- Email: [email protected]
Structural fire risk: The case of Portugal
Science of the Total Environment 573 (2016) 883–893
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Structural fire risk: The case of Portugal Joana Parente a,⁎, Mário G. Pereira a,b a b
Centre for Research and Technology of Agro-Environment and Biological Sciences, CITAB, University of Trás-os-Montes and Alto Douro, UTAD, Quinta de Prados, 5000-801 Vila Real, Portugal Instituto Dom Luiz, IDL, Faculdade de Ciências da Universidade de Lisboa, Campo Grande, Edifício C8, Piso 3, 1749-016, Lisboa, Portugal
H I G H L I G H T S
G R A P H I C A L
A B S T R A C T
• Verde and Zêzere susceptibility map was updated with longer and high resolution data. • Susceptibility model's performance and particularities were assessed and discussed. • Structural fire risk computed for updated potential damage and three hazard scenarios • Potential damage mapped with recent vulnerability and Horizon 3030 economic values • Higher fire risk in central, north and extreme south (Algarve) of Portugal
a r t i c l e
i n f o
Article history: Received 16 July 2016 Received in revised form 22 August 2016 Accepted 22 August 2016 Available online xxxx Editor: D. Barcelo Keywords: Vegetation fires Structural fire risk Susceptibility map Potential economic damage
a b s t r a c t This study is focused in mapping the structural fire risk in the vegetated area of Portugal using a deterministic approach based on the concept of fire risk currently accepted by the scientific community which consists in the combination of the fire hazard and the potential economic damage. The fire susceptibility map of Verde and Zêzere (2010) was adopted and updated by the use of a higher resolution digital terrain model, longer burnt area perimeter dataset (1975–2013) and the entire set of Corine land cover inventories. The susceptibility was mapped with five classes to be in accordance with the Portuguese law and the results confirms the good performance of this model not only in the favourability scores but also in the predictive values. Three different scenarios of (maximum, mean, and minimum annual) burnt area were consider to estimate the fire hazard which allow to estimate the likelihood of a pixel to burn (ranging between 0% and 20%) depending on the class to which it belongs and on the future burnt area scenario. The potential economic damage was estimated with the vulnerability scores and monetary values of species defined in the literature and by law. The obtained fire risk map identifies the areas more prone to be affected by fires in the future and provides an estimate of the economic damage of the fire which will be a valuable tool for forest and fire managers and to minimize the economic and environmental consequences of vegetation fires in Portugal. © 2016 Elsevier B.V. All rights reserved.
1. Introduction
⁎ Corresponding author at: Universidade de Trás-os-Montes e Alto Douro, Quinta de Prados, 5000-801 Vila Real, Portugal. E-mail addresses: [email protected] (J. Parente), [email protected] (M.G. Pereira).
http://dx.doi.org/10.1016/j.scitotenv.2016.08.164 0048-9697/© 2016 Elsevier B.V. All rights reserved.
Vegetation fires are part of the natural and human systems and, independently of their causes (lightning, accidental, negligent or intentional), may have a highly destructive effect on the ecosystems affecting its ability to recover, causing major environmental problems,
884
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
threaten human lives, activities and infrastructures (Blanchi et al., 2002; Chuvieco, 2009; Jaiswal et al., 2002). Along with the burning/removal of vegetation, fires have very important impacts on soil food web structure and carbon flow (Shaw et al., 2016), microbiological properties and enzymatic activities (Hedo et al., 2015), development of an ash cover (Pereira et al., 2015b), water repellency (Keesstra et al., 2016), terrestrial branch of the hydrological cycle (Pereira et al., 2016a; Van Eck et al., 2016), water quality (Pacheco et al., 2015; Santos et al., 2015a,b) and erosion (Cerdá and Doerr, 2005; Lasanta and Cerdà, 2005). The vegetation fires have a very negative image among the population (Pereira et al., 2014b) but are used as a tool for different objectives (Júnior et al., 2014) namely to increase soil nutrient pools promoting the relatively fast recover of the ecosystems after the fire (Bodí et al., 2014; Pereira et al., 2016b). Therefore, it's important to understand how the fire is transformed when interacting with its human and biophysical constraints/drivers to an integrated and sustainable risk management approach (Tedim and Carvalho, 2013). The existence of a fire risk map will be a valuable tool to support forest and fire management decisions, focus prevention activities, improve the efficiency of fire detection systems and manage resources and actions of firefighting with greater effectiveness (Catry et al., 2010; Freire et al., 2002). Therefore, many authors have worked to contribute for the development of fire risk maps all over the world and associated with specific hazards such as climate change and the Mediterranean region (Moriondo et al., 2006) and post-wildfire logging in USA (Donato et al., 2006). On this respect, Catry et al. (2010) used logistic regression models to predict the likelihood of ignition occurrence to produce an ignition risk map for the Portuguese mainland. Chuvieco et al. (2010) presented methods to generate the input variables and the risk integration to map wildland fire risk for several regions of Spain, using geographic information system and remote sensing technologies. Finney et al. (2011) conducted a simulation study in order to develop a large fire risk assessment system for the adjoining land area of the United States, by using stochastic and statistical models. Portugal is on the top of the European countries most affected by vegetation fires (Caetano et al., 2004; Pereira et al., 2014a) and the Portuguese authorities used two different methods to calculate the wildfire hazard based on the definition of Bachmann and Allgöwer (1999). The Portuguese Geographic Institute (IGP, 2016) uses a physical model and the variables that help to explain the spatial pattern of vegetation fires (namely, land cover, slope, roads, aspect and population density) to estimate the fire hazard (Tedim et al., 2014). On the other hand, the Institute for the Conservation of Nature and Forests (ICNF), (ICNF, 2016a) estimates the fire hazard using the deterministic approach of Verde and Zêzere (2010), which is based on the slope, land cover and
fire probability as well as the quintiles to define the 5 hazard classes to map the fire hazard in accordance with Portuguese Law – DecreeLaw No. 124/2006 of 28 June – (DL, 2006). Finally, the Portuguese Institute for the Sea and the Atmosphere (IPMA, 2016) computes and provides a daily vegetation fire risk index, combining the structural risk index provided by the ICNF and the Canadian fire weather index which is, in essence, a meteorological index of vegetation fire danger. However, to the best of our knowledge, there is no map of structural fire risk for entire Portugal using this deterministic approach and high resolution updated data which, therefore, is the main objective of this study. To ensure a consistent quantitative analysis of the risk of fire and the comparability of the results is necessary to use the concepts and terminology recognized, accepted and adopted by the community studying this kind of phenomenon (Bachmann and Allgöwer, 1999; Hardy, 2005). The definition of fire risk is very complex and have evolved and improved over the years, and currently the definition of fire risk accepted by the fire research community is the following: quantitative or qualitative indicator of the probability of an area being the source of ignition by natural or artificial means in a certain period of time, giving even information of the expected positive and negative impacts in that area (Chuvieco, 2009; Finney, 2005; Freire et al., 2002; Hardy, 2005; Jappiot et al., 2009). This definition includes two components (Fig. 1): (i) fire hazard, which is the probability of an area be affected by a fire during a certain period of time, i.e., it comprises the susceptibility of an area and the respective event probability to happen; (ii) potential damage, which is the damage caused by a fire as soon as such occurs, i.e., takes into account the vulnerability of an area to a fire and the economic value that entails (Jappiot et al., 2009). As there is no official method to combine the two main components of the structural risk, the approach adopted in this study is based on the combined use of the methods proposed by Antunes et al. (2011) to assess the potential economic damage of each parcel and of Verde and Zêzere (2010) to map the susceptibility. As from the publication of these studies, there were updated versions of the databases, more complete and with higher quality it is possible to evaluate the effects of new and better data and check the progress and performance of the assessment previously made. Therefore, the objectives of this study are: (i) to update the susceptibility mapping performed by Verde and Zêzere (2010) by using a Digital Terrain Model (DTM) with higher resolution, a longer fire history and the complete land use land cover inventories; (ii) to assess the fire risk in mainland Portugal; and, (iii) to discuss some of the characteristics of the adopted methodology namely their nature, the number and type of input variables/drivers, the effect of
Fig. 1. Framework to compute and map the structural fire risk in mainland Portugal in this study.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
the updated datasets and the development of fire susceptibility/risk map with five classes. Although it has been referred to the concept of fire risk, it is important to emphasize that this study will focus only on the structural fire risk, which measures the severity and the probability of a particular event (e.g., at the beginning of the season of the fires) based on the combination of geographic variables which do not change in the short term and can be used to improve planning activities and fire-fighting surveillance. We strongly believe that the structural fire risk mapping will be an added value and a basic tool to support the adaptation and mitigation of the economic and environmental effects of the vegetation fires. 2. Data and methods 2.1. Study area Mainland Portugal is located in the Iberian Peninsula, between Spain, at east, and the Atlantic Ocean at west. It has a land area of 89,000 km2 which is mostly occupied by forests and scrublands (48%) and agricultural areas (47%) (Pereira et al., 2014a). The spatial pattern of the altitude and population density is highly heterogeneous (Fig. 2). The altitude is very varied, ranging from sea level in the western and southern coast to about 2000 m, with the largest continental altitude located in the central region of the country (Pereira et al., 2015a). The distribution of the population (about 10.6 million habitants) within the territory is very asymmetric, with higher density in the north-western and southern coastal area of the country as well as around the major cities. 2.2. The datasets The data required to map the fire risk comprises (i) the National Mapping Burnt Area provided by ICNF (ICNF, 2016b) and later modified by (Parente et al., 2016); (ii) the Corine land cover (CLC) for 1990 (EEA, 2002), 2000 (EEA, 2014) and 2006 (DGT, 2016); (iii) spatial distribution of slope with 25 m resolution (Goncalves and Morgado, 2008); (iv) and the official economic values of forests to the levels of CLC's considered (CIAF, 2015). 2.2.1. The fire datasets There are two official fire datasets in Portugal: (i) Portugal Rural Fire Database (PRFD) which is based on ground measurements and includes, for each fire record, the fire location (in terms of the local administrative regions until 2000 and on the spatial coordinates afterwards), detailed
885
ignition and extinction date and time of the fire events, the area burnt (in forest, shrub lands and agricultural areas) and has been successively updated and corrected for different increasing covering periods (Parente et al., 2016; Pereira et al., 2011, 2015a); and, (ii) the National Mapping Burnt Area (NMBA) which is based on satellite imagery information, composed by the burnt area perimeters which comprises a detailed description of fire size and shape in an annual time scale and was recently reviewed to correct missing values and minor inconsistencies (Parente et al., 2016). Accordingly, the PRFD is more appropriate when the analysis to be performed requires high/detailed temporal resolution, while NMBA should be selected when the knowledge of the location, shape and size of the burned area is of fundamental importance (Parente et al., 2016). Therefore, for this study the NMBA was selected as the reference for the past history of burnt areas for 1975 to 2013 which comprises a total of 48,240 fire perimeters (Fig. 3). 2.2.2. Corine land cover Corine land cover (CLC) is one of the main components of the coordination of information on the environment of the European Union program which main objective is the production of occupation and/or land use maps of most of the European areas in 44 different classes starting in a year around 1990 (Caetano et al., 2009). The latest inventory represents the situation of land occupation by the year 2006 (CLC2006) with a minimum 25 ha map unit, showing build up areas, agriculture, forest and semi-natural areas, wetlands and water bodies (AuneLundberg and Strand, 2010). Since one of the main objectives of this study is to map the wildfire susceptibility and to compare the obtained results with the previous findings of Verde and Zêzere (2010) the same methodology was adopted which is based on the use of past burn scar maps and the fact that susceptibility is a measure/depends on the predisposition of the territory and where the wildfires can occur. Therefore, the following CLC thematic layers were excluded: artificial surfaces, wetlands and water bodies, corresponding to levels 1, 4 and 5, as well as the burnt areas, corresponding to the level 334(Fig. 4). 2.2.3. Slope The topography of a region is characterized by the altitude, slope and exposure, has a high impact on the local rainfall (orographic precipitation), air temperature (mean vertical gradient of − 0.6 °C/100 m in the troposphere) and, therefore, on the type and amount of vegetation covered as well as shaping the wind field thus contributing to drive the fire propagation (Chuvieco and Congalton, 1989; Freire et al., 2002; Verde and Zêzere, 2010).
