data). 

To infer the utility of backscatter coefficient for burn severity estimation 

determination coefficients (R 2 ) expressing the proportion of burn severity 

variance predicted by radar backscatter are presented in Table 1 for each 

sensor and polarization. The determination coefficients of were computed 

using all available pseudo-plots and for 5º intervals of local incidence 

angles.

262 


Sensor TerraSAR-X Sensor ASAR 

Polarization HH HV HH&HV Polarization VV HH HV HH&HV 

Scene data 12.1908 Scene data 09.12.08 01.08.09 03.19.09 

All pseudo-plots 0.529 0.085 0.546 0.101 0.176 0.539 0.542 

Pseudo-plots grouped by local incidence angle 

26°-30° 0.833 0.291 0.916 6°-10° 0.359 0.768 0.700 0.843 

31°-35° 0.707 0.234 0.861 10°-15° 0.338 0.555 0.733 0.758 

36°-40° 0.732 0.050 0.857 16°-20° 0.361 0.449 0.609 0.648 

41°-45° 0.666 0.291 0.815 21°-25° 0.186 0.354 0.693 0.752 

46°-50° 0.621 0.402 0.792 26°-30° 0.001 0.198 0.699 0.699 

51°-55° 0.612 0.520 0.824 31°-35° 0.007 0.154 0.767 0.767 

56°-60° 0.584 0.464 0.833 36°-40° 0.001 0.174 0.715 0.715 

61-65 0.476 0.582 0.843 41-45 0.050 0.109 0.744 0.744 

Table 1 - Determination coefficients explaining the agreement between dNBR and radar 

backscatter at X- and C-band. 

5 - Discussion 

Both sensors showed similar trends of the backscatter with respect to dNBR 

for co- and cross-polarized beams. Figures 1 and 2 show that for co-polarized 

beams the backscattering coefficient increased with burn severity. For 

X-band HH-polarization, the backscattering coefficient increased (Figure 2) 

from low to medium burn severity reaching saturation for high burn severities 

(i.e. dNBR 800). The HV-backscatter 

decreased with the increase of burn severity at both frequencies 

(Figures 1 and 2). This is explained by the larger amount of gaps in the 

canopy and the increased transmissivity with the increase of burn severity. 

Figure 3 shows the strong effects of the local incidence angle on the 

backscatter at all SAR frequencies and polarizations. For increasing local 

incidence angle decreasing trends were registered for co-polarized beams. 

This decrease is the result of a decreasing proportion of direct scatter from 

the ground to the total forest backscatter. Cross-polarized beams showed 

increase of the backscatter coefficient with increasing local incidence angle 

explained by the longer path traveled by the radar wave within the canopy. 

The strength of association (i.e. R 2 ) for co-polarized beams decreased with 

increasing incidence angle while the error increased (Table 1). The simultaneous 

use of cross and co-polarized channels (HH and HV) provided the 

highest R 2 coefficients and the smallest errors.


Backscatter properties of x- and c-band sar in a mediterranean pine forest affected by fire 263 

This work was a first step towards the understanding of the properties of 

SAR backscatter of burned areas. Aim of this study is to develop new methods 

for burn severity mapping. The study indicates that dual polarization 

data has good potential for burn severity mapping in the Mediterranean 

environments as long as the local incidence angle is accounted for. 


The work has been financed by the Spanish Ministry of Science and 

Education and the European Social Fund. ASAR and TerraSAR-X data were 

provided by the ESA and DLR. 

References 

Key, C.H. and Benson, N.C., 2006. Landscape assessment (LA). In Lutes, 

D.C., Keane, R.E., Caratti, J.F., Key, C.H., Benson, N.C., Sutherland S. & 

Gangi, L.J., (Eds.), FIREMON: Fire effects monitoring and inventory system. 

Fort Collins, CO: U.S. Department of Agriculture, Forest Service, 

Rocky Mountain Research Station, Gen. Tech. Rep. RMRS-GTR-164-CD: 

pp. 1-55. 

Tanase, M., Pérez-Cabello, F., Riva, J.d.l. and Santoro, M., 2009. TerraSAR- 

X data for burn severity evaluation in Mediterranean forests on sloped 

terrain. IEEE Trans. Geosci. Remote Sensing, accepted for publication. 

Ulander, L.M.H., 1996. Radiometric slope correction of synthetic-aperture 

radar images. IEEE Trans. Geosci. Remote Sensing, 34: 1115-1122.

Abstract: Long Term Data Record (LTDR) 5 km AVHRR data was used to generate 

burned area in Western Canada and compared to Local Area Coverage 

(LAC) 1.1 km and Pathfinder 8 km dataset (PAL) from 1984 to 1999. The 

algorithm uses a Bayesian Network (naive Bayes) trained with LAC burned 

area. The results matched very well LAC and improved the results over PAL. 

Additional research is needed to extrapolate results to other boreal areas, 

such as Siberia, where LTDR 5 km exists since 1981, but LAC 1.1 km is only 

available since the 90s. 


Boreal forests contribute to store carbon that is released to the atmosphere 

as a consequence of wildfires. Due to the slow vegetation recovery process, 

the increase in boreal forest burning can modify the carbon balance significantly 

(Balshi et al., 2009). Recently the LTDR project, funded by NASA, 

created a global daily dataset from the AVHRR Global Area Coverage (GAC) 

at 5 km (Nagol et al., 2009; Pedelty et al., 2007) from 1981 to 2006. LTDR 

has improved atmospheric correction and inter-calibration between sensors. 

2 - Methods 

BURNED AREA DATA TIME SERIES FROM LTDR DATASET 

FOR CANADA (1981-2000) 

J.A. Moreno Ruiz1, 2 D. Riaño1, 3 , A. Alonso-Benito4 , 

N.H.F. French5 & S.L. Ustin1 1 CSTARS, University of California, 250-N, The Barn, One Shields Avenue, Davis, CA 95616-8617, USA 

2 Universidad de Almería, Almería, Spain 

3 Centro de Ciencias Humanas y Sociales, CSIC, Madrid, Spain 

4 Universidad de La Laguna, La Laguna, Tenerife, Spain 

5 MTRI, Michigan Technological University, Ann Arbor, MI, USA 

jamoreno@ual.es; driano@cstars.ucdavis.edu; asaloben@gmail.com; 

nhfrench@mtu.edu; slustin@ucdavis.edu 

LTDR dataset version 2 has been available since October 15th, 2007 

(http://ltdr.nascom.nasa.gov/ltdr/productSearch.html, last accessed 

07/06/2009). A 10 day composite from daily data was built from 1984 to 

1999 with the minimum albedo criterion. We selected Chuvieco et al. 

