E

Dg

EMJMa

b

c

d

e

f

a

AA

KFFGRLHFL

1

tiibecfisiia

0d

ARTICLE IN PRESSG ModelCOMOD-5342 No of Pages 13

Ecological Modelling xxx (2009) xxxndashxxx

Contents lists available at ScienceDirect

Ecological Modelling

journa l homepage wwwe lsev ier com locate eco lmodel

evelopment of a framework for fire risk assessment using remote sensing andeographic information system technologies

milio Chuviecoalowast Inmaculada Aguadoa Marta Yebra a Heacutector Nietoa Javier Salasa Pilar Martiacutenab Lara Vilarb Javier Martiacutenezb Susana Martiacutenc Paloma Ibarrad

uan de la Rivad Jaime Baezae Francisco Rodriacuteguez f Juan R Molina figuel A Herrera f Ricardo Zamoraf

Departamento de Geografiacutea Universidad de Alcalaacute Colegios 2 28801 Alcalaacute de Henares SpainInstituto de Economiacutea Geografiacutea y Demografiacutea (IEGD) Consejo Superior de Investigaciones Cientiacuteficas (CSIC) Albasanz 26-28 28037 Madrid SpainDepartamento de Economiacutea y Gestioacuten Forestal ETSI de Montes Ciudad Universitaria sn 28040 Madrid SpainDepartamento de Geografiacutea y Ordenacioacuten del Territorio Universidad de Zaragoza CPedro Cerbuna 12 50009 Zaragoza SpainFundacioacuten Centro de Estudios Ambientales del Mediterraacuteneo CCharles Darwin 14 Parque Tecnoloacutegico Paterna 46980 Valencia SpainDepartamento de Ingenieriacutea Forestal Universidad de Coacuterdoba Avd Meneacutendez Pidal sn 14071 Coacuterdoba Spain

r t i c l e i n f o

rticle historyvailable online xxx

eywordsire riskire dangereographic information systemsemote sensing

a b s t r a c t

Forest fires play a critical role in landscape transformation vegetation succession soil degradation andair quality Improvements in fire risk estimation are vital to reduce the negative impacts of fire either bylessen burn severity or intensity through fuel management or by aiding the natural vegetation recoveryusing post-fire treatments This paper presents the methods to generate the input variables and therisk integration developed within the Firemap project (funded under the Spanish Ministry of Science andTechnology) to map wildland fire risk for several regions of Spain After defining the conceptual scheme forfire risk assessment the paper describes the methods used to generate the risk parameters and presents

ogistic regressionuman factorsuel moisture contentightning

proposals for their integration into synthetic risk indices The generation of the input variables was basedon an extensive use of geographic information system and remote sensing technologies since the projectwas intended to provide a spatial and temporal assessment of risk conditions All variables were mappedat 1 km2 spatial resolution and were integrated into a web-mapping service system This service wasactive in the summer of 2007 for semi-operational testing of end-users The paper also presents the firstvalidation results of the danger index by comparing temporal trends of different danger components and

erent

fire occurrence in the diff

Introduction

Forest fires are a major factor of environmental transforma-ion in a wide variety of ecosystems (FAO 2007) Fires have globalmpacts (Chuvieco 2008) affecting forested areas and having anmportant share in greenhouse gas emissions Although fire haseen historically used as a tool for land use management and manycosystems are well adapted to fire cycles recent changes in bothlimate and societal factors related to fire can transform traditionalre regimes increasing the negative effects of fire upon vegetation

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

oils and human values In this regard the impact of climate warm-ng on increasing fire frequency and intensity has been documentedn several ecosystems (Kasischke and Turetsky 2006 Westerling etl 2006) Current climate projections point towards worse con-

lowast Corresponding authorE-mail address emiliochuviecouahes (E Chuvieco)

304-3800$ ndash see front matter copy 2008 Elsevier BV All rights reservedoi101016jecolmodel200811017

study regionscopy 2008 Elsevier BV All rights reserved

ditions in the next decades for most Tropical and Boreal regions(Flannigan et al 2005)

In addition to global effects fires have also important localeffects which are commonly associated to fire frequency andintensity which imply soil degradation soil erosion lost of livesbiodiversity and infrastructures (Omi 2005) Recent changes inland use management in developed countries with an increasingabandonment of traditional rural practices (Veacutelez 2004 Whitlock2004) have implied a remarkable increase of fuel accumulationwhich lead to more severe and intense fires and consequently tohigher negative impacts on soils and vegetation resilience

Within this environmental context the interest of having bet-ter tools for fire prevention and assessment should be emphasized

ework for fire risk assessment using remote sensing and geographicodel200811017

Fire risk evaluation is a critical part of fire prevention since pre-fireplanning resources require objective tools to monitor when andwhere a fire is more prone to occur or when it will have more neg-ative effects Traditional fire danger systems rely on meteorologicalindices based on variables that are routinely measured by weather

INE

2 al Mod

splti

sppidfisibim(

sinqwpasstafT(pttgp

2

2

idtaw

a

b

ARTICLEG ModelCOMOD-5342 No of Pages 13

E Chuvieco et al Ecologic

tations However atmospheric conditions are only one of the com-onents of fire risk which should also consider human aspects fuel

oads and moisture status as well as values at stake Modelling fuelrends have been proposed to analyze spatial and temporal changesn fire risk conditions (He et al 2004 Shang et al 2004)

Traditional fire terminology does not put a strong empha-is on potential damages of fire but rather on the ignition andropagation potential For instance FAO defines fire risk as ldquotherobability of fire starting determined by the presence and activ-

ties of causative agenciesrdquo Fire danger on the other hand isefined as considering ldquoboth fixed and variable factors of there environment that determine the ease of ignition rate ofpread difficulty of control and fire impact often expressed as anndexrdquo (httpwwwfaoorgforestrysitefiremanagementen) Inoth cases these terms do not clearly distinguish between the phys-

cal probability that a fire occurs and the potential damage that itay cause which is a common practice in natural hazards literature

Bachmann and Allgoumlwer 2001)Following the activities of the European project Spread a revi-

ion of the traditional fire danger systems was suggested byncluding explicitly the vulnerability aspects of fire which wereeglected in previous approaches (Chuvieco et al 2003) Conse-uently a new scheme for fire risk assessment was developedhich included two aspects fire danger (ignition or propagationotential) and vulnerability (potential damage) being the total riskproduct of the two The Spread fire risk scheme was designed to becale-independent and therefore applicable to both local and globalcales The system was not fully developed at the initial stage beinghe focus the development of methods for generating the input vari-bles Further process was required to develop consistent methodsor data integration and to extend the vulnerability componenthese two aspects were developed within another research projectnamed Firemap) and will be the main objective of this paper Theaper will present the conceptual scheme for fire risk assessmenthen it will briefly comment the methods that were used to generatehe risk variables thirdly it will propose techniques for data inte-ration and it will finally present the results of the initial validationrocess

Methods

1 Fire risk scheme

The fire risk assessment method that we proposed in this paper

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

s based on considering fire occurrence probability and potentialamages (Fig 1) The former is named fire danger throughouthis paper and considers the potential that a fire ignites or prop-gates The two main sources of ignition human and naturalere considered The former is undoubtedly the most important

Fig 1 Framework for fire risk assessment

PRESSelling xxx (2009) xxxndashxxx

worldwide (FAO 2007) but fires caused by lightning are also veryrelevant in some regions (Nieto et al in press) In addition to igni-tion sources the moisture status of plants was also consideredsince plants are the main ignition material in a forest fire Thepropagation component of fire danger was associated to the firespread potential which is a result of fuel amount and continu-ity plus favourable terrain and weather conditions (mainly windspeed)

The second group of fire risk conditions was associated tothe vulnerability component which is the assessment of poten-tial damages caused by the fire The negative effects of fire weredivided in three aspects socio-economic values (properties woodresources recreational importance carbon stocks etc) degrada-tion potential (soil and vegetation conditions) and landscape value(uniqueness conservation status etc)

To obtain an operational assessment of fire risk conditions fol-lowing the proposed scheme the following steps were required