Fig. 2. Altitude of mainland Portugal taking into account the digital elevation model with 25 m resolution provided by Goncalves and Morgado (2008) and the parish population density for 2011 in mainland Portugal. Population data from census 2011 provided by the Statistics Portugal (INE, 2012).
886
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Fig. 3. Burnt area scars for 1975–2013 period according to the National Mapping Burnt Area (NMBA) dataset. Data provided by the Portuguese Institute for the Conservation of Nature and Forest (ICNF, 2016b) and by the Forest Research Centre, School of Agriculture, University of Lisbon.
The high resolution spatial pattern of the slope in mainland of Portugal can be obtained by using the DTM with a spatial resolution of 25 m (Goncalves and Morgado, 2008) and the raster techniques available on Quantum GIS (Goel et al., 2015). For this study, the slope was categorized in the same 6 classes used by Verde and Zêzere (2010): 0–2°, 2– 5°, 5–10°, 10–15°, 15–20°, ≥20° (Fig. 5).
2.3. Fire risk assessment The risk assessment is based on different ways of modelling (probabilistic, semi-probabilistic or determinist) and using several spatial analysis tools (namely, data layers combination using GIS, spatial interaction, extraction of data and classical statistical analysis), which
Fig. 4. Corine land cover 1990, 2000 and 2006 for mainland Portugal. Adapted from DGT (2016); EEA (2002) and EEA (2014).
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
887
where freq is the number of times the pixel burned in the study period (which cannot exceed one per year due to the fire dataset limitations) and Ny is the duration of the study period in years. The susceptibility is then computed, for each pixel, by integrating all the favourability scores of all variables and the fire probability as SP ¼ fp Fav1 Fav2 … Favn Due to multiplicative nature of the above equation all the pixels with null fire favourability scores were reclassified to one percent, thus becoming neutral values in the previous equation. The obtained value expresses all the possible combinations of the variables found for each pixel. To be in accordance with the Decree-Law No. 124/2006 of 28 June (DL, 2006), the fire susceptibility is mapped with just 5 classes based on quintiles of the susceptible values obtained in the previous iteration. For each of these classes the prediction values which represent the total area that will burn in the near future, were computed by predv ¼
Fig. 5. Slope map of mainland Portugal with the spatial resolution of 25 m.
integrate the factors that favours the occurrence and spread of the fire such as vegetation cover type, topography, climate and social variables (Antunes et al., 2011; Bachmann and Allgöwer, 1999; Blanchi et al., 2002; Finney, 2005). The next subsections will described the methods used to map the risk of fire in mainland Portugal. 2.3.1. Fire hazard assessment According to some authors the fire hazard has two types of variants: (i) spatial and long range related to the type of fuel and the topography (physical properties of the surface/landscape); (ii) temporal and low range, related to the content of the fuel mixture and with the atmospheric conditions (Fernandes et al., 2006; Leblon, 2001). Being one of the aims of this study to upgrade the susceptibility map of Verde and Zêzere (2010), the first variation of the definition was adopted for this study, i.e., the fire hazard will take into consideration the probability of a fire happening in a given place and the susceptibility of the site for the occurrence of this phenomenon. Usually, modelling the fire susceptibility requires a priori expert knowledge of a set of variables which can influence this phenomenon, the relative importance of each variable in the fire susceptibility model and, consequently, the resulting fire risk map is also dependent/function of these factors/variables (Gralewicz et al., 2012; Parisien et al., 2005; Renard et al., 2012). For comparison purposes, the deterministic model proposed by Verde and Zêzere (2010) which is based on the slope and vegetation cover was adopted and considered a benchmarking for this study, will be briefly described as follows. First, the favourability scores are computed for all variables, except fire probability, as Favx ¼
NBPx 100 TNPx
where Favx is the favourability score for each slope or CLC class x, NBPx is the total number of burnt pixels in class x, and TNPx is the total number of pixels in the class x. The fire probability of each pixel (fp) is estimated using the fire database and the classic definition of probability (André, 2008) according to fp ¼
freq 100 Ny
Burned pixels in susceptibility class x 100 Total burned pixels on entire territory
The hazard map will be produced with the same classes of the susceptibility map, but with probabilistic values, taking into account a given scenario of future burnt area computed by, BA f predv P ¼ 1− 1− TAx where P is the probability, TAx is the total area in the susceptibility class x, predv is the prediction value for that class and BAf are given by the following equation BA f ¼
Futureburntarea Areaof a singlepixel
2.3.2. Potential damage assessment The assessment of potential damage takes into account: (i) the fire intensity, which depends on climatic conditions, topography, and the available fuel load; (ii) the vulnerability of the analysed elements which depends on the ability of the structure withstand a certain amount of heat during a certain time without keeping a high damage, i.e. the degree of loss on a particular element or group of elements at Table 1 Economic value (€/ha) and the vulnerability coefficient (Antunes et al., 2011) of the vegetative species associated with the CLC classes used for computing the vulnerability. Forest type Scrubs Softwoods
Hardwoods
Mixed forest
Agricultural Open spaces with little or no vegetation
Species
Pinus pinaster Other softwoods Pinus pinea Quercus faginea Other hardwoods Quercus suber Quercus ilex Ceratonia siliqua Eucalyptus Castanea sativa Miller Arbutus unedo – –
Economic value (€/ha)
Vulnerability
53 93 90 497 87 1553
0.5 0.9 0.9 0.9 0.75 0.75
618 112 781
0.75 0.75 0.75
137 849
0.75 0.75
191 200 10
0.75 0.5 0.25
888
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Fig. 6. Favourability score for CLC and slope as well as the fire probability in the study period of 1975–2013 in a Digital Terrain Model with 25 m resolution.
risk as a result of the occurrence of a given natural phenomenon of a given magnitude (Chung and Fabbri, 2005); (iii) value of elements that interact with the phenomenon under study (Chung and Fabbri, 2005; Jappiot et al., 2009). The potential damage is computed, in each pixel, by the product of the vulnerability of the elements and their respective economic value (Antunes et al., 2011). The vulnerability is defined according to the level or degree of destruction of the elements caused by fire and the vulnerability scores adopted for each vegetation type (Table 1) were the ones proposed by Antunes et al. (2011). Only the economic value of the CLC layers used to map the susceptibility were considered to estimate the potential damage. The economic value was assessed from the monetary values defined in the resolution of the Portuguese Council of Ministers no. 6-B/2015 that updates the National Strategy for the Forests (DGRF, 2006) with a horizon to 2030, where the values of the different types of forest and scrubland are set (Table 1). The final value of the potential damage reflect a monetary loss per surface area.
3. Results 3.1. Hazard assessment The fire probability as well as the CLC and slope favourability scores obtained with the longer fire historical dataset, using all the CLC inventories and high resolution DTM are shown in Fig. 6. As expected and can be seen, the higher favourability scores are present in the regions where the fire probability is also higher. However, it is important to underline that the same class of CLC or slope present the same favourability score everywhere in the maps. To compare these findings with the ones on Verde and Zêzere (2010), the temporal average of the CLC and slope favourability scores obtained in both studies were represented simultaneously in the same plot (Figs. 7 and 8). It is important to underline the increase in the favourability scores in all CLC classes except for Permanently irrigated land (212), Mixed forest (313) and Beaches, dunes, sands (331) and in
Fig. 7. Medium favourability scores for the Corine land cover (CLC) classes obtained in this study (blue columns) and in Verde and Zêzere (2010) (orange columns).
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
889
Fig. 8. Medium favourability scores for slope classes obtained in this study for a 25 m resolution (blue columns) and for the ones obtained in Verde and Zêzere (2010) for a 80 m resolution (orange columns). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
the 3 classes of lower slope. This generalized increase in the favourability scores is essentially due to the increase in the total burnt area in the study period (1.9 Mha in 1975–1994 to 4.3 Mha in 1975– 2013) since the changes in the area occupied by each CLC class are positive for some classes and negative for others but the absolute value of the relative change class area is much small than the increase in the total burnt area in the class. The increase in the favourability scores in some CLC classes is much higher than in most slope classes. The fire susceptibility obtained by integrating the three variables (CLC, slope and fire probability) for the 1975–2013 period was mapped in 5 classes with the respective predictive values and shown in Fig. 9. The most susceptible areas are located in areas with, simultaneously, high favourability scores and high fire probability. The achieved susceptibility map is very similar with the one obtained by Verde and Zêzere (2010). However the predictive values of each class are quite different.
In essence, the predicted value increase 25.6% (52% to 77.6%) for the more susceptible (‘very high’) class and decreased for all other classes (Table 2). It is very important to underline these differences because the predictive values can be used to estimate how much of the future total area will burn in each susceptibility class. Therefore, the assessment of the fire hazard was performed considering the same five susceptibility classes and for three different future scenarios of annual burnt area (Table 3), namely corresponding to about: (i) the maximum annual burnt area ever registered in Portugal, i.e., 500,000 ha; (ii) the average annual burnt area, i.e., 250,000 ha; and the minimum annual burnt area, i.e., 125,000 ha which, curiously, ranging from a factor of 2 to each other. The hazard maps are not
Table 2 Predictive values of each of the five fire susceptibility classes obtained in the present study and in the past by Verde and Zêzere (2010). Predictive Value Susceptibility class
Past
Present
Very low Low Medium High Very high
3.0 5.0 12.0 28.0 52.0
0.4 1.3 5.1 15.6 77.6
Table 3 Hazard assessment for three different future scenarios for the five considered susceptibility classes. 1st case: 500,000 ha will burn in a year; 2nd case: 250,000 ha will burn in a year; 3rd case: 125,000 ha will burn in a year. Hazard (%)
Fig. 9. Fire susceptibility map for mainland Portugal with 5 classes defined by the quintiles of the susceptibility values in each pixel.