(2008) study site in Western Canada (UL: 62º16’N,-122º20’E; LR: 53ºN,- 

96º30’E), where the authors generated burned area maps. 

The algorithm to calculate burned area from LTDR was trained with the 

265

266 


Chuvieco et al. (2008) LAC burned area record resampled to 5 km using 

nearest neighbor to match LTDR. The year 1989 was used for training due 

to the largest burned area (~12,000 km 2 ). A total of 895 unburned pixels 

were selected in between those pixels that had a Global Environment 

Monitoring Index (GEMI) (Pinty and Verstraete, 1992) > 0.36. Out of 814 

burned pixels, we selected for training 298 under boreal forest vegetation 

type, with the following relaxed fix thresholds on reflectance channel 2 

(0.83 µm, r 2 ) < 0.125, brightness temperature (~3.75 µm, T 3 ) > 300 K and 

Vi3T (Barbosa et al., 1999) < -0.45. 

We considered from April to September, when wildfires are likely to occur, 

the year before, during and after the fire. Median, mean, maximum (Max) 

and minimum (Min) were computed for: reflectance channel 1 (0.63 µm, 

r 1 ), r 2 and T 3 and several spectral indexes: Vi3T, GEMI and a new Burned 

Boreal Forest Index (BBFI) (1/ρ 2 + T 3 /2). 

Bayes Net (naive Bayes) classifier in the machine learning package WEKA 

(http://www.cs.waikato.ac.nz/~ml/index.html) was applied to determine 

the probability of a pixel to be burned and the probability to be unburned. 

In the Bayesian Network modeling a gain ranking filter determined which 

variables had the highest significance. The following ones, ranked in order 

of significance, were selected: Median_BBFI__after, Max_BBFI__after, 

Median_GEMI_after, Max_BBFI_during, Min_Vi3t_after, Min_Vi3t_during, 

Min_ρ 2 _after, Min_ρ 2 _during, Max_T 3 _after. 

After the training process, the classifier outputs probability density functions 

for the selected variables for both classes, burned and unburned. 

Following Bayes´s theorem (Bayes 1763), the joint probability density function 

for a given class was written as a product of the individual density 

functions. 

Before running each pixel through the Bayesian Network model, we applied 

the following relaxed thresholds to avoid false burned detections: 

(Max_T 3 _during > 300) and (Min_ρ 2 _during < 0.1) and (Max_BBFI_during 

> 160) and (Max_BBFI_during 

Median_BBFI_before) and (Median_BBFI_after >= 158) and 

(Median_GEMI_after < Median_GEMI_before) and (Median_GEMI_before >= 

0.36). 

The Bayesian model provided the probability of a pixel to be burned and 

the probability to be unburned. One option was assigning the class to each 

pixel with higher probability, but in order to handle data uncertainty and 

avoid false detections, we computed the normalized probability ([0, 1]) for 

each class (burned/unburned) and calculated the difference between both, 

resulting a value in the range [-1, 1]. Pixels with final probabilities lower 

than 0 were flagged as unburned. Based on the training dataset, we used 

a classification tree J48 in WEKA to determine that a pixel should always 

be flagged burned, if the final probability was >0.997. Pixels within the 

range [0, 0.997] were flagged as potentially burned and assigned to burned 

or unburned based on the neighboring pixels. The results were finally 

smoothed by comparing it to the 4 neighbors to avoid false burned detec-

Burned area data time series from ltdr dataset for Canada (1981-2000) 267 

tions: 1) a burned pixel could not be neighbor to water; and 2) a burned 

pixel could not be isolated. In addition, to handle data uncertainty: 1) a 

potentially burned was flagged as burned if it had a neighbor burned; and 

2) a potentially burned pixel was flagged unburned if it did not have at 

least two burned neighbors. 

The algorithm trained with 298 burned pixels and 895 unburned in 1989 

was applied to the entire study site, from 1984 to 1999. After inspecting 

the results, we selected 1989, 1993, 1994, 1995 and 1998 for training. 

1989 was used to estimate burned area from 1984 to 1992, whereas 1998 

from 1996 to 1999. The method was the same as the one described above, 

defining new minimum requirements, selecting the most significant variables 

and keeping the same total probability thresholds for burned, potential 

and unburned pixels. 


Figure 1 shows the results of the LTDR burned area calculated for Western 

Canada in comparison to the LAC (Chuvieco et al., 2008) and PAL (Riaño et 

al., 2007) products. LTDR, when trained only with year 1989, showed similar 

trends to LAC but burned area was underestimated. When training with 

1989, 1993, 1994, 1995 and 1998 results matched much better LAC. 1984- 

1992 corresponds to AVHRR 7, 9 and 11 satellites. 1993, 1994 were the end 

of life for AVHRR 11 with high orbital degradation, 1995 was the beginning 

of AVHRR 14, and r 2 was inconsistent between AVHRR 11 and 14 (Moreno 

Ruiz et al., 2009). The algorithm does not overestimates burned area, but 

needs to be retested in other parts of Canada to ensure that it is not overfitting 

the test site. The use of the previous season to map burned area was 

not significant, but the year after the fire was critical. French et al., (1995) 

mapped burned area analyzing the burned scars the year following the fire, 

since boreal forest takes more than a year to recover from burning. 

Figure 1 - Burned 

area estimations) 

for different AVHRR 

products.

268 


The algorithm was trained using the boreal forest sites, so it could be tuned 

to other vegetation types or a general model could be tested for all of them 

or even new variables could be added. The total probability thresholds for 

burned, potentially burned and unburned pixels could be adjusted to each 

year instead of using the value computed for 1989. LTDR 5 km improved the 

results over PAL which was proven inaccurate for the boreal regions 

(Carmona-Moreno et al., 2005; Chuvieco et al., 2008). 