Generation of risk factors using a common geographical unit Atarget scale and spatial resolution needs to be defined in relationto what sources of data are available

Conversion of risk factors to a common risk scale To integratethe input risk variables the original measurement scale of eachinput variable should be first converted to a common risk metricOtherwise we could not generate synthetic indices

c Development of criteria to integrate risk factors The differentinput variables have different impacts on fire risk conditionsIdentifying which are more relevant and how they should beweighed to generate synthetic indices is a critical phase in riskassessment

Since fire risk is a spatial and temporal process it should beaddressed both spatially and temporally The use of geographicinformation system (GIS) is quite obvious in this regard since thesetools are ideal to manage spatial information provide adequate spa-tial processing and visualization of results For this reason severalprevious studies on fire risk estimation have been based on GIS (Yoolet al 1985 Chuvieco and Congalton 1989 Chou 1992 Abhineetet al 1996 Chuvieco and Salas 1996 Castro and Chuvieco 1998Vasconcelos et al 2001 Nourbakhsh et al 2006) To reduce thetotal length of the paper the generation of the input variables willbe presented briefly and will refer to more extended publicationsfor details (Table 1)

22 Study regions

Several research groups working on Mediterranean conditionsparticipated in developing the Firemap project Four study areaswere finally selected to develop the methods of data generation andintegration (Fig 2) Three of them are autonomous regions of SpainAragon with 47719 km2 Madrid with 8028 km2 and Valencia with23255 km2 while the fourth is a province of Andalusia (Huelva10148 km2) Total area covered for the four regions is 89131 km2which accounts for 18 of the total area of Spain Following end-user recommendations the minimum mapping unit was fixed at1 km2 using as a reference the standard UTM grid The regions wereselected to provide a good assortment of Spanish various fire con-ditions For instance Aragon has the most important proportionof natural-caused fires and one of the lowest population densitiesin the country while Madrid has the highest population densityand it is the most urbanized region of Spain Huelva is still a rural-

ework for fire risk assessment using remote sensing and geographicodel200811017

oriented area but has a strong contrast between the coastal regionand the interior highlands Valencia suffers the largest forest firesas an average with an important tourist pressure in the coast andnotable forest resources in the interior Climatic and ecological char-acteristics are also quite diverse within the general characteristics

ARTICLE ING ModelECOMOD-5342 No of Pages 13

E Chuvieco et al Ecological Mod

Table 1Input factors for the fire risk assessment system

Factor Input data Method

Human (Vilar et al 2008) Historical occurrence Logistic regressionLightning (Nieto et al in

press)Demographic dataVegetationmdashDTM

Dead fuels moisturecontent (Aguado et al2007)

Meteorological data Linear regression analysis

Live fuels moisture content(Chuvieco et al 2004bGarcia et al 2008)

Satellite images Statistical fitting

Inversion of RTM

Propagation danger(Martiacuten Fernaacutendez et al2002)

Fuel type maps Behave Simulation

Meteorological data

Socio-economic values(Rodriacuteguez y Silva et al2007)

Forest maps Empirical modelsRecreational areasQuestionnaires

Degradation potential Soil maps Ecological modelsDigital terrain model Qualitative cross-tabulationClimatic dataVegetation mapsField studies

L

oait

2

2

ti

while the factor associated to recreational land use was approached

andscape value(Martiacutenez-Vega et al2007)

Protected areas Landscape patternLand cover

f Mediterranean areas Generally speaking Madrid and Aragonre more continental while Huelva and Valencia have more mar-time influences being Huelva more rainy and with milder summeremperatures than Valencia

3 Generation of risk variables

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

31 Modelling the human factors of fire ignitionIn most countries human activities are in one way or the other

he main responsible for fire ignition Humans have used fire histor-cally for different purposes light heat cooking land clearing etc

Fig 2 Location of th

PRESSelling xxx (2009) xxxndashxxx 3

and still have a critical impact on fire regimes and vegetation distri-bution (Pyne 1995) In Mediterranean areas human factors causemore than 90 of fires (Leone et al 2003) In Spain 961 of all firesare human-caused (Direccioacuten General de Biodiversidad 2006)

In spite of the importance of these human aspects little work hasbeen devoted to this issue maybe because of the complexity of pre-dicting human behaviour both in space and time Most frequentlythe studies have focused on variables related to land use or landuse-change (rural abandonment agriculturalndashforest interface orurbanndashforest interface) population trends rural activities poten-tial conflicts that may lead to vengeances or arson (unemploymentenforcement of conservation areas reforestation in traditional pas-tured areas etc) (Vega-Garciacutea et al 1995 Cardille et al 2001Leone et al 2003 Martiacutenez et al 2009)

The approach to consider human factors in fire risk assessmenthas been commonly based on statistical models which have triedto explain historical human-caused fire occurrence from a setof independent variables (Martell et al 1989 Chou et al 1993Chuvieco et al 2003 Martiacutenez et al 2004) Most commonly thosevariables are previously mapped at a similar spatial resolutionof the fire databases using a GIS Logistic regression analysis hasbeen frequently used for prediction and explanation of human-caused fire occurrence (Chou et al 1993 Vega-Garciacutea et al 1995Vasconcelos et al 2001 Martiacutenez et al 2009)

For this project the analysis of human risk conditions werefirstly based on selecting the critical variables associated to human-cause fires in Spain following a detailed reviewed of specializedliterature General factors commonly identified by previous studiesneeded to be approached using single variables which shouldbe available for all study sites For instance fires associated tonegligence or arson were approached by considering distancesto roads and railroads electric lines and military establishments

ework for fire risk assessment using remote sensing and geographicodel200811017

by the presence of urbanndashwildland interfaces hotels and cabinsand camping sites

In a second phase variables expressing each factor were mappedat the target spatial resolution of the fire risk assessment system

e study areas

ARTICLE IN PRESSG ModelECOMOD-5342 No of Pages 13

4 E Chuvieco et al Ecological Modelling xxx (2009) xxxndashxxx

Table 2Percent of correct classification of the human-cause prediction model for the different study areas

Madrid Valencia Huelva Aragon

ccurrence Low occurrence High occurrence Low occurrence High occurrence

A 924 764 820 914G 844 868

(stfoart

aTatagotettmmm

ilru

rc(dvrifvevvNVuoltat

Table 3Variables selected and marginal effects for the logistic model of human factors inthe Madrid region (significance level 005)

Selected variables Marginal effects (dxdy)

Buffer forest trails 0064Natural protected areas 0155Urbanndashwildland interfase 0190Change in rural population minus0105

TP

AG

Low occurrence High occurrence Low occurrence High o

ccuracy 754 657 794 574lobal accuracy 706 684

1 km2) using a wide variety of GIS analysis Following the generalcheme of the risk index only structural factors were considered athis stage since the human component was intended to be stableor the whole fire season Other authors have proposed daily indicesf human danger based on meteorological conditions (Martell etl 1987 Vega-Garciacutea et al 1995) but this approach would haveequired many variables that were not available for such a largeerritory covered in our study regions

Logistic Regression techniques were used to estimate the prob-bility of occurrence from socio-economic explanatory variableshe dependent variable was the number of fires caused by humanctivities in the period 1990ndash2004 derived from official fire statis-ics which reference fire records to a UTM 10 km times 10 km UTM gridnd to the municipality where the fire started To improve thiseoreferencing the 10 km grid and the municipality layers wereverlaid and then interpolated to the target 1 km2 grid using spa-ial techniques previously developed for fire applications (de la Rivat al 2004) Since the dependent variable is continuous to usehe logistic regression model the original variable was split intowo groups using the upper and lower third of ordered values The

iddle third was discarded at this point to avoid including inter-ediate values of occurrence but they were used for validating theodelPrevious to building the statistical model correlations between

ndependent variables were tested to avoid multicolinearity prob-ems The model was based on a forward stepwise logisticegression analysis In each study area 60 of the input cells weresed for model calibration and the remaining 40 for validation