Susceptibility class
1st case
2nd case
3rd case
Very low Low Medium High Very high
0.14 0.32 1.66 4.19 23.02
0.07 0.16 0.83 2.09 11.51
0.04 0.08 0.42 1.05 5.76
890
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Table 4 Hazard assessment for the annual burnt area scenario of 500,000 ha in the past (Verde and Zêzere, 2010) and in this present study. Susceptibility class
Very low Low Medium High Very high
inner region of the country will have the lower fire risk and the central region the higher fire risk classes. 4. Discussion
Hazard (%) Past
Present
0.85 1.40 3.38 8.42 16.81
0.14 0.32 1.66 4.19 23.02
shown because their spatial patterns are exactly equal to the susceptibility map and only differing in the values of each class (Table 3). The hazard for the worst case scenario (annual burnt area of 500,000 ha) can be compared with the estimates by Verde and Zêzere (2010) revealing that the probability of a pixel burn and belonging to the very high class increased 6.2% (from 16.8% to 23.02%) while the probability of a pixel burn and belonging to any other class decreased − 70% in the very low class and about − 20% in the other classes (Table 4) 3.2. Potential damage and fire risk assessment The vulnerability values described in subsection 2.3.2 were mapped along with the potential damage computed with the economic values presented in the same section (Fig. 10). In general, there is a good spatial agreement between the vulnerability and potential economic damage in the sense that the regions with higher vulnerability also present higher economic potential damage. When computing the fire risk by integrating the fire hazard with the potential economic damage a wide range of values can be obtained as a consequence of the different potential economical values (0–390 €/ha) and hazard probability for the five classes of susceptibility and three burnt area case scenarios (0.1–19.6). To map the fire risk the obtained values can be classified in different classes using the quintiles or other method. However, to keep the association with the susceptibility map and to respect the Portuguese law, the fire risk was mapped using the same five classes and the three case scenarios considered in the previous subsection. The results of the risk assessment (Fig. 11) includes the fire risk probability as well as the weighted economic potential damage and reveal that whatever the case under study the south-
This study aims to map the fire risk in Portugal and for that purposes adopted a consensual conceptual framework within the fire research community (Bachmann and Allgöwer, 1999; Hardy, 2005), the methodology of Verde and Zêzere (2010) for the assessment of the wildfire susceptibility and hazard and the approach of Antunes et al. (2011) to assess the potential economic damage. At the first glance, the susceptibility mapping method seems to be a very simple model, based on just two variables – slope and vegetation cover – but, in fact, this is a very robust and resistant model for the following reasons: (i) the authors (Verde and Zêzere, 2010) tested (alternatively and/or additionally) other driver variables, namely elevation, temperature and precipitation, which are usually recognized as among the most important vegetation fire conditioning factors, without any significant increase in the prediction rates; and (ii) some of these variables can be proxies of each other. For example, slope is a measure of the altitude change. The altitude or elevation regulates the rainfall and temperature which, consequently, define the climatic/weather conditions that, in turn, determines the existence, type and state of the vegetation in each location. This means that vegetation cover is also a proxy for climate, altitude, and viceversa, i.e., when using the vegetation cover the climate information are also being considered in spite of not explicitly. Most likely because of its robustness this susceptibility model/map was adopted by the Portuguese authorities (i.e. ICNF and IPMA). As expected, the favourability scores are higher in specific CLC classes, namely Open spaces with little or no vegetation (explicitly, 333 Sparsely vegetated areas and 332 Bare rocks) and Scrub and/or herbaceous vegetation associations (specifically, 321 Natural grasslands and 322 Moors and heathland) as well as in the classes of higher slope (N 10°). In the first case, these results are in good agreement with the findings of (Pereira et al., 2014a) while in the second case it is a natural consequence of slope be one of the variables that most enhances the spread of fire not only because it favours the energy propagation and the ignition of adjacent combustible but also because these regions are generally more isolated and with less access for firefighting (Morandini et al., 2014; Tihay et al., 2014). It is also important to underline that the favourability scores do not necessarily allows to identify the classes that are most affected by fire in absolute terms but, since is
Fig. 10. Spatial distribution of vulnerability and potential economic damage for mainland of Portugal.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
891
Fig. 11. Spatial distribution of fire risk for the 3 cases in study as function of medium economic value for the quintiles of the susceptibility values in each pixel. 1st case: 500,000 ha will burn in a year; 2nd case: 250,000 ha will burn in a year; 3rd case: 125,000 ha will burn in a year.
defined as the ratio between the burnt area in that class and the total area of the class, is the class fire proneness, i.e., a relative measure of how each class is affected by fire (Pereira et al., 2014a). Another aspect that deserves some attention is the use of the areas burnt by fire in various stages of the framework: twice in the evaluation of the susceptibility, in particular (1) to compute the favourability scores and (2) explicitly in the expression of susceptibility in the form of fire probability in each pixel; the probability of fire turn is again used in the same sequence (3) to determine the fire hazard. The later use is a consequence of the fire hazard definition but the first two are also very easy to justify. The first use is to rank the CLC and slope classes in terms of their fire proneness while the second is to discriminate where, within the country, each class is more or less affected. On this respect it also worth noting that the fire probability is a measure of the fire recurrence but also a proxy for the human behaviour since the large majority of the fires are caused by the humans (Verde and Zêzere, 2010). It is also important to discuss the consequences of using a DTM with higher resolution, several inventories of land use and vegetation cover, and longer fire history. The use of a high resolution DTM allows mapping the susceptibility with higher spatial accuracy as a consequence of assessing all aspects of the fire risk (favourability scores, fire probability, etc.) with higher spatial resolution. Changes in the area occupied by each CLC class from 2000 to 2006 were recently assessed for Portugal (Pereira et al., 2014a) revealing a significant decrease (of about −20%) in the area of broad-leaved and coniferous forests as well as of the non-vegetated areas but also an even more substantial increase (20%–30%) in the artificial surfaces and in the area of sclerophyllous vegetation. This is a consequence of several factors such as rural abandonment, urbanization expansion and also vegetation fires and must be underlined also because of occur in a very short period (7 years) in comparison to the entire study period (39 years) which coincides with the period of higher changes in almost all aspects of the country. This means that the computation of the favourability scores may be biased. That is the main reason why, in this study, the favourability scores were computed using the three CLC inventories and not just one. Nevertheless, in a more recent study (Verde, 2015), the robustness of this susceptibility model was assessed in what respects to the use of single or multiple CLC inventories and the obtained results points to a relative independence of the model performance in relation to how many or which CLC inventory are/is used to access the favourability scores. This could be due to two reasons: the effective good quality of the model or the use of five classes to map the fire susceptibility which could mask the effects of eventual changes in the area of the CLC classes. Eventually, if those changes were taken into account, the estimated favourability scores could be different and imply changes in the sorting of the CLC classes (in terms of their susceptibility) but without implying
changing between susceptibility classes. However we are more convinced of the good quality of the model not just because of its essentially deterministic nature, large amount of reliable data, all tests and comparisons performed previously (Verde and Zêzere, 2010; Verde, 2015), because most significant changes in the CLC classes' areas were in the classes excluded in this study or not in the ones with higher fire proneness/favourability (Pereira et al., 2014a) but, essentially, for the large amount of high quality data used to calibrate and validate the model. In fact, one of the most conditioning factors when developing a model is the quality but also the quantity of data available which, in addition, need to be split for the independent calibration, testing and validation procedures (Pereira et al., 2013). The amount of data constrains the number of statistical significant predictors that can be used in a model, calibration data should be sufficient to assure the stable estimation of model parameters Validation should be performed with data variability within the range of values where the model will operate (Knight and Shamseldin, 2005; Ogundimu et al., 2016). However, this study uses 27 predictors (21 CLC and 6 slope classes) but 134 million pixels to model the fire susceptibility and risk. Another important and definitive indicator of the good quality of the Verde and Zêzere (2010) model is the increasing performance when using more/new/different data for an extended/future period. In general, the use of new data in a model (e.g. validation or prediction) lead to a decrease in the goodness of fit statistics obtained during the tuning procedure (i.e., calibration or test procedure). However, the use of a longer fire historical dataset in this study leads to an increase of the favourability scores, especially evident in the classes of higher susceptibility and, which is more relevant, to an increase of the predictive value for the very high susceptibility class at the expense of the decrease of the predictive value of lower susceptibility classes. In fact, the good quality of the susceptibility map is of fundamental importance since it determines the quality of the fire risk map. Finally, it is important to compare the obtained results with the findings of previously published studies. However, most of the previous fire risk assessments for Portugal were performed for different fire risk concepts, non-structural fire risk aspects, specific vegetation types or small regions such as assessing the relative importance of different factors on the spatial distribution of fire ignitions (Catry et al., 2008), annual fire risk maps fire as the probability of its occurrence (Turkman et al., 2014), fire probability in forestry stands (da Costa Ricardo, 2010), or assessing and mapping (structural, dynamical and/or integrated) fire risk in specific counties or districts (Antunes et al., 2011; Caetano et al., 2004; Teodoro and Duarte, 2013). Results for entire mainland Portugal includes: the forest fire risk at county level (Lourenco, 1994), the fire risk map adopted by the Portuguese Forest Service (Pereira and dos Santos, 2003). All these maps were produced with much short fire history, mostly before the sequence of devastating fires in 2003 and 2005,
892
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
and have a much lower spatial resolution to be properly compared with the results of this study. 5. Conclusion In summary, all objectives of this study were achieved in the sense that a fire risk map for the vegetative areas was produced for Portugal, based on an updated susceptibility map and the most important aspects of the adopted methodology were discussed. The implemented framework for the fire risk assessment is the most recent and gathers greater consensus within the research community. The susceptibility mapping model of Verde and Zêzere (2010) was adopted for its excellent performance, be in accordance with the Portuguese law and adopted by the national authorities. The susceptibility of the territory was derived from vegetation cover, slope and fire probability and mapped with just five classes. This model was updated in this study using higher spatial resolution DTM, longer and up to date fire history as well as several Corine land cover inventories. Updated susceptibility map is extremely similar to the previous in terms of the spatial patterns of susceptibility classes confirming the good quality of this model which present higher performance when using new fire data and higher resolution elevation map. In fact, the predictive value increase for the very high susceptibility class and decrease for the other classes of lower susceptibility. The fire hazard was estimated for three different scenarios of burnt area corresponding to years with minimum, mean and maximum annual burnt ever registered in the study period. The potential economic damage was computed with the vulnerability scores proposed for the different vegetation types in Portugal by Antunes et al. (2011) and the economic value of the vegetation layers defined by the Portuguese law (CIAF, 2015) which is based on the National Strategy for the Forests for the 2030 horizon. The obtained fire risk map is an estimate of the structural vegetation fire risk for the 1975–2013 period using a deterministic approach to compute the susceptibility and the potential economic damage with the highest possible spatial resolution (25 m) at this moment. This fire risk map reveal that the area with higher risk is located in the central and north part of the country as well as in extreme south (Algarve) whatever the burnt area scenario considered for the future. On the contrary the area with lower fire risk is located in the south of the country, specifically in the interior of Alentejo. These results are due mainly to the landscape and vegetation cover conditions conducive to the development of vegetation fires present in these two regions that do not exist in other areas of the country. This map could be very useful to take more effective the measures to prevent and combat vegetation fires, with the contribution and participation of all the agents of the territory management, including the populations. The plan for future work includes the assessment of the usefulness of other variables (e.g., meteorological, socioeconomic) to improve the susceptibility map, to update the vulnerability map, to consider human, landscape and ecological potential damage and, finally, to include the dynamical risk factors, namely weather conditions. Acknowledgements This work was supported by: (i) the Herbette Foundation of the University of Lausanne; (ii) the FIREXTR project, PTDC/ATPGEO/0462/ 2014; (iii) the project Interact - Integrative Research in Environment, Agro-Chain and Technology, NORTE-01-0145-FEDER-000017, research line BEST, co-financed by FEDER/NORTE 2020; and, (iv) European Investment Funds by FEDER/COMPETE/POCI – Operacional Competitiveness and Internacionalization Programme, under Project POCI-010145-FEDER-006958 and National Funds by FCT - Portuguese Foundation for Science and Technology, under the project UID/AGR/04033. We are especially grateful to ICNF and ISA for providing the fire data and to João Pereira for the final spelling and grammar review of the manuscript.