This work was supported in part by the Spanish Commission for Science and 

Technology (CICYT) under Grant CGL2007-66888-C02-02/CLI. Thank you to 

Emilio Chuvieco for providing the burned area record generated from LAC 

data. Useful comments on earlier drafts of the manuscript are acknowledged 

from Julia Marcia Medina. 

References 

Balshi, M.S., McGuire, A.D., Duffy, P., Flannigan, M., Kicklighter, D.W. and 

Melillo, J., 2009. Vulnerability of carbon storage in North American 

boreal forests to wildfires during the 21st century. Global Change 

Biology, 15 (6), 1491-1510. 

Barbosa, P.M., Grégoire, J.M. and Pereira, J.M.C., 1999. An algorithm for 

extracting burned areas from time series of AVHRR GAC data applied at 

a continental scale. Remote Sensing of Environment, 69, 253-263. 

Bayes, T., 1763. An essay towards solving a problem in the doctrine of 

chances. Philosophical Transactions of the Royal Society of London, 53, 

370-418. 

Carmona-Moreno, C., Belward, A., Malingreau, J.P., Hartley, A., Garcia- 

Alegre, M., Antonovskiy, M., Buchshtaber, V. and Pivovarov, V., 2005. 

Characterizing inter-annual variations in global fire calendar using data 

from Earth observing satellites. Global Change Biology, 11 (9), 1537- 

1555. 

Chuvieco, E., Englefield, P., Trishchenko, A.P. and Luo, Y., 2008. Generation 

of long time series of burn area maps of the boreal forest from NOAA- 

AVHRR composite data. Remote Sensing of Environment, 112 (5), 2381- 

2396. 

French, N.H.F., Kasischke, E.S., Bourgeauchavez, L.L. and Berry, D., 1995. 

Mapping the Location of Wildfires in Alaskan Boreal Forests Using Avhrr 

Imagery. International Journal of Wildland Fire, 5 (2), 55-62. 

Moreno Ruiz, J.A., Riaño, D., García Lázaro, J.R. and Ustin, S.L., 2009. 

Intercomparison of AVHRR PAL and LTDR Version 2 Long Term Datasets 

for Africa from 1982 to 2000 and Its Impact on Mapping Burned Area. 

IEEE Geoscience and Remote Sensing Letters, in press.

Burned area data time series from ltdr dataset for Canada (1981-2000) 269 

Nagol, J.R., Vermote, E.F. and Prince, S.D., 2009. Effects of atmospheric 

variation on AVHRR NDVI data. Remote Sensing of Environment 113, 

392-397. 

Pedelty, J., Devadiga, S., Masuoka, E., Brown, M., Pinzon, J., Tucker, C., 

Roy, D., Ju, J., Vermote, E., Prince, S., Nagol, J., Justice, C., Schaaf, C., 

Liu, J., Privette, J. and Pinheiro, A., 2007. Generating a long-term land 

data record from the AVHRR and MODIS instruments. Geoscience and 

Remote Sensing Symposium, 2007. IGARSS 2007 (pp. 1021-1025): IEEE 

International. 

Pinty, B. and Verstraete, M.M., 1992. GEMI: a non-linear index to monitor 

global vegetation from satellites. Vegetatio, 101, 15-20. 

Riaño, D., Moreno Ruiz, J.A., Isidoro, D. and Ustin, S.L., 2007. Spatial and 

temporal patterns of burned area at global scale between 1981-2000 

using NOAA-NASA Pathfinder. Global Change Biology, 13 (1), 40-50.

CORRECTION OF TOPOGRAPHIC EFFECTS INFLUENCING 

THE DIFFERENCED NORMALIZED BURN RATIO’S OPTIMALITY 

FOR ESTIMATING FIRE SEVERITY 

Abstract: The influence of illumination effects on the optimality of the 

dNBR (differenced Normalized Burn Ratio) was evaluated for the case of the 

2007 Peloponnese (Greece) wildfires using Landsat TM (Thematic Mapper) 

imagery. Well illuminated pixels exhibited more optimal displacements in 

the bi-spectral feature space than more shaded pixels. To correct for illumination 

effects, the c-correction method and a modified c-correction 

technique were applied. The resulting mean dNBR optimality of uncorrected, 

c-corrected and modified c-correction data was respectively 0.57, 0.59 

and 0.66. Applying a topographic correction significantly improves the reliability 

of change detection especially in rugged terrain and when low sun 

angle images are used. 


S. Veraverbeke, R. Goossens 

Department of Geography, Ghent University, Ghent, Belgium 

sander.veraverbeke@ugent.be 

W. Verstraeten 

Geomatics Engineering, Katholieke Universiteit Leuven, Leuven, Belgium 

S. Lhermitte 

Centro de Estudios Avanzados en Zonas Aridas (CEAZA), La Serena, Chile 

Fire severity is an important factor in post-fire assessment. The Normalized 

Burn Ratio (NBR) is the standard spectral index to estimate fire severity 

(Key et al., 2005): 

TM4 - TM7 

NBR = --------------------- (1) 

TM4 + TM7 

where TM4 and TM7 are respectively the NIR and MIR reflectance of Landsat 

TM imagery. Bi-temporal image results in the differenced Normalized Burn 

Ratio (dNBR). A pixel-based optimality measure which evaluates a pixel’s 

movement in the bi-spectral feature space has been developed (Roy et al., 

2006). An optimal fire severity spectral index needs to be sensitive to fireinduced 

vegetation changes and insensitive to perturbing factors such illumination 

effects. 

These illumination effects are initiated by both topography and solar position 

at the moment of image acquisition. Topographic effects in ratio- 

271

272 


images based analysis are assumed to be minimal and are not considered in 

most studies using the dNBR. Therefore the focus of this research is to 

evaluate illumination effects and to propose a topographic correction 

approach to become a more reliable dNBR fire severity assessment. 

2 - Methods 

2.1 - Study area and data-preprocessing 

The study area is situated at the Peloponnese (Greece). Large wildfires 

stroke the Peloponnese in the 2007 summer. The pre-fire image dates from 

23/07/2006, whereas the post-fire image was acquired at 28/09/2007. The 

images were subjected to geometric, radiometric and atmospheric correction 

(Chavez, 1996). 

