Increases in fire risk due to warmer summer temperatures and wildland urban interface changes do not necessarily lead to more fires

Applied Geography 56 (2015) 1e12

Contents lists available at ScienceDirect

Applied Geography journal homepage: www.elsevier.com/locate/apgeog

Increases in fire risk due to warmer summer temperatures and wildland urban interface changes do not necessarily lead to more fires
s, K. Emsellem, O. Ganga, F. Moebius, D.M. Fox*, N. Martin, P. Carrega, J. Andrieu, C. Adne N. Tortorollo, E.A. Fox UMR 7300 ESPACE CNRS, University of Nice Sophia Antipolis, 98 Blvd. Edouard Herriot, BP 3209, 06204 Nice Cedex 3, France

a r t i c l e i n f o

a b s t r a c t

Article history: Available online

Forest fire frequency in Mediterranean countries is expected to increase with land cover and climate changes as temperatures rise and rainfall patterns are altered. Although the cause of many Mediterranean fires remains poorly defined, most fires are of anthropogenic origin and are located in the wildland urban interface (WUI), so fire ignition risk depends on both weather and land cover characteristics. The objectives of this study were to quantify the overall trends in forest fire risk in the WUI of the AlpesMaritimes department in SE France over a period of almost 50 years (about 1960e2009) and relate these to changes in land cover and temperature changes. Land cover for two contrasting reference catchments (236 km2 and 289 km2, respectively) was mapped from available aerial photographs. Changes in fire risk over time were estimated using statistical relationships defined for each type of WUI, where isolated and scattered housing present a greater risk than dense and very dense housing. Summer monthly temperatures and spring and summer precipitation were quantified over the same temporal period as land cover. Finally, trends in fire frequency and burned area were analyzed over a shorter 37 year period (1973e2009) due to the lack of available fire data before 1973. Fire risk associated with WUI expansion increased by about 18%e80% over the 1960e2009 period (depending on the catchment). Similarly, mean summer minimum and maximum monthly temperatures increased by 1.8
C and 1.4
C, respectively, over the same period. Summer rainfall appears to decrease over time since about the 1970's but remains highly variable. Land cover and weather changes both suggest an overall increase in fire risk. However, the number of fires and burned area have decreased significantly since about 1990. This paradoxical result is due to a change in fire-fighting strategy which reinforced the systematic extinction of fires in their early stages. Technical support in the form of improved radio communication and helicopters contributed greatly to reducing fire frequency and burned area. Surveillance and legal reforms included the introduction of field patrols and restricted access to forests during high risk periods. Although this has proven highly successful in the short term, the risk of fuel load accumulation over time remains a risk which might contribute to the development of mega-fires in extreme climatic conditions in the future. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Wildland urban interface (WUI) Land cover change Climate change Forest fires Fire risk Fire suppression

Introduction Each year, forest fires in Euro-Mediterranean countries burn hundreds of thousands of hectares (Martinez, Vega-Garcia, & Chuvieco, 2009). These fires cause or contribute to human deaths, severe property damage, and increased risks in soil erosion, runoff and downstream flooding. Repetitive fires may also contribute to

* Corresponding author. E-mail address: [email protected] (D.M. Fox). http://dx.doi.org/10.1016/j.apgeog.2014.10.001 0143-6228/© 2014 Elsevier Ltd. All rights reserved.

soil degradation and biodiversity loss. Of the 30,000 to 60,000 fires that occur annually, it is estimated that more than 90% are of human origin (Martinez et al. 2009; Oliveira, Oehler, San-MiguelAyanz, Camia, & Pereira, 2012; San-Miguel-Ayanz, Moreno, & Camia, 2013). These ignitions are related to arson and a wide range of accidental causes which vary from one region to another. Factors such as land abandonment, socio-economic transformations in rural areas, persistence of fire producing traditional activities, negligence, landscape structure, land cover, population density, forest policy, greater recreational use of forests, and other human related factors all contribute to the frequency, size and spatial