References André, J., 2008. Probabilidades e Estatística para a Engenharia. Lidel. Antunes, C.C., Viegas, D.X., Mendes, J.M., 2011. Avaliação do Risco de Incêndio Florestal no Concelho de Arganil. Silva Lusitana. 19, pp. 165–179. Aune-Lundberg, L., Strand, G.-H., 2010. CORINE Land Cover 2006. The Norwegian CLC2006 project. Report from the Norwegian Forest and Landscape Institute. 11, p. 2010. Bachmann, A., Allgöwer, B., 1999. The need for a consistent wildfire risk terminology. The Joint Fire Science Conference and Workshop: Crossing the Millennium: Integrating Spatial Technologies and Ecological Principles for a New Age in Fire Management. 1, pp. 67–77. Blanchi, R., Jappiot, M., Alexandrian, D., 2002. Forest fire risk assessment and cartography. A methodological approach. Proceedings of the IV International Conference on Forest Fire Research. Luso, Portugal. Citeseer. Bodí, M.B., Martin, D.A., Balfour, V.N., Santín, C., Doerr, S.H., Pereira, P., et al., 2014. Wildland fire ash: production, composition and eco-hydro-geomorphic effects. Earth Sci. Rev. 130, 103–127. Caetano, M., Freire, S., Carrão, H., 2004. Fire risk mapping by integration of dynamic and structural variables. Remote Sens. Transit. 1. Caetano, M., Nunes, V., Nunes, A., 2009. CORINE Land Cover 2006 for Continental Portugal. Relatório técnico. Instituto Geográfico Português, Lisbon, Portugal. Catry, F., Rego, F., Moreira, F., Bacao, F., 2008. Characterizing and modelling the spatial patterns of wildfire ignitions in Portugal: fire initiation and resulting burned area. WIT Trans. Ecol. Environ. 119. Catry, F.X., Rego, F.C., Bação, F.L., Moreira, F., 2010. Modeling and mapping wildfire ignition risk in Portugal. Int. J. Wildland Fire 18, 921–931. Cerdá, A., Doerr, S.H., 2005. Influence of vegetation recovery on soil hydrology and erodibility following fire: an 11-year investigation. Int. J. Wildland Fire 14, 423–437. Chung, C., Fabbri, A.G., 2005. Systematic procedures of landslide hazard mapping for risk assessment using spatial prediction models. Landslide Hazard and Risk. Wiley, New York, pp. 139–177. Chuvieco, E., Congalton, R.G., 1989. Application of remote sensing and geographic information systems to forest fire hazard mapping. Remote Sens. Environ. 29, 147–159. Chuvieco, E., Aguado, I., Yebra, M., Nieto, H., Salas, J., Martín, M.P., et al., 2010. Development of a framework for fire risk assessment using remote sensing and geographic information system technologies. Ecol. Model. 221, 46–58. Chuvieco, E., 2009. Earth Observation of Wildland Fires in Mediterranean Ecosystems. Springer. CIAF, 2015. Resolução do Conselho de Ministros no. 6-B/2015. Diário da República, p. 24. da Costa Ricardo, A., 2010. Modelação da probabilidade de ocorrência de incêndio em povoamentos florestais de Portugal Continental. Universidade Técnica de Lisboa, Lisboa. DGRF, 2006. Resolução do Conselho de Ministros n.o 114/2006. Diário da República, 1.a série, p. 179. DGT, 2016. Corine Land Cover (CLC). DL, 2006. Decreto-Lei no. 124/2006 de 28 de Junho. Diário Da República — I SÉRIE-A, p. 123. Donato, D., Fontaine, J., Campbell, J., Robinson, W., Kauffman, J., Law, B.E., 2006. Post-wildfire logging hinders regeneration and increases fire risk. Science 311, 352. EEA, 2002. Corine Land Cover 1990 (CLC1990) and Corine Land Cover Changes 1975– 1990 in a 10 km Zone Around the Coast of Europe. EEA, 2014. Corine Land Cover 2000 Seamless Vector Data. Fernandes, P., Luz, A., Loureiro, C., Ferreira-Godinho, P., Botelho, H., 2006. Fuel modelling and fire hazard assessment based on data from the Portuguese National Forest Inventory. For. Ecol. Manag. 234, S229. Finney, M.A., 2005. The challenge of quantitative risk analysis for wildland fire. For. Ecol. Manag. 211, 97–108. Finney, M.A., McHugh, C.W., Grenfell, I.C., Riley, K.L., Short, K.C., 2011. A simulation of probabilistic wildfire risk components for the continental United States. Stoch. Env. Res. Risk A. 25, 973–1000. Freire, S., Carrão, H., Caetano, M.R., 2002. Produção de cartografia de risco de incêndio florestal com recurso a imagens de satélite e dados auxiliares. Lisboa, IGP. Goel, D., Pandey, N., Gupta, S., 2015. Proprietary and Open Source Geospatial Software. Organized by Department of Civil Engineering 48. Indian Institute of Technology (Banaras Hindu University), Varanasi-221005, Uttar Pradesh, India. Goncalves, J., Morgado, A., 2008. Use of the SRTM DEM as a geo-referencing tool by elevation matching. Int. Arch. Photogramm. Remote. Sens. Spat. Inf. Sci. 37, 879–883. Gralewicz, N.J., Nelson, T.A., Wulder, M.A., 2012. Factors influencing national scale wildfire susceptibility in Canada. For. Ecol. Manag. 265, 20–29. Hardy, C.C., 2005. Wildland fire hazard and risk: problems, definitions, and context. For. Ecol. Manag. 211, 73–82. Hedo, J., Lucas-Borja, M., Wic, C., Andrés-Abellán, M., de Las Heras, J., 2015. Soil microbiological properties and enzymatic activities of long-term post-fire recovery in dry and semiarid Aleppo pine (Pinus halepensis M.) forest stands. Solid Earth 6, 243. ICNF, 2016a. Cartografia de Risco - mapa de perigosidade de Incêndio Florestal. ICNF, 2016b. Informação Geográfica. IGP, 2016. Dados Abertos. INE, 2012. Censos 2011 resultados definitivos-Portugal. Instituto Nacional de Estatística, IP, Lisboa-Portugal. IPMA, 2016. Classes de Risco de Incêndio por Concelho. RCM - Hoje. Jaiswal, R.K., Mukherjee, S., Raju, K.D., Saxena, R., 2002. Forest fire risk zone mapping from satellite imagery and GIS. Int. J. Appl. Earth Obs. Geoinf. 4, 1–10. Jappiot, M., Gonzalez-Olabarria, J., Lampin-Maillet, C., Borgniet, L., 2009. Assessing wildfire risk in time and space. Living with wildfires: what science can tell us. A Contribution to the Science-Policy Dialogue, pp. 41–47.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893 Júnior, A.C.P., Oliveira, S.L., Pereira, J.M., Turkman, M.A.A., 2014. Modelling fire frequency in a Cerrado savanna protected area. PLoS One 9, e102380. Keesstra, S., Wittenberg, L., Maroulis, J., Sambalino, F., Malkinson, D., Cerdà, A., et al., 2016. The influence of fire history, plant species and post-fire management on soil water repellency in a Mediterranean catchment: the Mount Carmel range, Israel. Catena. Knight, D., Shamseldin, A., 2005. River Basin Modelling for Flood Risk Mitigation. CRC Press. Lasanta, T., Cerdà, A., 2005. Long-term erosional responses after fire in the central Spanish Pyrenees: 2. Solute release. Catena 60, 81–100. Leblon, B., 2001. Forest wildfire hazard monitoring using remote sensing: a review. Remote Sens. Rev. 20, 1–43. Lourenco, L., 1994. Risco de incendio florestal em Portugal Continental. Informação florestal. Morandini, F., Silvani, X., Honoré, D., Boutin, G., Susset, A., Vernet, R., 2014. Slope effects on the fluid dynamics of a fire spreading across a fuel bed: PIV measurements and OH* chemiluminescence imaging. Exp. Fluids 55, 1–12. Moriondo, M., Good, P., Durao, R., Bindi, M., Giannakopoulos, C., Corte-Real, J., 2006. Potential impact of climate change on fire risk in the Mediterranean area. Clim. Res. 31, 85–95. Ogundimu, E.O., Altman, D.G., Collins, G.S., 2016. Adequate sample size for developing prediction models is not simply related to events per variable. J. Clin. Epidemiol. Pacheco, F., Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., 2015. Controls and forecasts of nitrate yields in forested watersheds: a view over mainland Portugal. Sci. Total Environ. 537, 421–440. Parente, J., Pereira, M.G., Tonini, M., 2016. Space-time clustering analysis of wildfires: the influence of dataset characteristics, fire prevention policy decisions, weather and climate. Sci. Total Environ. 559, 151–165. Parisien, M.-A., Kafka, V., Hirsch, K., Todd, J., Lavoie, S., Maczek, P., 2005. Mapping Wildfire Susceptibility with the BURN-P3 Simulation Model. Pereira, J.M.C., dos Santos, M.T.N., 2003. Fire Risk and Burned Area Mapping in Portugal: DGF. Pereira, M., Malamud, B., Trigo, R., Alves, P., 2011. The history and characteristics of the 1980–2005 Portuguese rural fire database. Nat. Hazards Earth Syst. Sci. 11, 3343–3358. Pereira, M.G., Aranha, J., Amraoui, M., 2014a. Land cover fire proneness in Europe. 23, 13. Pereira, M.G., Calado, T.J., DaCamara, C.C., Calheiros, T., 2013. Effects of regional climate change on rural fires in Portugal. Clim. Res. 57, 187–200. Pereira, M.G., Caramelo, L., Orozco, C.V., Costa, R., Tonini, M., 2015a. Space-time clustering analysis performance of an aggregated dataset: the case of wildfires in Portugal. Environ. Model. Softw. 72, 239–249. Pereira, M.G., Fernandes, L.S., Carvalho, S., Santos, R.B., Caramelo, L., Alencoão, A., 2016a. Modelling the impacts of wildfires on runoff at the river basin ecological scale in a changing Mediterranean environment. Environ. Earth Sci. 75, 1–14.
893
Pereira, P., Cerdà, A., Lopez, A.J., Zavala, L.M., Mataix-Solera, J., Arcenegui, V., et al., 2016b. Short-term vegetation recovery after a grassland fire in Lithuania: the effects of fire severity, slope position and aspect. Land Degrad. Dev. Pereira, P., Cerdà, A., Úbeda, X., Mataix-Solera, J., Arcenegui, V., Zavala, L., 2015b. Modelling the impacts of wildfire on ash thickness in a short-term period. Land Degrad. Dev. 26, 180–192. Pereira, P., Mierauskas, P., Novara, A., 2014b. Stakeholders' perceptions about fire impacts on Lithuanian protected areas. Land Degradation & Development. Renard, Q., Pélissier, R., Ramesh, B., Kodandapani, N., 2012. Environmental susceptibility model for predicting forest fire occurrence in the western Ghats of India. Int. J. Wildland Fire 21, 368–379. Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., Pacheco, F., 2015a. A framework model for investigating the export of phosphorus to surface waters in forested watersheds: implications to management. Sci. Total Environ. 536, 295–305. Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., Pacheco, F., 2015b. Water resources planning for a river basin with recurrent wildfires. Sci. Total Environ. 526, 1–13. Shaw, E.A., Denef, K., de Tomasel, C.M., Cotrufo, M.F., Wall, D.H., 2016. Fire affects root decomposition, soil food web structure, and carbon flow in tallgrass prairie. Soil. Tedim, F., Carvalho, S., 2013. A vulnerabilidade aos incêndios florestais: reflexões em torno de aspetos conceptuais e metodológicos. Territorium. Tedim, F., Garcin, M., Vinchon, C., Carvalho, S., Desramaut, N., Rohmer, J., 2014. Comprehensive vulnerability assessment of forest fires and coastal erosion: evidences from case-study analysis in Portugal. Assessment of Vulnerability to Natural Hazards: A European Perspective. 149. Teodoro, A.C., Duarte, L., 2013. Forest fire risk maps: a GIS open source application–a case study in Norwest of Portugal. Int. J. Geogr. Inf. Sci. 27, 699–720. Tihay, V., Morandini, F., Santoni, P.-A., Perez-Ramirez, Y., Barboni, T., 2014. Combustion of forest litters under slope conditions: burning rate, heat release rate, convective and radiant fractions for different loads. Combust. Flame 161, 3237–3248. Turkman, K., Turkman, M.A., Pereira, P., Sá, A., Pereira, J., 2014. Generating annual fire risk maps using Bayesian hierarchical models. J. Stat. Theory Pract. 8, 509–533. Van Eck, C.M., Nunes, J.P., Vieira, D., Keesstra, S., Keizer, J.J., 2016. Physically-based modelling of the post-fire runoff response of a Forest catchment in Central Portugal: using field versus remote sensing based estimates of vegetation recovery. Land Degrad. Dev. Verde, J., Zêzere, J., 2010. Assessment and validation of wildfire susceptibility and hazard in Portugal. Nat. Hazards Earth Syst. Sci. 10, 485–497. Verde, J.C.R.M., 2015. Wildfire Susceptibility Modelling in Mainland Portugal. Universidade De Lisboa.