2.2 - dNBR optimality 

For evaluating the optimality of the bi-temporal change detection the TM4- 

TM7 bi-spectral space was considered. A displacement from unburned (U) 

to optimally (O) sensed burned is illustrated in figure 1. Perturbing factors 

decrease the performance of the index. The magnitude of change to which 

the index is insensitive is equal to the Euclidian distance. Thus the 

observed displacement vector UB can be decomposed in the sum of the vectors 

UO and OB, hence, following the expression of Roy et al. (2006) the 

index optimality is defined as: 

⎟OB⎥ 

optimality = 1 - --------------- (2) 

⎟UB⎥ 

Figure 1 - Example pre/post-fire trajectory 

of a pixel in the TM4-TM7 feature 

space.

Correction of topographic effects influencing the differenced normalized burn ratio’s optimality for estimating fire severity 273 

2.3 - Correcting for illumination effects 

Methods that correct for illumination effects are based on the cosine of the 

incidence angle, which is the angle between the normal to the ground and 

the sun rays (Teillet et al., 1982): 

cos γ i = cos θ p cos θ z + sin θ p sin θ z cos (φ a - φ o ) (3) 

where γ i is the incident angle; θ p is the slope angle; θ z is the solar zenith 

angle; φ a is the solar azimuth angle; and φ o is the aspect angle. 

In the c-correction method terrain corrected reflectance ρ t is defined as: 

( cos θ ) z + ck ρt = ρa ---------------cos 

γt + ck where c k is a band specific parameter. 

We propose a modified c-correction method that corrects reflectance to a 

maximum illumination cos γ t of one, instead of normalizing as a function 

of the solar zenith angle: 

( ) 

1 + c k 

ρ t = ρ a ----------------cos 

γ t + c k 

Index optimality was compared among eight aspect classes and among different 

classes of bi-temporally averaged illumination. 


Figures 2A-C depict the topographically uncorrected, c-corrected and modified 

c-correction dNBR optimality maps. The modified c-correction dNBR 

optimality (mean = 0.66) outperformed c-corrected and uncorrected optimality 

(means of respectively 0.59 and 0.57), whereas c-corrected optimality 

provided slightly better results than uncorrected optimality. This is also 

reflected when the respective histograms are inspected (see figures 2D-F). 

(4) 

(5)

274 


Figure 2 - Topographically uncorrected (a and d), c-corrected (b and e) and modified c-correction 

(c and f) dNBR optimality maps and histograms.

Correction of topographic effects influencing the differenced normalized burn ratio’s optimality for estimating fire severity 275 

Figure 3 - Mean topographically uncorrected, c-corrected and modified c-correction dNBR optimality 

score by aspect class (a), average illumination class (b). 

In comparison with topographically uncorrected data, both topographic 

correction techniques increased the optimality of badly illuminated areas 

but the modified c-correction method ameliorated more than the original 

c-correction technique. For well illuminated areas, however, dNBR optimality 

dropped after c-correction while the modified c-correction realized 

slightly better optimality scores than uncorrected data (see figure 3). 


Based on the spectral index theory, the effect of illumination on the dNBR 

optimality for assessing fire severity using pre- (2006) and post-fire (2007) 

Landsat TM imagery was evaluated for the 2007 Peloponnese wildfires. To 

improve the performance of the index, the c-correction method and a modified 

version of this technique were applied to derive terrain corrected 

reflectance. The resulting mean dNBR optimality scores of topographically 

uncorrected, c-corrected and modified c-correction data were respectively 

0.57, 0.59 and 0.66. The modified c-correction method resulted in a more

276 


reliable fire severity assessment, implying a big potential for other spectral 

indices based change detection studies in rugged terrain with low sun angle 

imagery. 

References 

Chavez, P., 1996. Image-based atmospheric corrections - revisited and 

improved. Photogrammetric Engineering & Remote Sensing, 62, 1025- 

1036. 

Key, C. and Benson, N., 2005. Landscape assessment: ground measure of 

severity; the Composite Burn Index, and remote sensing of severity, the 

Normalized Burn Index. In Lutes, D., Keane, R., Caratti, J., Key, C., 

Benson, N., Sutherland S. and Gangi, L., (Editors). FIREMON: Fire effects 

monitoring and inventory system (pp. 1-51). USDA Forest Service, Rocky 

Mountains Research Station, General Technical Report RMRS-GTR-164-CD 

LA. 

Roy, D., Boschetti, L. and Trigg, S., 2006. Remote sensing of fire severity: 

assessing the performance of the Normalized Burn Ratio. IEEE 

Transactions on Geoscience and Remote Sensing, 3, 112-116. 

Teillet, P., Guindon, B. and Goodenough, D., 1982. On the slope-aspect correction 

of multispectral scanner data. Canadian Journal of Remote 

Sensing, 8, 84-106.

BURNED AREAS MAPPING BY MULTISPECTRAL IMAGERY: 

A CASE STUDY IN SICILY, SUMMER 2007 

P. Conte & G. Bitelli 

DISTART - University of Bologna, Bologna, Italy 

paolo.conte2@unibo.it; gabriele.bitelli@unibo.it 

Abstract: This work reports some experiments in classifying and mapping 

the areas which were affected by wildfires in centre-eastern Sicily in June 

2007. The study exploited moderate resolution images acquired by ASTER 

sensor (Advanced Spaceborne Thermal Emission and Reflection Radiometer) 

and compared four types of classification, all of which based on spectral 

indices specially developed for this kind of application: BAI (Burned Area 

Index), NBR (Normalized Burned Ratio) and MIRBI (Mid-Infrared Bispectral 

Index). After assessing each index by means of direct interpretation and 

set-up of appropriate threshold values, the method was optimized by their 

multiple thresholding. This approach was used for both single images and 

multi-temporal series. For each index some confusion errors with different 

kinds of surface (e.g. farmed areas, urban areas, water bodies) were reported. 

The simultaneous use of the indices made the classification more accurate 

with fewer commission errors. 