Table 2 shows the results of the different models for each studyegion A 05 threshold value was used for classification of the inputases The number of correctly assigned cells varies from 684Valencia) to 868 (Aragon) As a general comment the low inci-ence of fires is better classified than the higher occurrence Theariables included in each model were in agreement with the expe-ience of the forest managers who participated in the project Fornstance in Madrid with a high presence of recreational activities inorest areas the urbanndashwildland interface was the most prominentariable in the model Table 3 illustrates for this study area the realffect of each independent variable in the variation of the responseariable (the lsquomarginal effectsrsquo after standardizing the independentariables) Urbanndashwildland interface is followed in importance byatural Protected Areas and unemployment rate In the region ofalencia the main explicative variable was the variation of the pop-lation followed by demographic potential (defined as a function

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

f a cellrsquos current population and its accessibility to other popu-ated cells) In the region of Huelva the obtained model shows thathe demographic potential the variation of the agrarian populationnd the buffer of roads in forest areas are the independent variableshat have more influence in the variation of the fire occurrence due

able 4ercent of correct classification of the lightning-cause prediction model for the different s

Madrid Aragon

No ignition Ignition No ignition Ign

ccuracy 684 727 710 670lobal accuracy 686 708

Farmers above 55 years minus0105Unemployment rate 0113Hotels minus0208

to human causes Finally in the Aragon region the most significantvariables were agriculturalndashforest interface land use change refor-estation electric lines in forested areas and common-lands in forestareas

232 Ignition potential from lightningIn spite of the lower importance of lightning over human fac-

tors for fire ignition lightning strikes are also an important factorto consider in fire danger estimation They tend to burn largerareas than human-caused fires because they occur in more iso-lated and steeper areas and frequently have various simultaneousignited spots and therefore are more difficult to control (Wottonand Martell 2005) Several previous studies have focused on ana-lyzing the geographical variables that are more prone to lightningcaused fires such as the topography (Diacuteaz-Avalos et al 2001) strikepolarity (Latham and Schlieter 1989) and fuel moisture content(Wotton and Martell 2005)

For this project the structural factors associated to historicallightning-caused fires were analyzed by comparing spatial pat-terns of affected and non-affected areas The dependent variablein this case was the number of lightning-caused fires during thelongest possible period of time (when both lightning sources andfire statistics are available) while the independent variables werethe total number of light strikes vegetation and terrain charac-teristics and moisture codes derived from the US National FireDanger Rating System (Bradshaw et al 1983) and the CanadianForest Fire Weather Index (Van Wagner 1987) The daily meteoro-logical database was only available at 3 km times 3 km resolution andfor the period of 2002ndash2004 to which the analysis was restricted

Similarly to the human factors a logistic regression model topredict and explain historical fire occurrence was derived for thelightning-caused fires In this case a pure binary variable was taken

ework for fire risk assessment using remote sensing and geographicodel200811017

into account (firenot fire) since the total number of fires was muchsmaller than the human-caused fires The outputs of the modelsshow good classification results with 70 of the cells correctly clas-sified (Table 4) The main explicative variable was the number of drystorms (with less than 2 mm)

tudy areas

Valencia Huelva

ition No ignition Ignition No ignition Ignition

696 651 810 600693 807

ARTICLE IN PRESSG ModelECOMOD-5342 No of Pages 13

E Chuvieco et al Ecological Modelling xxx (2009) xxxndashxxx 5

F adrid( utside

2

itofibf

flfietceee1ditgtdceditt

ig 3 Temporal evolution of dead FMC values during the summer of 2007 for the Mf) 1st September (White indicate areas that were not considered urban zones or o

33 Ignition potential associated to fuel moisture content statusFuel moisture content (FMC) is a critical variable to estimate

gnition and propagation danger since the amount of water inhe vegetation is inversely related to ignition potential and ratef spread (Nelson 2001) Following a common approach in forestre literature the estimation of FMC was divided in this projectetween dead and live components The former were estimatedrom meteorological variables and the later from satellite images

The estimation of FMC for dead materials lying on the forestoor (leaves branches and debris) is included in most operationalre danger rating systems (Camia et al 2003) It is most commonlystimated from meteorological variables since dead fuels changeheir water content in parallel to atmospheric conditions Weatherhanges affect the degree of water evaporation and absorptionspecially temperature rainfall and wind speed (Viney 1991) Thestimation of dead FMC for this project was performed from anmpirical approach based on field sampling developed between998 and 2003 in Central Spain (Aguado et al 2007) The indepen-ent variables in this case were two moisture codes routinely used

n fire danger estimation the Fine Fuel Moisture Code (FFMC) andhe 10-h code the former being part of the Canadian and US fire dan-er systems respectively Similar results were obtained from thewo moisture codes but finally the 10-h code was selected since itoes not require wind speed as an input and therefore it is easier toompute Once the empirical relations were established they were

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

xtended to a grid of 1 km times 1 km resolution interpolated from theata of the European Centre for Medium Range Weather Forecast-

ng (ECMWF)rsquos using local algorithms The interpolation algorithmook into account horizontal distances between the grid point andhe surrounding stations (quadratic inverse distance algorithm)

region (a) 15th June (b) 1st July (c) 15th July (d) 1st August (e) 15th August andthe study area)

The effect of altitude of each grid point over the value of the vari-able (temperature or humidity) was also considered (Aguado et al2007) The estimation of dead FMC was computed everyday basedon 12 (noon) forecasted data from the 8 am prediction (Fig 3)

Regarding the estimation of FMC of live species satellite remotesensing was used as an input The use of satellite data in live FMCestimation has been discussed by different authors in the last years(Chuvieco et al 2004b Danson and Bowyer 2004 Maki et al2004 Dennison et al 2005 Riano et al 2005 Stow et al 2005)In spite of the difficulty of extracting the influence of water absorp-tion over other factors affecting plant reflectance several studieshave found good relationships especially in grasslands and someshrub species Two approaches were used in this project one basedon empirical models for NOAA-AVHRR images using results fromprevious projects (Chuvieco et al 2004b) and the other one basedon simulation models for Terra-MODIS data (Yebra et al 2008) Theempirical method was found inappropriate for very dry years suchas 2005 when high overestimations were found Therefore a revi-sion of the empirical method was proposed The new functions tookinto account the rainfall conditions of the Spring season to choosewhether a dry or normal year equation should be applied The out-puts provide a more consistent estimation of FMC for contrastingyears than a single model (Garcia et al 2008)

The second approach to estimate FMC of live species was basedon the inversion of simulation models based on the radiative trans-

ework for fire risk assessment using remote sensing and geographicodel200811017

fer function (RTM Pinty et al 2004) The inputs were an 8-daycomposite of the first seven reflectance bands of MODIS (MOD09product (Vermote and Vermeulen 1999) as well as the vege-tation indices and the leaf area index product derived from thesame sensor MOD15 (Knyazikhin et al 1999) The performance of

INE

6 al Mod

RlwCpbo8tl

2

cueiccw

w1dg2taTcmm

apdmco

2

(tctpAta

ldquotTtiarmoout

mi

ARTICLEG ModelCOMOD-5342 No of Pages 13

E Chuvieco et al Ecologic

TM versus empirical models showed similar accuracies for grass-ands (root mean square error 245 and 274 respectively) and

orst accuracy for shrublands (2304 and 1430 respectively)onsidering the greater accessibility to AVHRR images and the gooderformance of the calibrated models finally the empirical modelased on these images was used in this project for the estimationf live FMC To avoid cloud coverage and off-nadir observations an-day compositing technique based on maximum daily tempera-ures was used (Chuvieco et al 2005) Therefore the estimation ofive FMC was updated every 8-day

34 Propagation potentialMost fire spread simulation models have been designed for local

onditions and for active fires that have occurred or have been sim-lated to occur For this project it was intended to produce anstimation of the average propagation potential of each cell assum-ng a fire may occur anytime in any cell of the study areas Anotherhallenge was that fire propagation values should be calculated foroarse grid cells since our model was addressed to regional scaleshich is uncommon in fire behaviour models

Within these two limitations average propagation conditionsere simulated using the Behave program (Andrews and Chase