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Structural fire risk: The case of Portugal Joana Parente a,⁎, Mário G. Pereira a,b a b
Centre for Research and Technology of Agro-Environment and Biological Sciences, CITAB, University of Trás-os-Montes and Alto Douro, UTAD, Quinta de Prados, 5000-801 Vila Real, Portugal Instituto Dom Luiz, IDL, Faculdade de Ciências da Universidade de Lisboa, Campo Grande, Edifício C8, Piso 3, 1749-016, Lisboa, Portugal
H I G H L I G H T S
G R A P H I C A L
A B S T R A C T
• Verde and Zêzere susceptibility map was updated with longer and high resolution data. • Susceptibility model's performance and particularities were assessed and discussed. • Structural fire risk computed for updated potential damage and three hazard scenarios • Potential damage mapped with recent vulnerability and Horizon 3030 economic values • Higher fire risk in central, north and extreme south (Algarve) of Portugal
a r t i c l e
i n f o
Article history: Received 16 July 2016 Received in revised form 22 August 2016 Accepted 22 August 2016 Available online xxxx Editor: D. Barcelo Keywords: Vegetation fires Structural fire risk Susceptibility map Potential economic damage
a b s t r a c t This study is focused in mapping the structural fire risk in the vegetated area of Portugal using a deterministic approach based on the concept of fire risk currently accepted by the scientific community which consists in the combination of the fire hazard and the potential economic damage. The fire susceptibility map of Verde and Zêzere (2010) was adopted and updated by the use of a higher resolution digital terrain model, longer burnt area perimeter dataset (1975–2013) and the entire set of Corine land cover inventories. The susceptibility was mapped with five classes to be in accordance with the Portuguese law and the results confirms the good performance of this model not only in the favourability scores but also in the predictive values. Three different scenarios of (maximum, mean, and minimum annual) burnt area were consider to estimate the fire hazard which allow to estimate the likelihood of a pixel to burn (ranging between 0% and 20%) depending on the class to which it belongs and on the future burnt area scenario. The potential economic damage was estimated with the vulnerability scores and monetary values of species defined in the literature and by law. The obtained fire risk map identifies the areas more prone to be affected by fires in the future and provides an estimate of the economic damage of the fire which will be a valuable tool for forest and fire managers and to minimize the economic and environmental consequences of vegetation fires in Portugal. © 2016 Elsevier B.V. All rights reserved.
1. Introduction
⁎ Corresponding author at: Universidade de Trás-os-Montes e Alto Douro, Quinta de Prados, 5000-801 Vila Real, Portugal. E-mail addresses: [email protected] (J. Parente), [email protected] (M.G. Pereira).
http://dx.doi.org/10.1016/j.scitotenv.2016.08.164 0048-9697/© 2016 Elsevier B.V. All rights reserved.
Vegetation fires are part of the natural and human systems and, independently of their causes (lightning, accidental, negligent or intentional), may have a highly destructive effect on the ecosystems affecting its ability to recover, causing major environmental problems,
884
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
threaten human lives, activities and infrastructures (Blanchi et al., 2002; Chuvieco, 2009; Jaiswal et al., 2002). Along with the burning/removal of vegetation, fires have very important impacts on soil food web structure and carbon flow (Shaw et al., 2016), microbiological properties and enzymatic activities (Hedo et al., 2015), development of an ash cover (Pereira et al., 2015b), water repellency (Keesstra et al., 2016), terrestrial branch of the hydrological cycle (Pereira et al., 2016a; Van Eck et al., 2016), water quality (Pacheco et al., 2015; Santos et al., 2015a,b) and erosion (Cerdá and Doerr, 2005; Lasanta and Cerdà, 2005). The vegetation fires have a very negative image among the population (Pereira et al., 2014b) but are used as a tool for different objectives (Júnior et al., 2014) namely to increase soil nutrient pools promoting the relatively fast recover of the ecosystems after the fire (Bodí et al., 2014; Pereira et al., 2016b). Therefore, it's important to understand how the fire is transformed when interacting with its human and biophysical constraints/drivers to an integrated and sustainable risk management approach (Tedim and Carvalho, 2013). The existence of a fire risk map will be a valuable tool to support forest and fire management decisions, focus prevention activities, improve the efficiency of fire detection systems and manage resources and actions of firefighting with greater effectiveness (Catry et al., 2010; Freire et al., 2002). Therefore, many authors have worked to contribute for the development of fire risk maps all over the world and associated with specific hazards such as climate change and the Mediterranean region (Moriondo et al., 2006) and post-wildfire logging in USA (Donato et al., 2006). On this respect, Catry et al. (2010) used logistic regression models to predict the likelihood of ignition occurrence to produce an ignition risk map for the Portuguese mainland. Chuvieco et al. (2010) presented methods to generate the input variables and the risk integration to map wildland fire risk for several regions of Spain, using geographic information system and remote sensing technologies. Finney et al. (2011) conducted a simulation study in order to develop a large fire risk assessment system for the adjoining land area of the United States, by using stochastic and statistical models. Portugal is on the top of the European countries most affected by vegetation fires (Caetano et al., 2004; Pereira et al., 2014a) and the Portuguese authorities used two different methods to calculate the wildfire hazard based on the definition of Bachmann and Allgöwer (1999). The Portuguese Geographic Institute (IGP, 2016) uses a physical model and the variables that help to explain the spatial pattern of vegetation fires (namely, land cover, slope, roads, aspect and population density) to estimate the fire hazard (Tedim et al., 2014). On the other hand, the Institute for the Conservation of Nature and Forests (ICNF), (ICNF, 2016a) estimates the fire hazard using the deterministic approach of Verde and Zêzere (2010), which is based on the slope, land cover and
fire probability as well as the quintiles to define the 5 hazard classes to map the fire hazard in accordance with Portuguese Law – DecreeLaw No. 124/2006 of 28 June – (DL, 2006). Finally, the Portuguese Institute for the Sea and the Atmosphere (IPMA, 2016) computes and provides a daily vegetation fire risk index, combining the structural risk index provided by the ICNF and the Canadian fire weather index which is, in essence, a meteorological index of vegetation fire danger. However, to the best of our knowledge, there is no map of structural fire risk for entire Portugal using this deterministic approach and high resolution updated data which, therefore, is the main objective of this study. To ensure a consistent quantitative analysis of the risk of fire and the comparability of the results is necessary to use the concepts and terminology recognized, accepted and adopted by the community studying this kind of phenomenon (Bachmann and Allgöwer, 1999; Hardy, 2005). The definition of fire risk is very complex and have evolved and improved over the years, and currently the definition of fire risk accepted by the fire research community is the following: quantitative or qualitative indicator of the probability of an area being the source of ignition by natural or artificial means in a certain period of time, giving even information of the expected positive and negative impacts in that area (Chuvieco, 2009; Finney, 2005; Freire et al., 2002; Hardy, 2005; Jappiot et al., 2009). This definition includes two components (Fig. 1): (i) fire hazard, which is the probability of an area be affected by a fire during a certain period of time, i.e., it comprises the susceptibility of an area and the respective event probability to happen; (ii) potential damage, which is the damage caused by a fire as soon as such occurs, i.e., takes into account the vulnerability of an area to a fire and the economic value that entails (Jappiot et al., 2009). As there is no official method to combine the two main components of the structural risk, the approach adopted in this study is based on the combined use of the methods proposed by Antunes et al. (2011) to assess the potential economic damage of each parcel and of Verde and Zêzere (2010) to map the susceptibility. As from the publication of these studies, there were updated versions of the databases, more complete and with higher quality it is possible to evaluate the effects of new and better data and check the progress and performance of the assessment previously made. Therefore, the objectives of this study are: (i) to update the susceptibility mapping performed by Verde and Zêzere (2010) by using a Digital Terrain Model (DTM) with higher resolution, a longer fire history and the complete land use land cover inventories; (ii) to assess the fire risk in mainland Portugal; and, (iii) to discuss some of the characteristics of the adopted methodology namely their nature, the number and type of input variables/drivers, the effect of
Fig. 1. Framework to compute and map the structural fire risk in mainland Portugal in this study.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
the updated datasets and the development of fire susceptibility/risk map with five classes. Although it has been referred to the concept of fire risk, it is important to emphasize that this study will focus only on the structural fire risk, which measures the severity and the probability of a particular event (e.g., at the beginning of the season of the fires) based on the combination of geographic variables which do not change in the short term and can be used to improve planning activities and fire-fighting surveillance. We strongly believe that the structural fire risk mapping will be an added value and a basic tool to support the adaptation and mitigation of the economic and environmental effects of the vegetation fires. 2. Data and methods 2.1. Study area Mainland Portugal is located in the Iberian Peninsula, between Spain, at east, and the Atlantic Ocean at west. It has a land area of 89,000 km2 which is mostly occupied by forests and scrublands (48%) and agricultural areas (47%) (Pereira et al., 2014a). The spatial pattern of the altitude and population density is highly heterogeneous (Fig. 2). The altitude is very varied, ranging from sea level in the western and southern coast to about 2000 m, with the largest continental altitude located in the central region of the country (Pereira et al., 2015a). The distribution of the population (about 10.6 million habitants) within the territory is very asymmetric, with higher density in the north-western and southern coastal area of the country as well as around the major cities. 2.2. The datasets The data required to map the fire risk comprises (i) the National Mapping Burnt Area provided by ICNF (ICNF, 2016b) and later modified by (Parente et al., 2016); (ii) the Corine land cover (CLC) for 1990 (EEA, 2002), 2000 (EEA, 2014) and 2006 (DGT, 2016); (iii) spatial distribution of slope with 25 m resolution (Goncalves and Morgado, 2008); (iv) and the official economic values of forests to the levels of CLC's considered (CIAF, 2015). 2.2.1. The fire datasets There are two official fire datasets in Portugal: (i) Portugal Rural Fire Database (PRFD) which is based on ground measurements and includes, for each fire record, the fire location (in terms of the local administrative regions until 2000 and on the spatial coordinates afterwards), detailed
885
ignition and extinction date and time of the fire events, the area burnt (in forest, shrub lands and agricultural areas) and has been successively updated and corrected for different increasing covering periods (Parente et al., 2016; Pereira et al., 2011, 2015a); and, (ii) the National Mapping Burnt Area (NMBA) which is based on satellite imagery information, composed by the burnt area perimeters which comprises a detailed description of fire size and shape in an annual time scale and was recently reviewed to correct missing values and minor inconsistencies (Parente et al., 2016). Accordingly, the PRFD is more appropriate when the analysis to be performed requires high/detailed temporal resolution, while NMBA should be selected when the knowledge of the location, shape and size of the burned area is of fundamental importance (Parente et al., 2016). Therefore, for this study the NMBA was selected as the reference for the past history of burnt areas for 1975 to 2013 which comprises a total of 48,240 fire perimeters (Fig. 3). 2.2.2. Corine land cover Corine land cover (CLC) is one of the main components of the coordination of information on the environment of the European Union program which main objective is the production of occupation and/or land use maps of most of the European areas in 44 different classes starting in a year around 1990 (Caetano et al., 2009). The latest inventory represents the situation of land occupation by the year 2006 (CLC2006) with a minimum 25 ha map unit, showing build up areas, agriculture, forest and semi-natural areas, wetlands and water bodies (AuneLundberg and Strand, 2010). Since one of the main objectives of this study is to map the wildfire susceptibility and to compare the obtained results with the previous findings of Verde and Zêzere (2010) the same methodology was adopted which is based on the use of past burn scar maps and the fact that susceptibility is a measure/depends on the predisposition of the territory and where the wildfires can occur. Therefore, the following CLC thematic layers were excluded: artificial surfaces, wetlands and water bodies, corresponding to levels 1, 4 and 5, as well as the burnt areas, corresponding to the level 334(Fig. 4). 2.2.3. Slope The topography of a region is characterized by the altitude, slope and exposure, has a high impact on the local rainfall (orographic precipitation), air temperature (mean vertical gradient of − 0.6 °C/100 m in the troposphere) and, therefore, on the type and amount of vegetation covered as well as shaping the wind field thus contributing to drive the fire propagation (Chuvieco and Congalton, 1989; Freire et al., 2002; Verde and Zêzere, 2010).