Every year Italy is affected by countless forest fires which have a considerable 

impact for number of human casualties and extent of economic, social 

and environmental damage; in spite of a greater financial and technological 

commitment to face such events (e.g. the duty for local government 

bodies to draw up and yearly update a cadastre of forest fires), the number 

and extent of burned areas have increased from 2006 to 2007. In fact two 

recent reports (Indagini Ecosistema Incendi, Legambiente, 2006 and 2007) 

show a change in the trend of recent past: in 2006 there had been 5,643 

events of wildfires for a total burned area of about 40,000 ha, with a little 

improvement in comparison with previous years. Instead, 2007 was a critical 

year with 10,614 events (top value since 1997) for a total burned area 

of 225,000 ha (top value since 1981), with 23 dead and 26 wounded apart 

from heavy economic and environmental damages. The South of Italy is 

277

278 


indeed the most affected area, with a peak in the number of wildfires in 

the summer because of climatic factors (e.g. high temperature) and fuel 

properties. 

In order to preserve and protect woodland areas, Law 353/2000 imposes 

the drafting of regional plans for the forecast, prevention and active fight 

of fires in addition to formative, informational and educational activities. 

It is worth noticing that, even if the number of towns with a cadastre of 

forest fires passed from 24% in 2006 to 46% in 2007, only 6% of them 

thoroughly comply with the directives of the Outline Law in force. Remote 

sensing can help to fulfil the requirements of the law in an economic and 

effective way as, along with other applications, it allows to detect and map 

fire-affected surfaces over wide areas, in inaccessible and dangerous places 

and at a low cost when compared to traditional survey techniques. 


This work is a qualitative evaluation of both potentials and limits of several 

algorithms for post-fire mapping of burned areas by means of specially 

designed spectral indices such as BAI (Burned Area Index), NBR 

(Normalized Burn Ratio) and MIRBI (Mid-Infrared Bispectral Index) 

(Chuvieco et al., 2002; Benson, 1999; Trigg & Flasse, 2001). In order to get 

over the problems highlighted by the single indexes two different approaches 

were used: in a mono-temporal approach we employed a single post-fire 

image for computing and combining the indices to be used in classification, 

while in a multi-temporal approach (change detection) we assessed 

for each pixel the difference between the index values computed for the 

pre-fire and post-fire images. 

The analysis was performed by using multispectral moderate resolution 

images from ASTER sensor, operating on NASA’s Terra platform; in particular 

we made use of AST_07XT products (On-Demand L2 Surface Reflectance), 

which, apart from geometric and radiometric corrections, are characterized 

by a further adjustment taking atmospheric conditions and changing in 

earth-sun relative positions into account - which allows to transform radiance 

value of image pixels in surface reflectance of the corresponding 

object on the earth surface. We used files including both the 3 bands of 

VNIR sub-system (spatial resolution 15 m) and the 6 bands of SWIR subsystem 

(spatial resolution 30 m). 

In the mono-temporal approach a single post-fire image acquired on 28th 

June 2007 at 09.53.36 a.m. was processed: the image was chosen for the 

absence of clouds, the availability of data about wildfires occurred in the 

previous period, and mainly because it is immediately following the peak 

temperature period occurred in Sicily between 23rd and 26th June 2007 

(Figure 1). For the change-detection approach we made also use of another 

image acquired on 9th April 2007 at 09.53.47 a.m., in a period with 

scarce wildfire events.

Burned areas mapping by multispectral imagery: a case study in Sicily, summer 2000 279 

Figure 1 - Tmax /Tmed 26th June 2007 (data: SIAS). 

The zone chosen for experimentation (Figure 2) is in centre-eastern Sicily. 

In the single image approach the extension of the analysed area is the same 

as the one covered by the ASTER image, that is 60 Km x 60 Km, and 

includes almost completely the province of Enna and portions of the 

provinces of Palermo, Catania and Caltanissetta, whilst in the changedetection 

technique it is much smaller, just about 2873 square kilometres, 

as it consists only of the area resulting from the overlapping of the two 

images used. 

Figure 2 - Area of study.

280 


We started with a resampling, which, by means of a bilinear interpolation, 

transformed the six bands of SWIR subsystem to the same resolution of 

VNIR bands (obtaining a file with nine bands all with a 15 m spatial resolution). 

The following step was the choice of the most suitable RGB combination 

for photo-interpretation (Chirici & Corona, 2005; Epting et al., 

2005; Grégoire & Brivio, 2001; Lentile et al., 2006). After a preliminary 

analysis we decided to make use of 8-3-2 (SWIR-NIR-R) bands and a simulated 

Natural Colour Composite representation. Whilst the latter simulates 

a sham band corresponding to the blue region by means of a weighted 

average of green and red values - which permits to remove all doubts for 

surfaces with a response similar to the one connected with burned areas in 

the infrared region - the former is more sensitive to fire effects. 

The first step in the mono-temporal approach consisted in computing index 

values in post-fire image pixels by means of band-math functions in ENVI. 

Reference formulas are: 

1000000 b2 + b3 

BAI = ---------------------------------- NBR = --------------- MIRBI = 10·b3-9,8·b4+2,0 

(100-b1) 2 + (60-b2) 2 b2-b3 

where b1=band 2 (red), b2=band 3 (NIR) b3=band 8 (SWIR) and b4=band 

4 (SWIR). 

This procedure allowed us to carry out three different pixel-based binary 

classifications thanks to independent thresholding of every single computed 

index: for BAI and NBR, predefined threshold values were chosen 

(Zaffaroni et al., 2007); instead, for MIRBI we computed frequency histograms 

both for the whole image and for assessed wildfires and we chose 

threshold values so that they could contain almost every actually burned 

pixel. 

Therefore we assigned a pixel to the burned class when it satisfied the following 

conditions: 

DN > 40 for BAI index; 

-0,3 < DN < 0.1 for NBR index; 

-650 < DN < 0 for MIRBI index. 

In order to solve the confusion errors peculiar to each index an AND logic 

operation of the three resulting classifications was applied: a pixel was 

assigned to the class of fire-affected surfaces only if it was flagged as 

potentially burned by all the three algorithms, therefore if it satisfied the 

condition: 

(-0,3 < NBR < 0.1) & (BAI > 40) & (-650 < MIRBI < 0) 

A further attempt to optimise the single-image approach consisted in modifying 

the threshold values used in this algorithm on the basis of the statistical 

index distribution for assessed wildfires. 