990) A total of 5525 simulations were run for the 13 fuel typesefined within this program by modifying systematically the sloperadients from 0 to 90 and the wind speeds from 4 kmh to0 kmh Standard values of FMC were considered 5 for 1-h mois-ure fuels 10 for 10-h moisture fuels 12 for 100-h moisture fuelsnd 50 for live fuels (Veacutelez 2000 Martiacuten Fernaacutendez et al 2002)hose input conditions were selected by considering the worst-ase scenario that is the fire is potentially propagated along theaximum slope gradient and the wind speed is the average of theaximum speeds for the summer timeThe simulated values of flame length and rate of spread were

veraged for each fuel type and slope interval as to generate aotential propagation map of the study sites Fuel type models wereerived from the forest inventory maps by selecting the most com-on fuel in the target cell size of 1 km times 1 km Slope intervals were

omputed from the 250 m times 250 m digital terrain model of Spainriginally derived from 1200000 scale topographic maps

35 Socio-economic valuesIn the Firemap project topics associated to values at stake

vulnerability) were divided in two groups those associatedo economic and social factors and those related to ecologicalomponents The former were intended to evaluate what poten-ial damages from the fire could be related to losses of woodroducts hunting revenues recreational and tourist resourcesdditionally the potential economic impacts of carbon seques-

ration soil erosion and landscape conservation were taken intoccount

Different approaches were used for deriving each factor Thetangiblerdquo resources were evaluated using direct methods such ashe market price the age of the forest stand and the rotation lengthhe wood resources were assessed following a mixture procedurehat considers the American approach (only natural regenerations considered) and the European one (man-induced regenerations well) The ldquointangiblerdquo resources were evaluated using indi-ect methods such as the cost-travel and the contingent valuationethods The former has been used to assess the recreational value

f the landscape while the latter was the basis to evaluate the costf no-use and wildlife conservation of endangered species The val-

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

es associated to hunting and CO2 sinks were priced according tohe forest inventory

To illustrate the methods two examples can be comment inore detail one of the tangible resources and the other on the

ntangible ones An example of the tangible resource is the use of

PRESSelling xxx (2009) xxxndashxxx

the acorn of evergreen oaks which is an important component forfeeding high-quality Iberian pig species The sustainable use of thisresource requires computing the adequate density of animals perhectare which was based on the density of oaks and their sanitarystatus Once the overall production was estimated the potentialdamage of losing those resources by a forest fire was computedfollowing

V = PR((1 + i)(Tminuse) minus 1)

i(1 + i)(Tminuse)

where V is the assessment of potential losses P the production ofacorns (kg haminus1) R the price (D kgminus1) i the yearly rate of devaluationT the rotation for the oaks (in years) and e is the age of the speciesfor the year of the fire (in years)

Another example of the socio-economic vulnerability assess-ment refers to an intangible resource which is the valuationof recreational resources These resources were assessed usingthe travel-cost model (Riera-Font 2000) A demand function wasderived to account for the preferences of population to access dif-ferent natural areas The demand function was formulated as

Dij = f (Cij Rij Vij Eij Iij)

where Dij is the number of days where visitor ldquoirdquo goes to place ldquojrdquo Cijthe cost associated to move to ldquojrdquo for visitor ldquoirdquo R the rent of visitorldquoirdquo (according to four ranges) Vij the number of times that visitorldquoirdquo goes to place ldquojrdquo E a weighting factor on whether the visitor iswilling to pay a fee to enter the natural area ldquojrdquo and I is the quali-tative importance of forest areas for visitor ldquoirdquo (questionnaire scalefrom 1 to 4) The formula is additionally weighed according to thenumber of visitors in each natural area and provides a total esti-mation of economic interest for each cell of the study areas Woodresources were estimated from current cost of wood products anddifferent scenarios of potential fire intensity level Net productivityand reforestation costs were also considered

The economic assessment of all the resources considered inthe socio-economic vulnerability was included into a dedicatedgeographic information system Some of the variables were com-puted in quantitative terms (mostly in D haminus1) while others werecalculated in qualitative values (vulnerability categories) Obvi-ously the more intense the fire the more important the damagesand therefore the model considered also different fire behaviourscenarios Six fire intensity levels were considered and average con-ditions were considered for the final evaluation of socio-economicresources at stake

236 Degradation potentialThe vulnerability associated to ecological factors was focused on

the assessment of vegetation response to fire effects This responsewas set up for two different time periods short term (less than 1year) focused on identifying the most erodible areas and mediumterm (25 years) to identify changes in vegetation structure andcomposition caused by the fire As a result of both a syntheticindex of the degradation potential associated with fire was obtained(Alloza et al 2006) Since vulnerability evaluation needs to be donebefore a fire occurs no previous knowledge of fire characteristicsand post-fire climatology is available Consequently risk scenariosneed to be created In our case we chose the worst-case scenarioaccording to typical Mediterranean conditions the fire occurs insummer the fuel has low humidity and post-fire climatic conditionsare similar to the historical average

ework for fire risk assessment using remote sensing and geographicodel200811017

For the short-term evaluation the post-fire ecosystem responsecapacity was determined by physical environment characteristicsin terms of erodibility and characteristics of the affected vegetation(a comprehensive scheme of the evaluation process is included inFig 4)

ARTICLE IN PRESSG ModelECOMOD-5342 No of Pages 13

E Chuvieco et al Ecological Modelling xxx (2009) xxxndashxxx 7

ent o

bull

bull

Fig 4 Methodology diagram for the assessm

Erodibility refers to the potential erosion as a result of post-firevegetation loss In spite of the numerous modifications and criti-cism on the Universal Soil Loss Equation (USLE) structure it stillconstitutes a reference to assess the magnitude of soil loss inburnt areas (Giovannini 1999) Consequently the same factorsconsidered by the USLE (soil erodibility slope vegetation and cli-mate) were considered in our model Soil erodibility analysis wasbased on organic matter content surface structure and soil crust-ing risk The slope factor was quantified from the digital elevationmodels and the post-fire cover factor by estimating density andstructure of the vegetation communities from the National For-est map For the climate factor the Fournier index was used as anindicator of the climate erosive ability Due to data limitations aqualitative approach was finally selected by classifying erodibil-ity in three categories high medium and low sensitivity to fireeffectsVegetation response ability is critical to explain post-fire soil ero-sion since a minimum vegetation cover of 30ndash40 is commonlyaccepted as the limit protective role of vegetation against erosion(Francis and Thornes 1990) To approach vegetation responsethe post-fire ecological strategies of different functional groupswere considered like the resprout ability the seed bank per-sistency or the growth or dispersal ability (Lavorel et al 1999McIntyre et al 1999) To predict the response ability post-firereproductive strategy was considered as a predictive attributebased on the information available of long-term post-fire regen-eration patterns in Mediterranean forest (Baeza et al 2007 Baeza

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

and Vallejo 2008) Based on the National forest map the mainvegetation communities were grouped according to the verticalcomposition structure (trees andor shrubs) and the reproductivestrategy For each community a vulnerability value was assignedas the inverse of its response ability to the short-term effects

f ecological vulnerability in the short term

(eg seeder shrubland = very high resprouter shrubland = lowdeficient seeder tree covered + seeder shrubland = very highresprouter tree covered + mixed shrubland = medium vulnerabil-ity) The climatic limits to post-fire regeneration were based onhistorical water deficit indicators

The integration of the different components of post-fire short-term degradation potential was determined by soil erodibility andvegetation vulnerability and water limitations (Fig 4) Scenariosof fire intensity were estimated for the Rothermelrsquos standard fuelmodels (Anderson 1982) contrasted on experimental fires (Baezaet al 2002) and fire simulations carried out with the FARSITE firesimulator (Finney 1998) The final characterization was 1 8 9 = lowintensity 2 5 6 7 10 11 = medium intensity and 3 4 12 13 = highintensity