Fig. 2. Altitude of mainland Portugal taking into account the digital elevation model with 25 m resolution provided by Goncalves and Morgado (2008) and the parish population density for 2011 in mainland Portugal. Population data from census 2011 provided by the Statistics Portugal (INE, 2012).
886
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Fig. 3. Burnt area scars for 1975–2013 period according to the National Mapping Burnt Area (NMBA) dataset. Data provided by the Portuguese Institute for the Conservation of Nature and Forest (ICNF, 2016b) and by the Forest Research Centre, School of Agriculture, University of Lisbon.
The high resolution spatial pattern of the slope in mainland of Portugal can be obtained by using the DTM with a spatial resolution of 25 m (Goncalves and Morgado, 2008) and the raster techniques available on Quantum GIS (Goel et al., 2015). For this study, the slope was categorized in the same 6 classes used by Verde and Zêzere (2010): 0–2°, 2– 5°, 5–10°, 10–15°, 15–20°, ≥20° (Fig. 5).
2.3. Fire risk assessment The risk assessment is based on different ways of modelling (probabilistic, semi-probabilistic or determinist) and using several spatial analysis tools (namely, data layers combination using GIS, spatial interaction, extraction of data and classical statistical analysis), which
Fig. 4. Corine land cover 1990, 2000 and 2006 for mainland Portugal. Adapted from DGT (2016); EEA (2002) and EEA (2014).
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
887
where freq is the number of times the pixel burned in the study period (which cannot exceed one per year due to the fire dataset limitations) and Ny is the duration of the study period in years. The susceptibility is then computed, for each pixel, by integrating all the favourability scores of all variables and the fire probability as SP ¼ fp Fav1 Fav2 … Favn Due to multiplicative nature of the above equation all the pixels with null fire favourability scores were reclassified to one percent, thus becoming neutral values in the previous equation. The obtained value expresses all the possible combinations of the variables found for each pixel. To be in accordance with the Decree-Law No. 124/2006 of 28 June (DL, 2006), the fire susceptibility is mapped with just 5 classes based on quintiles of the susceptible values obtained in the previous iteration. For each of these classes the prediction values which represent the total area that will burn in the near future, were computed by predv ¼
Fig. 5. Slope map of mainland Portugal with the spatial resolution of 25 m.
integrate the factors that favours the occurrence and spread of the fire such as vegetation cover type, topography, climate and social variables (Antunes et al., 2011; Bachmann and Allgöwer, 1999; Blanchi et al., 2002; Finney, 2005). The next subsections will described the methods used to map the risk of fire in mainland Portugal. 2.3.1. Fire hazard assessment According to some authors the fire hazard has two types of variants: (i) spatial and long range related to the type of fuel and the topography (physical properties of the surface/landscape); (ii) temporal and low range, related to the content of the fuel mixture and with the atmospheric conditions (Fernandes et al., 2006; Leblon, 2001). Being one of the aims of this study to upgrade the susceptibility map of Verde and Zêzere (2010), the first variation of the definition was adopted for this study, i.e., the fire hazard will take into consideration the probability of a fire happening in a given place and the susceptibility of the site for the occurrence of this phenomenon. Usually, modelling the fire susceptibility requires a priori expert knowledge of a set of variables which can influence this phenomenon, the relative importance of each variable in the fire susceptibility model and, consequently, the resulting fire risk map is also dependent/function of these factors/variables (Gralewicz et al., 2012; Parisien et al., 2005; Renard et al., 2012). For comparison purposes, the deterministic model proposed by Verde and Zêzere (2010) which is based on the slope and vegetation cover was adopted and considered a benchmarking for this study, will be briefly described as follows. First, the favourability scores are computed for all variables, except fire probability, as Favx ¼
NBPx 100 TNPx
where Favx is the favourability score for each slope or CLC class x, NBPx is the total number of burnt pixels in class x, and TNPx is the total number of pixels in the class x. The fire probability of each pixel (fp) is estimated using the fire database and the classic definition of probability (André, 2008) according to fp ¼
freq 100 Ny
Burned pixels in susceptibility class x 100 Total burned pixels on entire territory
The hazard map will be produced with the same classes of the susceptibility map, but with probabilistic values, taking into account a given scenario of future burnt area computed by, BA f predv P ¼ 1− 1− TAx where P is the probability, TAx is the total area in the susceptibility class x, predv is the prediction value for that class and BAf are given by the following equation BA f ¼
Futureburntarea Areaof a singlepixel
2.3.2. Potential damage assessment The assessment of potential damage takes into account: (i) the fire intensity, which depends on climatic conditions, topography, and the available fuel load; (ii) the vulnerability of the analysed elements which depends on the ability of the structure withstand a certain amount of heat during a certain time without keeping a high damage, i.e. the degree of loss on a particular element or group of elements at Table 1 Economic value (€/ha) and the vulnerability coefficient (Antunes et al., 2011) of the vegetative species associated with the CLC classes used for computing the vulnerability. Forest type Scrubs Softwoods
Hardwoods
Mixed forest
Agricultural Open spaces with little or no vegetation
Species
Pinus pinaster Other softwoods Pinus pinea Quercus faginea Other hardwoods Quercus suber Quercus ilex Ceratonia siliqua Eucalyptus Castanea sativa Miller Arbutus unedo – –
Economic value (€/ha)
Vulnerability
53 93 90 497 87 1553
0.5 0.9 0.9 0.9 0.75 0.75
618 112 781
0.75 0.75 0.75
137 849
0.75 0.75
191 200 10
0.75 0.5 0.25
888
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Fig. 6. Favourability score for CLC and slope as well as the fire probability in the study period of 1975–2013 in a Digital Terrain Model with 25 m resolution.
risk as a result of the occurrence of a given natural phenomenon of a given magnitude (Chung and Fabbri, 2005); (iii) value of elements that interact with the phenomenon under study (Chung and Fabbri, 2005; Jappiot et al., 2009). The potential damage is computed, in each pixel, by the product of the vulnerability of the elements and their respective economic value (Antunes et al., 2011). The vulnerability is defined according to the level or degree of destruction of the elements caused by fire and the vulnerability scores adopted for each vegetation type (Table 1) were the ones proposed by Antunes et al. (2011). Only the economic value of the CLC layers used to map the susceptibility were considered to estimate the potential damage. The economic value was assessed from the monetary values defined in the resolution of the Portuguese Council of Ministers no. 6-B/2015 that updates the National Strategy for the Forests (DGRF, 2006) with a horizon to 2030, where the values of the different types of forest and scrubland are set (Table 1). The final value of the potential damage reflect a monetary loss per surface area.
3. Results 3.1. Hazard assessment The fire probability as well as the CLC and slope favourability scores obtained with the longer fire historical dataset, using all the CLC inventories and high resolution DTM are shown in Fig. 6. As expected and can be seen, the higher favourability scores are present in the regions where the fire probability is also higher. However, it is important to underline that the same class of CLC or slope present the same favourability score everywhere in the maps. To compare these findings with the ones on Verde and Zêzere (2010), the temporal average of the CLC and slope favourability scores obtained in both studies were represented simultaneously in the same plot (Figs. 7 and 8). It is important to underline the increase in the favourability scores in all CLC classes except for Permanently irrigated land (212), Mixed forest (313) and Beaches, dunes, sands (331) and in
Fig. 7. Medium favourability scores for the Corine land cover (CLC) classes obtained in this study (blue columns) and in Verde and Zêzere (2010) (orange columns).
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
889
Fig. 8. Medium favourability scores for slope classes obtained in this study for a 25 m resolution (blue columns) and for the ones obtained in Verde and Zêzere (2010) for a 80 m resolution (orange columns). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
the 3 classes of lower slope. This generalized increase in the favourability scores is essentially due to the increase in the total burnt area in the study period (1.9 Mha in 1975–1994 to 4.3 Mha in 1975– 2013) since the changes in the area occupied by each CLC class are positive for some classes and negative for others but the absolute value of the relative change class area is much small than the increase in the total burnt area in the class. The increase in the favourability scores in some CLC classes is much higher than in most slope classes. The fire susceptibility obtained by integrating the three variables (CLC, slope and fire probability) for the 1975–2013 period was mapped in 5 classes with the respective predictive values and shown in Fig. 9. The most susceptible areas are located in areas with, simultaneously, high favourability scores and high fire probability. The achieved susceptibility map is very similar with the one obtained by Verde and Zêzere (2010). However the predictive values of each class are quite different.
In essence, the predicted value increase 25.6% (52% to 77.6%) for the more susceptible (‘very high’) class and decreased for all other classes (Table 2). It is very important to underline these differences because the predictive values can be used to estimate how much of the future total area will burn in each susceptibility class. Therefore, the assessment of the fire hazard was performed considering the same five susceptibility classes and for three different future scenarios of annual burnt area (Table 3), namely corresponding to about: (i) the maximum annual burnt area ever registered in Portugal, i.e., 500,000 ha; (ii) the average annual burnt area, i.e., 250,000 ha; and the minimum annual burnt area, i.e., 125,000 ha which, curiously, ranging from a factor of 2 to each other. The hazard maps are not
Table 2 Predictive values of each of the five fire susceptibility classes obtained in the present study and in the past by Verde and Zêzere (2010). Predictive Value Susceptibility class
Past
Present
Very low Low Medium High Very high
3.0 5.0 12.0 28.0 52.0
0.4 1.3 5.1 15.6 77.6
Table 3 Hazard assessment for three different future scenarios for the five considered susceptibility classes. 1st case: 500,000 ha will burn in a year; 2nd case: 250,000 ha will burn in a year; 3rd case: 125,000 ha will burn in a year. Hazard (%)
Fig. 9. Fire susceptibility map for mainland Portugal with 5 classes defined by the quintiles of the susceptibility values in each pixel.