We slightly modified values to better fit BAI, NBR and MIRBI ranges to the 

histograms of assessed burned areas. This choice allowed us to obtain the 

condition


(-0,15 < NBR < 0.09) & (30 < BAI < 150) & (-650 < MIRBI < -150) 

to decide whether a pixel is burned or not. 

For the multi-temporal approach the first step was a geometric co-registration 

which made it possible to compute temporal differences in indices for 

every pixel; for this purpose, 40 Ground Control Points present on both 

images were selected. The following step was a radiometric normalization 

by an empirical method, with the aim to minimise differences between 

reflectance values of invariant pixels and consequently obtain index values 

as alike as possible for objects having analogous spectral characteristics in 

both images. 

The values of BAI and NBR indices were derived for pre- and post- fire 

images and, thanks to the pre-processing operations applied, the differences 

dBAI and dNBR were computed, and then the change detection 

(IndexPost-IndexPre). The empiric threshold values to be applied were chosen 

following the same procedure used in the previous approach, i.e. applying 

an AND logic operation between the two classifications we had 

obtained: a pixel belonged to the burned class if it satisfied 

(dBAI > 250) & (-2.0 < dNBR < -0.55) 

The last stage was a post-classification: firstly sieving and clumping filters 

were applied - which respectively permit to remove isolated points and 

aggregate contiguous polygons - and then the results of the 6 binary classifications 

were converted in vector form. 


Classification algorithms using the thresholding of single indexes are certainly 

very quick; they don’t need elaborate procedures but they show heavy 

confusion errors due to surfaces having almost the same spectral 

reflectance as burned areas in the bands used. In the case study in fact BAI 

results included lakes and other water bodies in the burned class; NBR 

reduced this kind of misidentification only to coastal areas, but showed 

also relevant errors with both urban and rural areas. Finally, MIRBI presented 

commission errors with both urban areas and water bodies but was 

more accurate with rural areas. This situation is shown in Figure 3 for a portion 

of the area. 

Multiple thresholding with fixed values significantly reduced the errors met 

in previous methods and permitted an accurate mapping of wildfires with 

just a slightly greater operative complexity. However, when dealing with old 

or greatly heterogeneous fires, spectral sensitivity may range in value 

between burned and intact areas - thus leading to contradictory results.

282 


Figure 3 - (a) RGB 8-3-2 ASTER combination; commission errors with (b) BAI, (c) NBR, (d) 

MIRBI. 

Optimizing this algorithm by a semi-empiric choice of threshold values 

requires well-trained operators and the availability of data on ground 

assessed fires, but significantly improves the quality of classification: the 

above mentioned surfaces are mapped with greater accuracy and no relevant 

errors are found; an example of this situation is shown in Figure 4. 

Figure 4 - From left to right: old wildfire; multiple threshold; optimized algorithm. 

The multi-temporal approach is certainly the most complex of all methods 

we dealt with, because image pre-processing is needed: results are similar 

to the ones obtainable by multiple thresholding, with few commission 

errors probably due to changes in land cover such as water level fluctuations 

in water bodies (Figure 5).



Figure 5 - From left to right: RGB 8-3-2 ASTER combination; multiple threshold; multi-temporal 

approach. 

However quick and comparatively easy to process they are, algorithms 

based on a single index produced errors with different surfaces and cannot 

always assure reliable classifications. The scrutiny of frequency histograms 

highlights that burned areas have a smaller variance with BAI than with 

other indices; therefore BAI is the most sensitive to the spectral response 

of burned vegetation and permits a better analysis as it computes the bispectral 

distance from a reference point in the R-NIR domain for every pixel 

(Chuvieco et al., 2002; Epting et al., 2005). 

The multiple thresholding of all three indices with fixed threshold values 

which do not depend on the statistical distribution of image pixels requires 

a more complex processing but allows greatly improving accuracy in classification. 

In fact a simultaneous use of four spectral bands in three equations 

instead of only two bands in a single index equation allows to remove 

errors with both water bodies and urban areas; classifications may still be 

uncertain in the case of either very heterogeneous fires or of events 

occurred long before the image was acquired). Furthermore it should be 

noted that this method does not require either qualified operators or 

assessed data on previous fires. 

Accuracy in classification with this approach might be slightly improved by 

means of a meticulous selection of threshold values - which would, however, 

require accurate statistical studies of histograms from well trained operators 

and ground validation of data. The quality of the results would not 

justify such a greater operational complexity. 

Finally, the classification based on a multi-temporal approach gives almost 

the same results which can be obtained when using the multiple threshold 

method in a mono-temporal approach, but requires fairly long and complex 

pre-processing procedures and does not succeed in removing every commission 

error. Here errors are connected with changes in the spectral sensitivity 

of the same area/s in each of the two different images, whereas in 

the mono-temporal approach they are due to surfaces having almost the 

same reflectance as burned pixels.

284 


For this study case, in conclusion, the multiple thresholding with fixed values 

seems to be the best classification method for this application: it 

requires relatively fast processing, does not need either ground validation 

or operators’ great expertise and gives fairly accurate results (especially 

when images with a good temporal coverage are used). 

References 

Chirici, G., Corona, P., 2005. An overview of passive remote sensing for postfire 

monitoring. Forest@, 2(3), 282-289. 

Chuvieco E., Martin M. P., Palacios A., 2002. Assessment of different spectral 

indices in the red- near- infrared spectral domain for burned land discrimination. 

Int. J. of Remote Sensing, 23, 5103-5110. 

Epting, J., Verbyla, D., Sorbel, B., 2005. Evaluation of remotely sensed 

indices for assessing burn severity in interior Alaska using Landsat TM and 

ETM+. Remote Sensing of Environment, 96, 328-339. 

Grégoire, J-M., Brivio, P.A., 2001. Mapping of burnt areas at global level: 

current possibilities offered by optical Earth Observation Systems 

(http://web.crii.uninsubria.it/infoterra/docs/ Insubria_Brivio.ppt) . 