In the medium-term 25 years after the fire the affected veg-etation communityrsquos vulnerability was determined by its abilityto persist with no substantial changes (community structure spe-cific composition and relative presence of the species) Taking intoaccount the vegetation communitiesrsquo grouping carried out and thefire historical frequency ecological vulnerability in the mediumterm was rated seeder shrubland = medium resprouter shrub-land = low deficient seeder tree covered + seeder shrubland = highresprouter tree covered + mixed shrubland = low vulnerability Thesynthetic index of the degradation potential is obtained by quali-tative cross-tabulation between the short and medium term withfour categories (lowndashmoderatendashhighndashextreme)

ework for fire risk assessment using remote sensing and geographicodel200811017

237 Landscape valueLandscape value was the third component to account for fire

vulnerability Fire managers take into account the intrinsic qual-ity of the landscape to rank the pre-fire planning obviously along

INE

8 al Mod

wtbwK(pbta

pn(mm

mv2Ewvac

cwoe2

2

2

tort

otacitsmm(

ftaihgbc

vttab

ARTICLEG ModelCOMOD-5342 No of Pages 13

E Chuvieco et al Ecologic

ith other variables associated to human settlements and poten-ial life threads The evaluation of landscape characteristics haseen approached by many authors in the last years including aide range of criteria ecological (Kato et al 1997 Nakagoshi andondo 2002) aesthetic population preferences visual propertiesMartiacutenez-Vega et al 2003 Arriaza et al 2004) For the Firemaproject the consideration of landscape properties in fire vulnera-ility assessment was approached from a weighed combination ofhe intrinsic value of the landscape and the legal status of protectedreas (Martiacutenez-Vega et al 2007)

The intrinsic value of the landscape took into account commonrocedures in landscape evaluation considering degree of ldquounique-essrdquo proximity to the ldquonaturalrdquo conditions and pattern conditionsdiversity patch connectivity and mixture) The input variables for

easuring those landscape properties were the CORINE land coverap (Buumlttner et al 2000) and the potential vegetation map of SpainThe consideration of legal protection figures for each cell was

easured as whether it was within any of the designated conser-ation areas (National and Regional parks Natural reserves Nature000 selected areas sites of Community Importance and otheruropean conservation figures) Additionally the communal forestsere also considered Each protected area was evaluated for fire

ulnerability by local managers For the integration of single evalu-tion values a weighed sum based on firersquos experts knowledge wasomputed

Final results showed an important proportion of the study areasovered by high or very high vulnerability to wildland fires Areasithin the highest ranks covered 24 of Huelva 15 of Madrid 12

f Aragon and 11 of Valencia The integration models have beenvaluated qualitatively by the forest services during the summer of007 with satisfactory agreement with their own evaluations

4 Model integration

41 Creating a common danger scalesOnce the input risk variables were generated two additional

asks were required to obtain an integrated fire risk index On thene hand the input variables needed to be converted to a commonisk scale on the other hand they should be properly weighed sohe importance of the different factors was taken into account

Several methods have been proposed to find common scalesf fire risk being variable normalization qualitative categoriza-ion and probabilistic approaches the most common (Chuvieco etl 2003) Variable normalization generates a common scale byonverting each variable to a zero-one range using either the min-mum and maximum value or the mean and standard deviation ofhe input variable Qualitative groups imply to convert the originalcale to a categorical or ordinal one using categories such as lowedium and high risk Finally the probabilistic approach requires toodel the variables using any of the standard probability functions

normal Poisson Binary etc)For this project it was not possible to obtain a common risk scale

or all the input variables especially for the complexity to quantifyhe ecological vulnerability While further developments find anppropriate way to solve this problem as a first step vulnerabil-ty values were categorized in four ordinal groups low mediumigh and extreme This decision conditioned the rest of the inte-ration scheme since the socio-economic vulnerability needed toe expressed in similar categories Cross tabulation process wasarried out to obtain final ratings following end-users knowledge

Regarding the fire danger branch of the fire risk scheme all the

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

ariables were converted to a 0ndash1 scale using probability func-ions For the consideration of the causes (human and lightning)he estimation models were based on logistic regression analysisnd therefore the predictions were already expressed in proba-ilistic terms For the fuel moisture content the conversion of FMC

PRESSelling xxx (2009) xxxndashxxx

to ignition potential (IP 0ndash1 scale) was based on a physical modelusing the concept of moisture of extinction (ME Simard 1968)This value expresses the maximum moisture value above which afire is not sustained and differs for each fuel type The conversionfrom FMC to IP was based on a linear relation from the FMC mini-mum value found in the historical data series (PI = 1) to ME (PI = 02)(Chuvieco et al 2004a) Strictly speaking the ignition potential ofa fuel when FMC equals to ME should be 0 However in our projecta conservative approach was adopted and the probability of a fuelwith FMC equals to ME was set to 02 to avoid eliminating areaswith mixed fuels

Finally the conversion of the propagation variables (rate ofspread and flame length) to a propagation potential danger wasbased on a normalization procedure The normalization was basedon the cumulative proportion of both rate of spread and flamelength in all grid cells of the study areas For each cell the max-imum probability value between rate of spread and flame lengthwas selected as representative of the worst-case conditions

242 Integration of risk indicesOnce the risk variables have a common scale of danger they

can be combined in many different ways and using a wider rangeof techniques qualitative cross-tabulation multicriteria evalua-tion regression techniques or probabilistic models (Chuvieco et al2003) Different choices were made for this project

The integration of the causative agents (human and lightning)was based on the Kolmogorov probabilistic rule (Tarantola 2005)which indicates that the probability of two independent events canbe expressed as

P(A cup B) = P(A) + P(B) minus P(A)P(B)

where P(A cup B) is the integrated probability P(A) the probability ofignition derived from human variables and P(B) is the probabilityof ignition derived from lightning

The integration of live and dead FMC was performed by averag-ing both FMC ignition potential values weighted by the percentagecover of both dead and live fuels

For the integration of causative agents and FMC a multicrite-ria evaluation technique (Gomez-Delgado and Barredo-Cano 2006)was adopted It was assumed that high risk probability shouldbe associated to situations when both high probability of havingcausative agents and FMC ignition potential occur Assuming thatboth of these two variables are expressed in a Cartesian axis thedistance to the maximums should be a good indicator of risk con-ditions since that point expresses the highest probability of bothfactors (Fig 5a)

In the case of the integration between ignition and propagationdanger a similar approach was adopted although in this case itwas assumed that the worst conditions would occur either whenthe maximum ignition or propagation danger occur Therefore inthis case the maximum danger values should be those more dis-tant from the origin (Fig 5b) In both the integration of ignitiondanger components and between ignition and propagation dan-ger the dynamic factors (FMC) were weighed higher (four times)than the static factors (human lighting and propagation) as to bemore sensitive to variables than change rapidly

For vulnerability variables the criterion to convert the origi-nal quantitative scale of the socio-economic aspects and landscapevalues to a risk scale was based on qualitative weighing The finalintegration of the vulnerability component was based on four qual-

ework for fire risk assessment using remote sensing and geographicodel200811017

itative risk categories (lowndashmoderatendashhighndashextreme) as to putthose factors in relation to the soil degradation factor which wasalready expressed in these four categories A similar weigh wasapplied to the three factors considered (socio-economic degrada-tion potential and landscape value) since they were considered

ARTICLE IN PRESSG ModelECOMOD-5342 No of Pages 13

E Chuvieco et al Ecological Modelling xxx (2009) xxxndashxxx 9

Fig 5 The multicriteria evaluation axis for integrating ignit

Table 5Mahalanobis distances between fire and non-fire cells for different components ofthe fire risk scheme

Madrid Huelva Aragon Valencia

II

tcw

2

motwhT(timca

on ignition points collected within 4 months of daily data The total

TR

I

I

I

TL

I

I

I

gnition danger 0276368 037556 0191812 0109428ntegrated danger 0232113 0388367 0149574 0088152

o have a similar impact in the estimation of potential damagesaused by fires Additional scenarios will be considered in futureorks

5 Development of a dedicated web-mapping service

The Firemap project was intended to develop operationalethodologies for fire managers Therefore the participation

f end-users was always encouraged To facilitate this par-icipation a dedicated web mapping service was developedithin the project using public domain software (mapserverttpmapservergisumnedu last accessed 22 October 2008)he final server was tested during the fire season of 2007June to September) and it was successfully reported byhe end-users It included all the input risk variables and