Susceptibility class
1st case
2nd case
3rd case
Very low Low Medium High Very high
0.14 0.32 1.66 4.19 23.02
0.07 0.16 0.83 2.09 11.51
0.04 0.08 0.42 1.05 5.76
890
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
Table 4 Hazard assessment for the annual burnt area scenario of 500,000 ha in the past (Verde and Zêzere, 2010) and in this present study. Susceptibility class
Very low Low Medium High Very high
inner region of the country will have the lower fire risk and the central region the higher fire risk classes. 4. Discussion
Hazard (%) Past
Present
0.85 1.40 3.38 8.42 16.81
0.14 0.32 1.66 4.19 23.02
shown because their spatial patterns are exactly equal to the susceptibility map and only differing in the values of each class (Table 3). The hazard for the worst case scenario (annual burnt area of 500,000 ha) can be compared with the estimates by Verde and Zêzere (2010) revealing that the probability of a pixel burn and belonging to the very high class increased 6.2% (from 16.8% to 23.02%) while the probability of a pixel burn and belonging to any other class decreased − 70% in the very low class and about − 20% in the other classes (Table 4) 3.2. Potential damage and fire risk assessment The vulnerability values described in subsection 2.3.2 were mapped along with the potential damage computed with the economic values presented in the same section (Fig. 10). In general, there is a good spatial agreement between the vulnerability and potential economic damage in the sense that the regions with higher vulnerability also present higher economic potential damage. When computing the fire risk by integrating the fire hazard with the potential economic damage a wide range of values can be obtained as a consequence of the different potential economical values (0–390 €/ha) and hazard probability for the five classes of susceptibility and three burnt area case scenarios (0.1–19.6). To map the fire risk the obtained values can be classified in different classes using the quintiles or other method. However, to keep the association with the susceptibility map and to respect the Portuguese law, the fire risk was mapped using the same five classes and the three case scenarios considered in the previous subsection. The results of the risk assessment (Fig. 11) includes the fire risk probability as well as the weighted economic potential damage and reveal that whatever the case under study the south-
This study aims to map the fire risk in Portugal and for that purposes adopted a consensual conceptual framework within the fire research community (Bachmann and Allgöwer, 1999; Hardy, 2005), the methodology of Verde and Zêzere (2010) for the assessment of the wildfire susceptibility and hazard and the approach of Antunes et al. (2011) to assess the potential economic damage. At the first glance, the susceptibility mapping method seems to be a very simple model, based on just two variables – slope and vegetation cover – but, in fact, this is a very robust and resistant model for the following reasons: (i) the authors (Verde and Zêzere, 2010) tested (alternatively and/or additionally) other driver variables, namely elevation, temperature and precipitation, which are usually recognized as among the most important vegetation fire conditioning factors, without any significant increase in the prediction rates; and (ii) some of these variables can be proxies of each other. For example, slope is a measure of the altitude change. The altitude or elevation regulates the rainfall and temperature which, consequently, define the climatic/weather conditions that, in turn, determines the existence, type and state of the vegetation in each location. This means that vegetation cover is also a proxy for climate, altitude, and viceversa, i.e., when using the vegetation cover the climate information are also being considered in spite of not explicitly. Most likely because of its robustness this susceptibility model/map was adopted by the Portuguese authorities (i.e. ICNF and IPMA). As expected, the favourability scores are higher in specific CLC classes, namely Open spaces with little or no vegetation (explicitly, 333 Sparsely vegetated areas and 332 Bare rocks) and Scrub and/or herbaceous vegetation associations (specifically, 321 Natural grasslands and 322 Moors and heathland) as well as in the classes of higher slope (N 10°). In the first case, these results are in good agreement with the findings of (Pereira et al., 2014a) while in the second case it is a natural consequence of slope be one of the variables that most enhances the spread of fire not only because it favours the energy propagation and the ignition of adjacent combustible but also because these regions are generally more isolated and with less access for firefighting (Morandini et al., 2014; Tihay et al., 2014). It is also important to underline that the favourability scores do not necessarily allows to identify the classes that are most affected by fire in absolute terms but, since is
Fig. 10. Spatial distribution of vulnerability and potential economic damage for mainland of Portugal.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
891
Fig. 11. Spatial distribution of fire risk for the 3 cases in study as function of medium economic value for the quintiles of the susceptibility values in each pixel. 1st case: 500,000 ha will burn in a year; 2nd case: 250,000 ha will burn in a year; 3rd case: 125,000 ha will burn in a year.
defined as the ratio between the burnt area in that class and the total area of the class, is the class fire proneness, i.e., a relative measure of how each class is affected by fire (Pereira et al., 2014a). Another aspect that deserves some attention is the use of the areas burnt by fire in various stages of the framework: twice in the evaluation of the susceptibility, in particular (1) to compute the favourability scores and (2) explicitly in the expression of susceptibility in the form of fire probability in each pixel; the probability of fire turn is again used in the same sequence (3) to determine the fire hazard. The later use is a consequence of the fire hazard definition but the first two are also very easy to justify. The first use is to rank the CLC and slope classes in terms of their fire proneness while the second is to discriminate where, within the country, each class is more or less affected. On this respect it also worth noting that the fire probability is a measure of the fire recurrence but also a proxy for the human behaviour since the large majority of the fires are caused by the humans (Verde and Zêzere, 2010). It is also important to discuss the consequences of using a DTM with higher resolution, several inventories of land use and vegetation cover, and longer fire history. The use of a high resolution DTM allows mapping the susceptibility with higher spatial accuracy as a consequence of assessing all aspects of the fire risk (favourability scores, fire probability, etc.) with higher spatial resolution. Changes in the area occupied by each CLC class from 2000 to 2006 were recently assessed for Portugal (Pereira et al., 2014a) revealing a significant decrease (of about −20%) in the area of broad-leaved and coniferous forests as well as of the non-vegetated areas but also an even more substantial increase (20%–30%) in the artificial surfaces and in the area of sclerophyllous vegetation. This is a consequence of several factors such as rural abandonment, urbanization expansion and also vegetation fires and must be underlined also because of occur in a very short period (7 years) in comparison to the entire study period (39 years) which coincides with the period of higher changes in almost all aspects of the country. This means that the computation of the favourability scores may be biased. That is the main reason why, in this study, the favourability scores were computed using the three CLC inventories and not just one. Nevertheless, in a more recent study (Verde, 2015), the robustness of this susceptibility model was assessed in what respects to the use of single or multiple CLC inventories and the obtained results points to a relative independence of the model performance in relation to how many or which CLC inventory are/is used to access the favourability scores. This could be due to two reasons: the effective good quality of the model or the use of five classes to map the fire susceptibility which could mask the effects of eventual changes in the area of the CLC classes. Eventually, if those changes were taken into account, the estimated favourability scores could be different and imply changes in the sorting of the CLC classes (in terms of their susceptibility) but without implying
changing between susceptibility classes. However we are more convinced of the good quality of the model not just because of its essentially deterministic nature, large amount of reliable data, all tests and comparisons performed previously (Verde and Zêzere, 2010; Verde, 2015), because most significant changes in the CLC classes' areas were in the classes excluded in this study or not in the ones with higher fire proneness/favourability (Pereira et al., 2014a) but, essentially, for the large amount of high quality data used to calibrate and validate the model. In fact, one of the most conditioning factors when developing a model is the quality but also the quantity of data available which, in addition, need to be split for the independent calibration, testing and validation procedures (Pereira et al., 2013). The amount of data constrains the number of statistical significant predictors that can be used in a model, calibration data should be sufficient to assure the stable estimation of model parameters Validation should be performed with data variability within the range of values where the model will operate (Knight and Shamseldin, 2005; Ogundimu et al., 2016). However, this study uses 27 predictors (21 CLC and 6 slope classes) but 134 million pixels to model the fire susceptibility and risk. Another important and definitive indicator of the good quality of the Verde and Zêzere (2010) model is the increasing performance when using more/new/different data for an extended/future period. In general, the use of new data in a model (e.g. validation or prediction) lead to a decrease in the goodness of fit statistics obtained during the tuning procedure (i.e., calibration or test procedure). However, the use of a longer fire historical dataset in this study leads to an increase of the favourability scores, especially evident in the classes of higher susceptibility and, which is more relevant, to an increase of the predictive value for the very high susceptibility class at the expense of the decrease of the predictive value of lower susceptibility classes. In fact, the good quality of the susceptibility map is of fundamental importance since it determines the quality of the fire risk map. Finally, it is important to compare the obtained results with the findings of previously published studies. However, most of the previous fire risk assessments for Portugal were performed for different fire risk concepts, non-structural fire risk aspects, specific vegetation types or small regions such as assessing the relative importance of different factors on the spatial distribution of fire ignitions (Catry et al., 2008), annual fire risk maps fire as the probability of its occurrence (Turkman et al., 2014), fire probability in forestry stands (da Costa Ricardo, 2010), or assessing and mapping (structural, dynamical and/or integrated) fire risk in specific counties or districts (Antunes et al., 2011; Caetano et al., 2004; Teodoro and Duarte, 2013). Results for entire mainland Portugal includes: the forest fire risk at county level (Lourenco, 1994), the fire risk map adopted by the Portuguese Forest Service (Pereira and dos Santos, 2003). All these maps were produced with much short fire history, mostly before the sequence of devastating fires in 2003 and 2005,
892
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893
and have a much lower spatial resolution to be properly compared with the results of this study. 5. Conclusion In summary, all objectives of this study were achieved in the sense that a fire risk map for the vegetative areas was produced for Portugal, based on an updated susceptibility map and the most important aspects of the adopted methodology were discussed. The implemented framework for the fire risk assessment is the most recent and gathers greater consensus within the research community. The susceptibility mapping model of Verde and Zêzere (2010) was adopted for its excellent performance, be in accordance with the Portuguese law and adopted by the national authorities. The susceptibility of the territory was derived from vegetation cover, slope and fire probability and mapped with just five classes. This model was updated in this study using higher spatial resolution DTM, longer and up to date fire history as well as several Corine land cover inventories. Updated susceptibility map is extremely similar to the previous in terms of the spatial patterns of susceptibility classes confirming the good quality of this model which present higher performance when using new fire data and higher resolution elevation map. In fact, the predictive value increase for the very high susceptibility class and decrease for the other classes of lower susceptibility. The fire hazard was estimated for three different scenarios of burnt area corresponding to years with minimum, mean and maximum annual burnt ever registered in the study period. The potential economic damage was computed with the vulnerability scores proposed for the different vegetation types in Portugal by Antunes et al. (2011) and the economic value of the vegetation layers defined by the Portuguese law (CIAF, 2015) which is based on the National Strategy for the Forests for the 2030 horizon. The obtained fire risk map is an estimate of the structural vegetation fire risk for the 1975–2013 period using a deterministic approach to compute the susceptibility and the potential economic damage with the highest possible spatial resolution (25 m) at this moment. This fire risk map reveal that the area with higher risk is located in the central and north part of the country as well as in extreme south (Algarve) whatever the burnt area scenario considered for the future. On the contrary the area with lower fire risk is located in the south of the country, specifically in the interior of Alentejo. These results are due mainly to the landscape and vegetation cover conditions conducive to the development of vegetation fires present in these two regions that do not exist in other areas of the country. This map could be very useful to take more effective the measures to prevent and combat vegetation fires, with the contribution and participation of all the agents of the territory management, including the populations. The plan for future work includes the assessment of the usefulness of other variables (e.g., meteorological, socioeconomic) to improve the susceptibility map, to update the vulnerability map, to consider human, landscape and ecological potential damage and, finally, to include the dynamical risk factors, namely weather conditions. Acknowledgements This work was supported by: (i) the Herbette Foundation of the University of Lausanne; (ii) the FIREXTR project, PTDC/ATPGEO/0462/ 2014; (iii) the project Interact - Integrative Research in Environment, Agro-Chain and Technology, NORTE-01-0145-FEDER-000017, research line BEST, co-financed by FEDER/NORTE 2020; and, (iv) European Investment Funds by FEDER/COMPETE/POCI – Operacional Competitiveness and Internacionalization Programme, under Project POCI-010145-FEDER-006958 and National Funds by FCT - Portuguese Foundation for Science and Technology, under the project UID/AGR/04033. We are especially grateful to ICNF and ISA for providing the fire data and to João Pereira for the final spelling and grammar review of the manuscript.