Key, C.H., Benson, N., 1999. The Normalized Burn Ratio (NBR): a Landsat 

TM radiometric index of burn severity incorporating multitemporal differencing 

(http://nrmsc.usgs.gov/research/nbr.htm) . 

Lentile, L., Holden, Z., Smith A., Falkowski M., Hudak, A., Morgan, P., Lewis, 

S., Gessler, P., Benson, N., 2006. Remote sensing techniques to assess 

active fire characteristics and post-fire effects. Int. J. of Wildland Fire, 

15, 319-345. 

Trigg, S., Flasse, S., 2001. An evaluation of different bi-spectral spaces for 

discriminating burned shrub-savanna. Int. J. of Remote Sensing, 22, 

2641-2647. 

Zaffaroni, P., Stroppiana, D., Brivio, P.A., Boschetti, M., 2007. Utilizzo di 

immagini ASTER per la delimitazione di aree percorse da incendio. Rivista 

italiana di Telerilevamento, 39, 93-101.

Abstract: In this paper NDVI VEGETATION time series were analysed using 

the Detrended Fluctuation Analysis (DFA) to estimate post fire vegetation 

recovery.The DFA is a well-known methodology, which allows the detectin 

of long-range power-law correlations in signals possibly characterized by 

nonstationarity, which features most of the observational and experimental 

signals. Results from our analysis point out that the persistence of vegetation 

dynamics is significantly increased by the occurrence of fires. In particular, 

a scaling behavior of two classes of vegetation (burned and 

unburned) has been revealed. The estimated scaling exponents of both 

classes suggest a persistent character of the vegetation dynamics. But, the 

burned sites show much larger exponents than those calculated for the 

unburned sites. 


POST FIRE VEGETATION RECOVERY ESTIMATION USING 

SATELLITE VEGETATION TIME SERIES 

R. Lasaponara , R. Coluzzi , F. Desantis, 

A. Lanorte, L. Telesca 

CNR-IMAA, Tito Scalo (PZ) Italy 


The dynamics of vegetation covers in burned and unburned areas can be 

monitored by using satellite data, which provide a wide spatial coverage 

and internal consistency of data sets. Several indices can be used to perform 

such kind of remote sensing monitoring. In particular, NDVI 

(Normalized Difference Vegetation Index) obtained from the visible (Red) 

and near infrared (NIR) by using the following formula NDVI= (NIR- 

Red)/(NIR+ Red), is the most widely used index to follow the process of 

recovery after fire. 

This investigation aims to perform a dynamical characterization of burned 

and unburned vegetation covers, using time series of remotely sensed data 

of two fire-affected and two fire-unaffected sites. For this purpose, we used 

the Detrended Fluctuation Analysis (DFA), which permits the detection of 

persistent properties in nonstationary signals. 

285

286 


2 - Methods 

The DFA method works as follows. In order to analyze the NDVI time series, 

we briefly present an introduction to the detrended fluctuation analysis 

(DFA), which is constituted by the following steps: 

1) Consider the decadal NDVI signal x(i), where i=1,…,N, and N is the total 

number of decades. We integrate the signal x(i) and obtain 

where is the mean value of x. 

2) The integrated signal y(k) is divided into boxes of equal length n. 

3) For each n-size box, we fit y(k), using a linear function, which represents 

the trend in that box. The y coordinate of the fitting line in each 

box is indicated by yn(k). 

4) The integrated signal y(k) is detrended by subtracting the local trend 

yn(k) in each box of length n. 

5) For given n-size box, the root-mean-square fluctuation, F(n), for this 

integrated and detrended signal is given by 

6) The above procedure is repeated for all the available scales (n-size box) 

to furnish a relationship between F(n) and the box size n, which for 

long-range power-law correlated signals is a power-law F(n)~nα. (3) 

7) The scaling exponent a quantifies the strength of the long-range powerlaw 

correlations of the signal: if a=0.5, the signal is uncorrelated; if 

a>0.5 the correlations of the signal are persistent, where persistence 

means that a large (small) value (compared to the average) is more likely 

to be followed by a large (small) value; if a

Post Fire vegetation recovery estimation using satellite VEGETATION time series 287 

ular, we analysed the ten-day (decadal) maximum value of daily NDVI maps. 

The data were subjected to atmospheric corrections performed by CNES on 

the basis of the Simplified Method for Atmospheric Corrections (SMAC). 

Figure 1 - Location of study test sites: red colour indicates fire affected areas and green fire 

un-affected areas. 

We performed the DFA for all the pixels for each site. As an example figures 

2 show two test sites: Bolotana (fire affected) and Orsomarso (fire unaffected).

288 


Figure 2a - NDVI for Bolotana and Orsomarso.

Post Fire vegetation recovery estimation using satellite VEGETATION time series 289 

Figure - 2b, 2c. Upper fig. 2b shows the NDVId and fig. 2c show the values of the a coefficients 

for the two test sites. 