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

ntegrated indices plus several vector variables as auxiliary infor-ation Zoom roam consults and download facilities were

reated (httpwwwgeograuahes8080cartofireindexphp lastccessed on 22 October 2008)

able 6esults of the MannndashWhitney U-test Significance of differences between fire and non-fir

ndex Statistics Madrid

gnition danger MannndashWhitney U-test 73428174Z minus6424Significance (two-tailed) 000

ntegrated danger MannndashWhitney U-test 76460728Z minus5967Significance (two-tailed) 000

able 7ogistic regression analysis for the validation exercise

ndex Madrid

gnition danger R2L

014Hosmer and Lemeshow test 892Coefficient (significance) 2976 (000)

ntegrated danger R2L

013Hosmer and Lemeshow test 834Coefficient (significance) 3432 (000)

ion danger (a) probability of ignition (b) fire danger

26 Validation

Two types of assessment should be considered in a fire riskframework the validation of the input risk variables and the val-idation of the final risk indices The former assessment should beassociated to the actual variable rather than to the fire occurrencevalues For instance the FMC estimation should be assessed againstFMC field measurements and not in relation to fire statistics sinceFMC is not the only factor affecting fire ignition or propagation Infact even with very low values of FMC fires will not occur in theabsence of a causative agent Consequently validating FMC withfire statistics may be misleading

However the assessment of integrated indices should be basedon fire statistics since an integrated index should consider the mainfactors of risk and therefore should properly predict fire ignitionandor propagation Since fire occurrence changes in space andtime the validation of integrated indices should be done with longtime series because short periods may bias some of the theoreticalassumptions that are required to build the model In spite of this afirst approximation of validating fire risk indices may be based onshorter time periods when enough spatial samples are availableIn this case the fire risk server was server for the summer of 2007in the four study regions The first assessment is therefore based

ework for fire risk assessment using remote sensing and geographicodel200811017

sample was more than 7 million observations (60000 cells of 1 km2

times 120 days)This preliminary assessment was focused on evaluating the exis-

tence of significant differences between the risk values of fire and

e cells

Huelva Aragon Valencia

56640631 504403051 223141072minus6358 minus6155 minus4455000 000 000

55444348 523491786 231831563minus6609 minus5489 minus3781000 000 000

Huelva Aragon Valencia

018 008 005

414 025 0862453 (000) 3207 (000) 1820 (000)

018 007 004

489 043 3232657 (000) 3769 (000) 1758 (000)

INE

1 al Mod

ntatbies

ARTICLEG ModelCOMOD-5342 No of Pages 13

0 E Chuvieco et al Ecologic

o fire cells For validation purposes two indices were consideredhe ignition danger which included human and lightning factorsnd the integrated danger which included ignition and propaga-ion danger The vulnerability components were not considered

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

ecause they are not related to fire occurrence but rather to thempacts of fire once it occurs They could not be assessed sincestimations of fire effects are not routinely collected by official firetatistics

Fig 6 Box graphs showing differences in the ignition danger fo

PRESSelling xxx (2009) xxxndashxxx

The assessment was based on fire reports generated by theregional forest fire services involved in the Firemap project Igni-tion points extracted from GPS survey were available for most of thefires as well as the starting date and time and burned area Total

ework for fire risk assessment using remote sensing and geographicodel200811017

ignition points that were used for validation were 173 in Madrid111 in Huelva 188 in Aragon and 158 in Valencia

Several statistics to estimate significance of differences betweenfire and non-fire cells were computed (1) the distances of Maha-

r cells with and without fires during the summer of 2007

INE

al Mod

l(eu

gs

ARTICLEG ModelCOMOD-5342 No of Pages 13

E Chuvieco et al Ecologic

anobis (2) the MannndashWhitney U-test (Mann and Whitney 1947)3) the Nagelkerke R2 coefficient from logistic regression fittings for

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

ach integrated index (Andrews et al 2003) Processing was donesing the R statistical software (R Development Core Team 2007)

Tables 5ndash7 show the results of this validation The two inte-rated indices showed similar values in the regions althoughome discrepancies were observed Ignition danger generally shows

Fig 7 Box graphs showing differences in the integrated danger f

PRESSelling xxx (2009) xxxndashxxx 11

higher Mahalanobis distances than integrated Danger However theresults were very close between the two in Huelva (with the highest

ework for fire risk assessment using remote sensing and geographicodel200811017

values among the different regions) and Valencia (the worst) TheU values confirm those results since both Ignition danger and Inte-grated danger provide significant differences in all study regionsThe results are poorer for Valencia than for other regions The logis-tic regression analysis showed similar trends as the U test which

or cells with and without fires during the summer of 2007

INE

1 al Mod

sptdn

3

eatwcairatrcecTatr

fmviitttSoavoor

sfihafaAiitaid

A

SErv

ARTICLEG ModelCOMOD-5342 No of Pages 13

2 E Chuvieco et al Ecologic

ignificant differences for all regions and indices The two risk com-onents show higher values for fire cells than for non-fires showinghe potential of these indices to predict fire occurrence in veryifferent regions although a wide diversity of risk values withinon-fire areas was observed (Figs 6 and 7)

Discussion and conclusions

The paper has proposed an integration framework for fire riskvaluation The system was based on two groups of factors thosessociated to the probability that a fire occurs and those related tohe potential damages of fires Within the former causative agentsere considered as well as fuel moisture status and propagation

onditions This terminology is quite innovative in the fire riskssessment literature which has traditionally relied on monitor-ng weather conditions Most operational fire danger systems areestricted to meteorological danger indices (San Miguel-Ayanz etl 2003) The potential changes in fire danger conditions associatedo climate warming may be easily estimated using these meteo-ological danger indices based on the different climate scenariosurrently available (Gillett et al 2004) Recent papers have mod-lled spatial and temporal changes in fuel characteristics leading tohanges in fire risk conditions (He et al 2004 Shang et al 2004)hese prototypes are based on modelling biophysical conditionsnd fire history leading to fuel accumulation They are very usefulo predict future scenarios and propose fuel treatments for fire riskeduction

Our model does not provide yet the capacity of modellinguture conditions but it is more comprehensive than the for-

er approaches since it includes socio-economic aspects andulnerability factors Although most fire managers recognize themportance of human factors there are not operational systemsncluding this component either by lack of input information orhe difficult integration between socio-economic and weather fac-ors This paper has addressed this issue and proposed mechanismso integrate human characteristics into integrated risk indicesimilarly the vulnerability aspects have not been considered inperational fire danger indices but they are a relevant part ofssessment systems for others natural hazard (earthquakes floodsolcano eruptions etc) The potential damages associated to fireccurring in a particular area and period should lead the decisionsn fire suppression by prioritizing those areas with more valuableesources at stake

The limited assessment that has been available for this study hashown significant differences between two integrated indices andre ignition in four study regions located in Spain These regionsave different ecological and fire conditions and may be consideredrepresentative sample of the potentials of these indices However

urther assessment is required work is required in other regionsnd periods to check consistency and generalization potentialdditional work is also needed to improve procedures of data

ntegration and sensitivity analysis of input variables in the finalntegration In spite of those limitations the scheme proposed inhis paper should provide a sound procedure to obtain syntheticnd spatially explicit assessment of fire risk conditions to helpmproving pre-fire management and to take more appropriateecisions about rehabilitation of areas affected by wildland fires

cknowledgements

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

The Firemap project was funded by the Spanish Ministry ofcience and Education (CGL2004-060490C04-01CLI) through thenvironment and Climate program Very useful comments wereeceived from end-users of the project Civil Protection Forest ser-ices in Madrid Aragon Valencia and Andalusia regions

PRESSelling xxx (2009) xxxndashxxx

References

Abhineet J Ravan SA Singh RK Das KK Roy PS 1996 Forest fire risk modellingusing remote sensing and geographic information system Current Science 70928ndash933

Aguado I Chuvieco E Boren R Nieto H 2007 Estimation of dead fuel moisturecontent from meteorological data in Mediterranean areas Applications in firedanger assessment IJWF 16 390ndash397