References André, J., 2008. Probabilidades e Estatística para a Engenharia. Lidel. Antunes, C.C., Viegas, D.X., Mendes, J.M., 2011. Avaliação do Risco de Incêndio Florestal no Concelho de Arganil. Silva Lusitana. 19, pp. 165–179. Aune-Lundberg, L., Strand, G.-H., 2010. CORINE Land Cover 2006. The Norwegian CLC2006 project. Report from the Norwegian Forest and Landscape Institute. 11, p. 2010. Bachmann, A., Allgöwer, B., 1999. The need for a consistent wildfire risk terminology. The Joint Fire Science Conference and Workshop: Crossing the Millennium: Integrating Spatial Technologies and Ecological Principles for a New Age in Fire Management. 1, pp. 67–77. Blanchi, R., Jappiot, M., Alexandrian, D., 2002. Forest fire risk assessment and cartography. A methodological approach. Proceedings of the IV International Conference on Forest Fire Research. Luso, Portugal. Citeseer. Bodí, M.B., Martin, D.A., Balfour, V.N., Santín, C., Doerr, S.H., Pereira, P., et al., 2014. Wildland fire ash: production, composition and eco-hydro-geomorphic effects. Earth Sci. Rev. 130, 103–127. Caetano, M., Freire, S., Carrão, H., 2004. Fire risk mapping by integration of dynamic and structural variables. Remote Sens. Transit. 1. Caetano, M., Nunes, V., Nunes, A., 2009. CORINE Land Cover 2006 for Continental Portugal. Relatório técnico. Instituto Geográfico Português, Lisbon, Portugal. Catry, F., Rego, F., Moreira, F., Bacao, F., 2008. Characterizing and modelling the spatial patterns of wildfire ignitions in Portugal: fire initiation and resulting burned area. WIT Trans. Ecol. Environ. 119. Catry, F.X., Rego, F.C., Bação, F.L., Moreira, F., 2010. Modeling and mapping wildfire ignition risk in Portugal. Int. J. Wildland Fire 18, 921–931. Cerdá, A., Doerr, S.H., 2005. Influence of vegetation recovery on soil hydrology and erodibility following fire: an 11-year investigation. Int. J. Wildland Fire 14, 423–437. Chung, C., Fabbri, A.G., 2005. Systematic procedures of landslide hazard mapping for risk assessment using spatial prediction models. Landslide Hazard and Risk. Wiley, New York, pp. 139–177. Chuvieco, E., Congalton, R.G., 1989. Application of remote sensing and geographic information systems to forest fire hazard mapping. Remote Sens. Environ. 29, 147–159. Chuvieco, E., Aguado, I., Yebra, M., Nieto, H., Salas, J., Martín, M.P., et al., 2010. Development of a framework for fire risk assessment using remote sensing and geographic information system technologies. Ecol. Model. 221, 46–58. Chuvieco, E., 2009. Earth Observation of Wildland Fires in Mediterranean Ecosystems. Springer. CIAF, 2015. Resolução do Conselho de Ministros no. 6-B/2015. Diário da República, p. 24. da Costa Ricardo, A., 2010. Modelação da probabilidade de ocorrência de incêndio em povoamentos florestais de Portugal Continental. Universidade Técnica de Lisboa, Lisboa. DGRF, 2006. Resolução do Conselho de Ministros n.o 114/2006. Diário da República, 1.a série, p. 179. DGT, 2016. Corine Land Cover (CLC). DL, 2006. Decreto-Lei no. 124/2006 de 28 de Junho. Diário Da República — I SÉRIE-A, p. 123. Donato, D., Fontaine, J., Campbell, J., Robinson, W., Kauffman, J., Law, B.E., 2006. Post-wildfire logging hinders regeneration and increases fire risk. Science 311, 352. EEA, 2002. Corine Land Cover 1990 (CLC1990) and Corine Land Cover Changes 1975– 1990 in a 10 km Zone Around the Coast of Europe. EEA, 2014. Corine Land Cover 2000 Seamless Vector Data. Fernandes, P., Luz, A., Loureiro, C., Ferreira-Godinho, P., Botelho, H., 2006. Fuel modelling and fire hazard assessment based on data from the Portuguese National Forest Inventory. For. Ecol. Manag. 234, S229. Finney, M.A., 2005. The challenge of quantitative risk analysis for wildland fire. For. Ecol. Manag. 211, 97–108. Finney, M.A., McHugh, C.W., Grenfell, I.C., Riley, K.L., Short, K.C., 2011. A simulation of probabilistic wildfire risk components for the continental United States. Stoch. Env. Res. Risk A. 25, 973–1000. Freire, S., Carrão, H., Caetano, M.R., 2002. Produção de cartografia de risco de incêndio florestal com recurso a imagens de satélite e dados auxiliares. Lisboa, IGP. Goel, D., Pandey, N., Gupta, S., 2015. Proprietary and Open Source Geospatial Software. Organized by Department of Civil Engineering 48. Indian Institute of Technology (Banaras Hindu University), Varanasi-221005, Uttar Pradesh, India. Goncalves, J., Morgado, A., 2008. Use of the SRTM DEM as a geo-referencing tool by elevation matching. Int. Arch. Photogramm. Remote. Sens. Spat. Inf. Sci. 37, 879–883. Gralewicz, N.J., Nelson, T.A., Wulder, M.A., 2012. Factors influencing national scale wildfire susceptibility in Canada. For. Ecol. Manag. 265, 20–29. Hardy, C.C., 2005. Wildland fire hazard and risk: problems, definitions, and context. For. Ecol. Manag. 211, 73–82. Hedo, J., Lucas-Borja, M., Wic, C., Andrés-Abellán, M., de Las Heras, J., 2015. Soil microbiological properties and enzymatic activities of long-term post-fire recovery in dry and semiarid Aleppo pine (Pinus halepensis M.) forest stands. Solid Earth 6, 243. ICNF, 2016a. Cartografia de Risco - mapa de perigosidade de Incêndio Florestal. ICNF, 2016b. Informação Geográfica. IGP, 2016. Dados Abertos. INE, 2012. Censos 2011 resultados definitivos-Portugal. Instituto Nacional de Estatística, IP, Lisboa-Portugal. IPMA, 2016. Classes de Risco de Incêndio por Concelho. RCM - Hoje. Jaiswal, R.K., Mukherjee, S., Raju, K.D., Saxena, R., 2002. Forest fire risk zone mapping from satellite imagery and GIS. Int. J. Appl. Earth Obs. Geoinf. 4, 1–10. Jappiot, M., Gonzalez-Olabarria, J., Lampin-Maillet, C., Borgniet, L., 2009. Assessing wildfire risk in time and space. Living with wildfires: what science can tell us. A Contribution to the Science-Policy Dialogue, pp. 41–47.
J. Parente, M.G. Pereira / Science of the Total Environment 573 (2016) 883–893 Júnior, A.C.P., Oliveira, S.L., Pereira, J.M., Turkman, M.A.A., 2014. Modelling fire frequency in a Cerrado savanna protected area. PLoS One 9, e102380. Keesstra, S., Wittenberg, L., Maroulis, J., Sambalino, F., Malkinson, D., Cerdà, A., et al., 2016. The influence of fire history, plant species and post-fire management on soil water repellency in a Mediterranean catchment: the Mount Carmel range, Israel. Catena. Knight, D., Shamseldin, A., 2005. River Basin Modelling for Flood Risk Mitigation. CRC Press. Lasanta, T., Cerdà, A., 2005. Long-term erosional responses after fire in the central Spanish Pyrenees: 2. Solute release. Catena 60, 81–100. Leblon, B., 2001. Forest wildfire hazard monitoring using remote sensing: a review. Remote Sens. Rev. 20, 1–43. Lourenco, L., 1994. Risco de incendio florestal em Portugal Continental. Informação florestal. Morandini, F., Silvani, X., Honoré, D., Boutin, G., Susset, A., Vernet, R., 2014. Slope effects on the fluid dynamics of a fire spreading across a fuel bed: PIV measurements and OH* chemiluminescence imaging. Exp. Fluids 55, 1–12. Moriondo, M., Good, P., Durao, R., Bindi, M., Giannakopoulos, C., Corte-Real, J., 2006. Potential impact of climate change on fire risk in the Mediterranean area. Clim. Res. 31, 85–95. Ogundimu, E.O., Altman, D.G., Collins, G.S., 2016. Adequate sample size for developing prediction models is not simply related to events per variable. J. Clin. Epidemiol. Pacheco, F., Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., 2015. Controls and forecasts of nitrate yields in forested watersheds: a view over mainland Portugal. Sci. Total Environ. 537, 421–440. Parente, J., Pereira, M.G., Tonini, M., 2016. Space-time clustering analysis of wildfires: the influence of dataset characteristics, fire prevention policy decisions, weather and climate. Sci. Total Environ. 559, 151–165. Parisien, M.-A., Kafka, V., Hirsch, K., Todd, J., Lavoie, S., Maczek, P., 2005. Mapping Wildfire Susceptibility with the BURN-P3 Simulation Model. Pereira, J.M.C., dos Santos, M.T.N., 2003. Fire Risk and Burned Area Mapping in Portugal: DGF. Pereira, M., Malamud, B., Trigo, R., Alves, P., 2011. The history and characteristics of the 1980–2005 Portuguese rural fire database. Nat. Hazards Earth Syst. Sci. 11, 3343–3358. Pereira, M.G., Aranha, J., Amraoui, M., 2014a. Land cover fire proneness in Europe. 23, 13. Pereira, M.G., Calado, T.J., DaCamara, C.C., Calheiros, T., 2013. Effects of regional climate change on rural fires in Portugal. Clim. Res. 57, 187–200. Pereira, M.G., Caramelo, L., Orozco, C.V., Costa, R., Tonini, M., 2015a. Space-time clustering analysis performance of an aggregated dataset: the case of wildfires in Portugal. Environ. Model. Softw. 72, 239–249. Pereira, M.G., Fernandes, L.S., Carvalho, S., Santos, R.B., Caramelo, L., Alencoão, A., 2016a. Modelling the impacts of wildfires on runoff at the river basin ecological scale in a changing Mediterranean environment. Environ. Earth Sci. 75, 1–14.
893
Pereira, P., Cerdà, A., Lopez, A.J., Zavala, L.M., Mataix-Solera, J., Arcenegui, V., et al., 2016b. Short-term vegetation recovery after a grassland fire in Lithuania: the effects of fire severity, slope position and aspect. Land Degrad. Dev. Pereira, P., Cerdà, A., Úbeda, X., Mataix-Solera, J., Arcenegui, V., Zavala, L., 2015b. Modelling the impacts of wildfire on ash thickness in a short-term period. Land Degrad. Dev. 26, 180–192. Pereira, P., Mierauskas, P., Novara, A., 2014b. Stakeholders' perceptions about fire impacts on Lithuanian protected areas. Land Degradation & Development. Renard, Q., Pélissier, R., Ramesh, B., Kodandapani, N., 2012. Environmental susceptibility model for predicting forest fire occurrence in the western Ghats of India. Int. J. Wildland Fire 21, 368–379. Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., Pacheco, F., 2015a. A framework model for investigating the export of phosphorus to surface waters in forested watersheds: implications to management. Sci. Total Environ. 536, 295–305. Santos, R., Fernandes, L.S., Pereira, M., Cortes, R., Pacheco, F., 2015b. Water resources planning for a river basin with recurrent wildfires. Sci. Total Environ. 526, 1–13. Shaw, E.A., Denef, K., de Tomasel, C.M., Cotrufo, M.F., Wall, D.H., 2016. Fire affects root decomposition, soil food web structure, and carbon flow in tallgrass prairie. Soil. Tedim, F., Carvalho, S., 2013. A vulnerabilidade aos incêndios florestais: reflexões em torno de aspetos conceptuais e metodológicos. Territorium. Tedim, F., Garcin, M., Vinchon, C., Carvalho, S., Desramaut, N., Rohmer, J., 2014. Comprehensive vulnerability assessment of forest fires and coastal erosion: evidences from case-study analysis in Portugal. Assessment of Vulnerability to Natural Hazards: A European Perspective. 149. Teodoro, A.C., Duarte, L., 2013. Forest fire risk maps: a GIS open source application–a case study in Norwest of Portugal. Int. J. Geogr. Inf. Sci. 27, 699–720. Tihay, V., Morandini, F., Santoni, P.-A., Perez-Ramirez, Y., Barboni, T., 2014. Combustion of forest litters under slope conditions: burning rate, heat release rate, convective and radiant fractions for different loads. Combust. Flame 161, 3237–3248. Turkman, K., Turkman, M.A., Pereira, P., Sá, A., Pereira, J., 2014. Generating annual fire risk maps using Bayesian hierarchical models. J. Stat. Theory Pract. 8, 509–533. Van Eck, C.M., Nunes, J.P., Vieira, D., Keesstra, S., Keizer, J.J., 2016. Physically-based modelling of the post-fire runoff response of a Forest catchment in Central Portugal: using field versus remote sensing based estimates of vegetation recovery. Land Degrad. Dev. Verde, J., Zêzere, J., 2010. Assessment and validation of wildfire susceptibility and hazard in Portugal. Nat. Hazards Earth Syst. Sci. 10, 485–497. Verde, J.C.R.M., 2015. Wildfire Susceptibility Modelling in Mainland Portugal. Universidade De Lisboa.