Results from all the investigated sites showed that the scaling exponents 

for fire-affected pixels range around the mean value of ~1.14, while those 

for fire-unaffected pixels vary around the mean value of ~0.77 (with t- 

Student test, p

AUTHOR INDEX 

A 

ABDALLAH C. pag. 45 

AGUADO I. pag. 133 

ALMOUSTAFA T.A. pag. 91 

ALONSO-BENITO A. pag. 209, 265 

AMATULLI G. pag. 33 

ANTONUCCI F. pag. 151 

ARBELO M. pag. 209 

ARCA A. pag. 75 

ARINO O. pag. 139, 233 

ARMAS R. pag. 139 

ARMENTERAS D. pag. 227 

ARMITAGE R.P. pag. 91 

ARNDT N. pag. 51, 57 

ARPACI A. pag. 51, 57 

ASNER G.P. pag. 221 

B 

BACCIU V. pag. 75 

BARRETO A. pag. 209 

BASILE G. pag. 151 

BERNA T. pag. 175 

BISQUERT M.M. pag. 109, 145 

BITELLI G. pag. 277 

BONORA L. pag. 203 

BOSCHETTI L. pag. 155 

BOSCHETTI M. pag. 121, 215 

BRIVIO P. A. pag. 121, 215 

BRUNDU G. pag. 75 

C 

CADAU E. pag. 193 

CALADO T.J. pag. 127 

CALLE A. pag. 161 

CAMIA A. pag. 33, 79, 85 

CARLÀ R. pag. 203 

CARRARA P. pag. 215 

CASADIO S. pag. 233 

CASANOVA J-L. pag. 161 

CASELLES V. pag. 109, 145 

CHUVIECO E. pag. 21, 133 

COLUZZI R. pag. 39, 95, 99, 

103, 151, 285 

CONESE C. pag. 203 

CONTE P. pag. 277 

CRUZ LÓPEZ I. pag. 181 

CSISZAR I. pag. 187 

D 

DACAMARA C.C. pag. 127, 237 

DANESE M. pag. 39 

DANSON F.M. pag. 15, 91 

DAVIDSON P. pag. 171 

DE LA RIVA J. pag. 115, 253, 259 

DE SANTIS A. pag. 95, 221 

DE SANTIS F. pag. 285 

DESMAZIÈRES Y. pag. 139 

DI BARTOLA C. pag. 193 

DIMITRAKOPOULOS K. pag. 69 

DRAXLER R. pag. 171 

DUCE P. pag. 75 

F 

FERRUCCI F. pag. 193 

FLANNIGAN M.D. pag. 25 

FLORSCH G. pag. 139 

291

292 

FORTUNATO G. pag. 191 

FRENCH N.H.F. pag. 265 

FROST P. pag. 167 

G 

GALIDAKI G. pag. 69 

GARCÍA-MARTÍN A. pag. 115 

GHENT D. pag. 63 

GIGLIO L. pag. 187 

GITAS I.Z. pag. 69 

GONZÁLEZ-ALONSO F. pag. 227 

GONZALEZ-CALVO A. pag. 209 

GOOSSENS R. pag. 271 

GOSSOW H. pag. 51 

GOUVEIA C. pag. 127, 237 

GOWMAN L.M. pag. 25 

GUARIGLIA A. pag. 103 

GUARINO A. pag. 193 

H 

HERNANDEZ-LEAL P.A. pag. 209 

HIM B. pag. 193 

HUESCA M. pag. 227 

I 

IAVARONE L. pag. 193 

J 

JOSEPHINE I.T. pag. 167 

JUSTICE C. pag. 187 

JUSTICE C. O. pag. 155 

K 

KERAMITSOGLOU I. pag. 139 

KNAPP D. pag. 221 

KONDRAGUNTA S. pag. 171 

KONTOES C. pag. 139 

KOUTSIAS N. pag. 243 

KUCERA J. pag. 247 

L 

LANEVE G. pag. 193 

LANORTE A. pag. 39, 95, 99, 

103, 151, 285 

LASAPONARA R. pag. 39, 95, 99, 

103, 151, 285 

LEGIDO J.L. pag. 109 

LEXER M.J. pag. 57 

LHERMITTE S. pag. 271 

LI P. pag. 173 

LOIZZO R. pag. 193 

LOPERTE G. pag. 151 

LÓPEZ SALDAÑA G. pag. 181 

LÓPEZ-SERRANO F.R. pag. 145 

M 

MALICO P. pag. 247 

MALLINIS G. pag. 69 

MANASSERO F. pag. 175 

MARTÍN M.P. pag. 85 

MATIUZZI, M. pag. 57 

MERINO DE MIGUEL S. pag. 227 

MOMBERG A. pag. 167 

MONTELEONE M. pag. 95 

MONTESANO T. pag. 95 

MONTORIO R. pag. 115 

MORENO M.V. pag. 21 

MORENO RUIZ J.A. pag. 265 

MOTA B. pag. 247 

MÜLLER M. pag. 57 

MURGANTE B. pag. 39 

N 

NIETO H. pag. 133 

NUNEZCASILLAS L. pag. 209 

O 

OLIVEIRA S. pag. 79 

OPAZO S. pag. 227 

P 

PAGANINI M. pag. 139 

PALACIOS V. pag. 115 

PAZ ANDRADE M.I. pag. 109 

PELLIZZARO G. pag. 75 

PEPE M. pag. 215

PEREIRA J.M.C. pag. 247 

PÉREZ-CABELLO F. pag. 115, 253, 259 

PETRUCCI B. pag. 215 

PLENIOU M. pag. 243 

POLYCHRONAKI A. pag. 69 

PRIOLO A. pag. 139 

R 

RAMPINI A. pag. 215 

RESSL R. pag. 181 

RIAÑO D. pag. 265 

RONGO R. pag. 193 

ROY D. P. pag. 155 

RUBIO E. pag. 145 

RUIZ RODRIGO P. pag. 51 

RUMINSKI M. pag. 171 

S 

SÁ A. pag. 139 

SALIS M. pag. 75 

SALVADOR P. pag. 161 

SÁNCHEZ J.M. pag. 109, 145 

SANDHOLT I. pag. 133 

SAN-MIGUEL AYANZ J. pag. 247 

SAN-MIGUEL J. pag. 33, 79 

SANTORO M. pag. 253, 259 

SANTURRI L. pag. 203 

SANZ J. pag. 161 

SCHROEDER W. pag. 187 

SIMKO J. pag. 171 

SPANO D. pag. 75 

SPESSA A. pag. 63 

STERGIOPOULOS I. pag. 69 

STROPPIANA D. pag. 121, 215 

T 

TANASE M. pag. 253, 259 

TELESCA L. pag. 285 

TRIGO R. pag. 237 

U 

USTIN S.L. pag. 265 

V 

VACIK H. pag. 51, 57 

VAUGHAN P.J. pag. 221 

VENTURA A. pag. 75 

VERAVERBEKE S. pag. 271 

VERSTRAETEN W. pag. 271 

VILAR L. pag. 85 

VOSLOO H. pag. 167 

W 

WEGMÜLLER U. pag. 253 

WOTTON B.M. pag. 25 

Z 

ZAFFARONIi P. pag. 215 

ZENG J. pag. 171 

293

Finito di stampare nel mese di agosto 2009 

da Il Segno Arti grafiche 

Soc. coop. sociale 

Potenza

ISBN 978-88-904367-0-3

7th Workshop on Forest Fire Management - EARSeL, European ...

Create successful ePaper yourself

Delete template?

Save as template?