Alloza JA Baeza MJ De la Riva J Duguy B Echeverriacutea MT Ibarra P Llovet JPeacuterez-Cabello F Rovira P Vallejo VR 2006 A model to evaluate the ecologicalvulnerability to forest fires in Mediterranean ecosystems In Viegas DX (Ed)Proceedings of the 5th Internartional conference on Forest Fire Research (vol GS203-12) Elsevier Coimbra

Anderson HE 1982 Aids to determining fuel models for estimating fire behaviorUSDA Forest Service General Technical Report INT-122 Ogden UT

Andrews PL Chase CH 1990 The BEHAVE fire behavior prediction system TheCompiler 8 4ndash9

Andrews PL Loftsgaarden DO Bradshaw LS 2003 Evaluation of fire dangerrating indexes using logistic regression and percentile analysis IJWF 12 213ndash226

Arriaza M Canas-Ortega JF Canas-Madueno JA Ruiz-Aviles P 2004 Assess-ing the visual quality of rural landscapes Landscape and Urban Planning 69115ndash125

Bachmann A Allgoumlwer B 2001 A consistent wildland fire risk terminology isneeded Fire Management Today 61 28ndash33

Baeza MJ Luis MD Raventos J Escarre A 2002 Factors infuencing fire behaviourin shrublands of different stand ages and the implications for using prescribedburning to reduce wildfire risk Journal of Environmental Management 65199ndash208

Baeza MJ Valdecantos A Alloza JA Vallejo R 2007 Human disturbance andenvironmental factors as drivers of long-term post-fire regeneration patterns inMediterranean forests Journal of Vegetation Science 18 243ndash252

Baeza MJ Vallejo R 2008 Vegetation recovery after fuel management in Mediter-ranean shrublands Applied Vegetation Science 11 151ndash158

Bradshaw L Deeming J Burgan RE Cohen J 1983 The 1978 National Fire-DangerRating System Technical Documentation No GTR INT -169 USDA Forest ServiceOgden Utah

Buumlttner G Feranec J Jaffrain G 2000 Corine land cover update 2000 In TechnicalGuidelines Europan Environmental Agency Copenhagen

Camia A Leblon B Cruz M Carlson JD Aguado I 2003 Methods used to esti-mate moisture content of dead wildland fuels In Chuvieco E (Ed) WildlandFire Danger Estimation and Mapping The Role of Remote Sensing Data WorldScientific Publishing Singapore pp 91ndash117

Cardille JA Ventura SJ Turner MG 2001 Environmental and social factors influ-encing wildfires in the Upper Midwest United States Ecological Applications 11111ndash127

Castro R Chuvieco E 1998 Modeling forest fire danger from geographic informa-tion systems GI 13 15ndash23

Chou YH 1992 Management of wildfires with a geographical information systemIJGIS 6 123ndash140

Chou YH Minnich RA Chase RA 1993 Mapping probability of fire occur-rence in San Jacinto Mountains California USA Environmental Management17 129ndash140

Chuvieco E 2008 Satellite observation of biomass burning implications in globalchange research In Chuvieco E (Ed) Earth Observation and Global ChangeSpringer New York pp 109ndash142

Chuvieco E Allgoumlwer B Salas FJ 2003 Integration of physical and human factorsin fire danger assessment In Chuvieco E (Ed) Wildland Fire Danger Estimationand Mapping The Role of Remote Sensing Data World Scientific PublishingSingapore pp 197ndash218

Chuvieco E Aguado I Dimitrakopoulos A 2004a Conversion of fuel moisture con-tent values to ignition potential for integrated fire danger assessment CanadianJournal of Forest Research 34 (11) 2284ndash2293

Chuvieco E Cocero D Riano D Martiacuten MP Martiacutenez-Vega J de la Riva JPeacuterez F 2004b Combining NDVI and surface temperature for the estima-tion of live fuel moisture content in forest fire danger rating RSE 92 322ndash331

Chuvieco E Congalton RG 1989 Application of remote sensing and geographicinformation systems to forest fire hazard mapping RSE 29 147ndash159

Chuvieco E Salas FJ 1996 Mapping the spatial distribution of forest fire dangerusing GIS IJGIS 10 333ndash345

Chuvieco E Ventura G Martiacuten MP Gomez I 2005 Assessment of multitemporalcompositing techniques of MODIS and AVHRR images for burned land mappingRSE 94 450ndash462

Danson FM Bowyer P 2004 Estimating live fuel moisture content from remotelysensed reflectance RSE 92 309ndash321

de la Riva J Peacuterez-Cabello F Lana-Renault N Koutsias N 2004 Mapping wildfireoccurrence at regional scale RSE 92 363ndash369

Dennison PE Roberts Dar A Peterson SH Rechel J 2005 Use of Normal-ized Difference Water Index for monitoring live fuel moisture content IJRS 26

ework for fire risk assessment using remote sensing and geographicodel200811017

1035ndash1042Diacuteaz-Avalos C Peterson DL Alvarado E Ferguson SA Besag JE 2001

Spacendashtime modelling of lightning-caused ignitions in the Blue Mountains Ore-gon Canadian Journal of Forest Research 31 1579ndash1593

Direccioacuten General de Biodiversidad 2006 Los incendios forestales en Espana Min-isterio de Medio Ambiente Madrid

INE

al Mod

F

F

F

F

G

G

G

G

H

K

K

K

L

L

L

M

M

M

M

M

M

M

M

M

M

N

Yebra M Chuvieco E Riano D 2008 Estimation of live Fuel Moisture Content from

ARTICLEG ModelCOMOD-5342 No of Pages 13

E Chuvieco et al Ecologic

AO 2007 Fire Management-Global Assessment 2006 A Thematic Study Preparedin the Framework of the Global Forest Resources Assessment 2005 FAO Rome

inney MA 1998 FARSITE Fire Area Simulator - Model development and evalua-tion USDA Forest Service RMRS-RP-4 51 pp

lannigan MD Logan KA Amiro BD Skinner WR Stocks BJ 2005 Future areaburned in Canada Climatic Change 72 1ndash16

rancis C Thornes JB 1990 Runoff hydrograms from three mediterranean vegeta-tion cover types In Thornes JB (Ed) Vegetation and Erosion Wiley Chichesterpp 363ndash384

arcia M Aguado I Chuvieco E 2008 Combining AVHRR and Meteorological Datafor Estimating Live Fuel Moisture Content in Forest Fire Danger Rating RSE enprensa

illett NP Weaver AJ Zwiers FW Flannigan MD 2004 Detecting the effect ofclimate change on Canadian forest fires Geophysical Research Letters 31 L18211doi182101102912004GL020876

iovannini G 1999 Post-fire soil erosion risk How to predict and how prevent InProceedings of the Advanced Study Course on Wildfire Management EuropeanCommission Athens pp 305ndash321

omez-Delgado M Barredo-Cano JI 2006 Sistemas de informacion geografica yevaluacion multicriterio en la ordenacion del territorio RA-MA Madrid

e HS Shang BZ Crow TR Gustafson EJ Shifley SR 2004 Simulating for-est fuel and fire risk dynamics across landscapesmdashLANDIS fuel module designEcological Modelling 180 135ndash151

asischke ES Turetsky MR 2006 Recent changes in the fire regime across theNorth American boreal regionmdashspatial and temporal patterns of burning acrossCanada and Alaska Geophysical Research Letters 33 1ndash5

ato Y Yokohari M Brown RD 1997 Integration and visualization of the ecolog-ical value of rural landscapes in maintaining the physical environment of JapanLandscape and Urban Planning 39

nyazikhin Y Glassy J Privette JL Tian Y Lotsch A Zhang Y Wang Y MorisetteJT Votava P Myneni RB Nemani RR Running SW 1999 MODIS leafarea index (LAI) and fraction of photosynthetically active radiation absorbedby vegetation (FPAR) product (MOD15) Algorithm Theoretical Basis Documenthttpeospsogsfcnasagovatbdmodistableshtml

atham DJ Schlieter JA 1989 Ignition probabilities of wildland fuels based onsimulated lightning discharges No Research Paper INT-411 US Department ofAgriculture Forest Service Intermountain Research Station Ogden UT

avorel S McIntyre S Grigulis K 1999 Plant response to disturbance in a Mediter-ranean grassland how many functional groups Journal of Vegetation Science10 661ndash672

eone V Koutsias N Martiacutenez J Vega-Garciacutea C Allgoumlwer B Lovreglio R 2003The human factor in fire danger assessment In Chuvieco E (Ed) WildlandFire Danger Estimation and Mapping The Role of Remote Sensing Data WorldScientific Publishing Singapore pp 143ndash196

aki M Ishiahra M Tamura M 2004 Estimation of leaf water status to monitorthe risk of forest fires by using remotely sensed data RSE 90 441ndash450

ann HB Whitney DR 1947 On a test of whether one of two random vari-ables is stochastically larger than the other Annals of Mathematical Statistics 1850ndash60

artell DL Bevilacqua E Stocks BJ 1989 Modelling seasonal variation in dailypeople-caused forest fire occurrence Canadian Journal of Forest Research 191555ndash1563

artell DL Otukol S Stocks BJ 1987 A logistic model for predicting daily people-caused forest fire occurrence in Ontario Canadian Journal of Forest Research 17394ndash401

artiacuten Fernaacutendez S Martiacutenez Falero E Peacuterez JM 2002 Optimization of resourcesmanagement in wildfire combat Environmental Management 30 336ndash352

artiacutenez-Vega J Martiacuten MP Romero R 2003 Valoracioacuten del paisaje en la Zonade Especial Proteccioacuten de Aves Carrizales y Sotos de Aranjuez (Comunidad deMadrid) GeoFocus 3 1ndash21

artiacutenez-Vega J Romero R Echavarriacutea P 2007 Valoracioacuten paisajiacutestica y ecoloacutegicade la Comunidad de Madrid su integracioacuten en un iacutendice sinteacutetico de riesgo deincendios forestales Revista de Teledeteccioacuten 28 43ndash60

artiacutenez J Chuvieco E Martin MP 2004 Estimating human risk factors in wild-land fires in Spain using logistic regression II International Symposium on FireEconomics Planning and Policy A Global Vision University of Cordoba CD-RomCoacuterdoba p 15

artiacutenez J Vega-Garciacutea C Chuvieco E 2009 Human-caused wildfire risk ratingfor prevention planning in Spain Journal of Environmental Management 90

Please cite this article in press as Chuvieco E et al Development of a framinformation system technologies Ecol Model (2009) doi101016jecolm

1241ndash1252cIntyre S Lavorel S Landsberg J Forbes TDA 1999 Disturbance response in

vegetationmdashtowards a global perspective on functional traits Journal of Vegeta-tion Science 10 621ndash630

akagoshi N Kondo T 2002 Ecological land evaluation for nature redevelopmentin river areas Landscape Ecology 17 (Suppl 1) 83ndash93

PRESSelling xxx (2009) xxxndashxxx 13

Nelson RM 2001 Water relations of forest fuels In Johnson EA Miyanishi K(Eds) Forest Fires Behavior and Ecological Effects Academic Press San DiegoCA pp 79ndash149

Nieto H Aguado I Chuvieco E in press Lightning-caused fires in Central Spaindevelopment of short-term and long-term probability models of occurrence intwo regions of Spain IJWF

Nourbakhsh I Sargent R Wright A Cramer K McClendon B Jones M 2006Mapping disaster zones Nature 439 787ndash788

Omi PN 2005 Forest Fires A Reference Handbook ABC-CLIO Santa Barbara CAxviii 347 pp

Pinty B Widlowski JL Taberner M Gobron N Verstraete MM Disney MGascon F Gastellu JP Jiang L Kuusk A 2004 Radiation Transfer ModelIntercomparison (RAMI) exercise results from the second phase Journal ofGeophysical Research 109 D06210 doi062100102902003JD004252

Pyne SJ 1995 World fire In The Culture of Fire on Earth University of WashingtonPress Seattle and London

R Development Core Team 2007 R A Language and Environment for StatisticalComputing R Foundation for Statistical Computing Vienna

Riano D Vaughan P Chuvieco E Zarco-Tejada P Ustin SL 2005 Estimationof fuel moisture content by inversion of radiative transfer models to simulateequivalent water thickness and dry matter content analysis at leaf and canopylevel IEEE Transactions on Geoscience and Remote Sensing 43 819ndash826

Riera-Font A 2000 Mass tourism and the demand for protected natural areas atravel cost approach Journal of Environmental Economics and Management 3997ndash116

Rodriacuteguez y Silva F Molina Martiacutenez JR Herrera Machuca M Zamora Diacuteaz R2007 Vulnerabilidad socioeconoacutemica de los espacios forestales frente al impactode los incendios aproximacioacuten metodoloacutegica mediante sistemas de informacioacutengeograacuteficos (proyecto Firemap) In IV International Wildland Fire ConferenceProceedings Sevilla

San Miguel-Ayanz J Carlson JD Alexander M Tolhurst K Morgan GSneeuwjagt R Dudley M 2003 Current methods to assess fire danger poten-tial In Chuvieco E (Ed) Wildland Fire Danger Estimation and MappingThe Role of Remote Sensing Data World Scientific Publishing Singaporepp 21ndash61

Shang BZ He HS Crow TR Shifley SR 2004 Fuel load reductions and fire risk incentral hardwood forests of the United States a spatial simulation study designEcological Modelling 180 89ndash102

Simard AJ 1968 The moisture content of forest fuels-A review of the basic conceptsInformation Report No FF-X-14 Forest Fire Research Institute Ottawa Ontario

Stow D Niphadkar M Kaiser J 2005 MODIS-derived visible atmospherically resis-tant index for monitoring chaparral moisture content IJRS 26 3867ndash3873

Tarantola A 2005 Inverse Problem Theory and Methods for Model Parameter Esti-mation Society for Industrial and Applied Mathematics Philadelphia 342 pp

Van Wagner CE 1987 Development and structure of the Canadian Forest FireWeather Index System Forestry Technical Report No 35 Canadian Forest Ser-vice Otawa

Vasconcelos MJP Silva S Tomeacute M Alvim M Pereira JMC 2001 Spatial pre-diction of fire ignition probabilities comparing logistic regression and neuralnetworks PERS 67 73ndash83

Vega-Garciacutea C Woodard T Adamowicz Lee B 1995 A logit model for predictingthe daily occurence of human caused forest fires IJWF 5 101ndash111

Veacutelez R 2000 La defensa contra incendios forestales Fundamentos y experienciasMcGraw-Hill Interamericana de Espana SAU Madrid 1281 pp

Veacutelez R 2004 Europe Development and fire II International Symposium on FireEconomics Planning and Policy A Global Vision University of Cordoba CD-RomCoacuterdoba 6 pp

Vermote EF Vermeulen A 1999 Atmospheric Correction Algorithm SpectralReflectances (MOD09) NASA

Vilar L MartinIsabel MP MartiacutenezVega FJ 2008 Empleo de teacutecnicas de regresioacutenlogiacutestica para la obtencioacuten de modelos de riesgo humano de incendio forestal aescala regional Boletiacuten de la Asociacioacuten de Geoacutegrafos Espanoles 47 5ndash29

Viney NR 1991 A review of fine fuel moisture modelling IJWF 1 215ndash234Westerling AL Hidalgo HG Cayan DR Swetnam TW 2006 Warming and ear-

lier spring increase western US forest wildfire activity Science 313 940ndash943Whitlock C 2004 Land management-Forests fires and climate Nature 432 28ndash29Wotton BM Martell DL 2005 A lightning fire occurrence model for Ontario

Canadian Journal of Forest Research 35 1389ndash1401

ework for fire risk assessment using remote sensing and geographicodel200811017

MODIS images for fire risk assessment Agricultural and Forest Meteorology 148523ndash536

Yool SR Eckhardt DW Estes JE Cosentino MJ 1985 Describing the brushfirehazard in southern California Annals of the Association of American Geogra-phers 75 417ndash430