2

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

distribution of fires (Badia, Serra, & Modugno, 2011; Catry, Rego, Baç~ ao, & Moreira, 2009; Ganteaume & Jappiot, 2013; Martinez et al., 2009). Most authors agree that the wildland urban interface (WUI) plays a critical role in Euro Mediterranean forest fires (Catry et al., 2009; Chas-Amil, Touza, & García-Martínez, 2013; Lampin-Maillet et al., 2010), and there are fears that fire frequency and burned area will increase in the future as the WUI expands. The WUI corresponds to the zone where housing and natural vegetation share a common space (Theobald & Romme, 2007). Although WUI categories vary from one author to another, they can include such variables as population density, housing density, number of houses, and neighborhood characteristics (Stewart, Radeloff, Hammer, & Hawbaker, 2007), and a combination of some form of housing density and vegetation continuity is frequently used to define categories (Bar-Massada, Stewart, Hammerc, Mockrin, & Radeloff, 2013; Chas-Amil et al., 2013; Lampin-Maillet et al., 2010). Due to the wide range in fire causes and vegetation characteristics, relationships between WUI categories and fire frequency or burned area tend to be site specific, where accidental causes and arson related fires can vary greatly from one area to another (Chas-Amil et al., 2013; Lampin-Maillet et al., 2010; Pezzatti, Zumbrunnen, Bürgi, Ambrosetti, & Conedera, 2013; Syphard et al., 2007). WUI fire analyses must therefore be site specific, even though the following tendencies are true for much of the Euro Mediterranean area: dense urban areas generally present high opportunities for fire ignition but low vegetation continuity and fire propagation, and as housing density decreases and vegetation continuity increases, there are fewer causes for fire ignition but greater risks of fire propagation. Hence, fire frequency may be greater in higher density areas, but fire size may increase as housing density decreases (Curt, Borgniet, & Bouillon, 2013). Fires in low density pastoral areas are frequently related to agricultural activities, especially traditional burning by shepherds to maintain grass cover, so intentional fire rates can be high (Chas-Amil et al., 2010; Nunes, 2012). Mediterranean areas continue to attract transient (tourists), temporary (secondary homes) and permanent (including retirees) residents for cultural, environmental, and climatic reasons (Benoit & Comeau, 2005; Serra, Pons, & Sauri, 2008). Migration from other European countries tends to favor Mediterranean areas (Brunetta & Rotondi, 1996), just as aging populations tend to migrate toward coastal zones when possible (Van Eetvelde & Antrop, 2004). In a study of Tavernes in the South of France, Van Eetvelde and Antrop (2004) found that urban expansion was concentrated along the coast, and arable land in foothills was progressively abandoned to the benefit of residential and secondary housing on traditional terraced foothills. These trends can also be observed in the Cannes-Nice area in the Alpes-Maritimes department of SE France, where urban development is concentrated along the coast, but where individual villas with swimming pools inland provide an alternative to seafront property. In addition, Moreira et al. (2011) cite several studies highlighting increased fire hazard as forest and shrub land replace agricultural and pastoral land. Therefore, the WUI behind the dense coastal urban area has evolved considerably over the past decades and is expected to continue to change (Roy, Fox, & Emsellem, 2014). This has had and will continue to have major implications for forest fire risk. Climatic variables have long been known to affect fire frequency and burned area and the need for operational fire risk indices has led to considerable research on relationships between weather €lders, 2010; Moriondo et al., conditions and fire occurrence (Mo 2006; Sharples, McRae, Weber, & Gill, 2009). At finer temporal and spatial scales, temperature, rainfall, relative humidity, and wind speed have all proven to be related to fire risk in addition to soil and vegetation water contents (Baeza, De Luis, Raventos, &

Escarre, 2002; Holsten, Dominic, Costa, & Kropp, 2013). At greater temporal and spatial scales, rainfall in both fire and off seasons has been proven important. Lower rainfall during fire season increases fire risk, but greater rainfall in the off season can also increase fire risk through increased biomass production (Ganteaume & Jappiot, 2013; Oliveira et al., 2012; Zumbrunnen, Bugmann, Conedera, & Bürgi, 2009). Maximum temperature and relative humidity have also been related to fire occurrence at the European scale (Oliveira et al., 2012). Mega fires are partially driven by extreme weather conditions and can often only be extinguished when conditions improve or when there is no more fuel to burn (Ganteaume & Jappiot, 2013; San-Miguel-Ayaz et al., 2013). Climate change scenarios predict an increase in temperatures for the Euro-Mediterranean zone, and this might lead to a longer fire season and greater number of days with extreme fire danger, so there is good reason to believe fire hazard will increase significantly with climate change (ESPON, 2013; Moreira et al., 2011; Moriondo et al., 2006; Mouillot, Ratte, Joffre, Mouillot, & Rambal, 2005; Pausas, 2004; Theobald & Romme, 2007). The objectives of this study are to estimate the land cover and temperature driven forest fire risk trends over a period of about 50 years (about 1960e2009) in the Alpes-Maritimes department in SE France and to compare fire risk to fire frequency and burned area (1973e2009). Methods Site description The Alpes-Maritimes department is located in the extreme SE of France and possesses an extensive Mediterranean coastline and shared border with Italy. Total surface area is about 4300 km2 and altitudes range from sea level to 3143 m. The coastal area is highly developed and forms an almost continuous narrow band of built area. The highland area is comprised mainly of wildland and more or less remote villages. Between these two extremes is an extensive band of intermingled built and forested area in which most of the WUI is concentrated. Due to the high urban pressure in the lowlands and steep slopes and shallow soils upland, there is virtually no agriculture in the department outside local olive production. Mapping land cover for the entire department was beyond the scope of this paper, so two representative and contrasting catchments were selected for the study (Fig. 1). These are typical of the two major WUI land cover scenarios in the department: a catchment (Paillon) immediately upstream of a large city (Nice in this case) under intense urban pressure, and a catchment (Loup) subject to construction of individual villas and scattered housing in a more intermediate zone between the coastal area and mountainous hinterland. Neither the Paillon nor the Loup has any significant agricultural activity apart from occasional patches of olive groves, so both catchments can be considered natural or within some form of wildland urban interface. The northern limits of the reference catchments shown in Fig. 1 correspond roughly to the northern limit of much of the potential departmental WUI area, so the reference catchments cover about 20e30% of the departmental WUI area. The Paillon river flows directly through the center of Nice and the catchment has a surface area of about 236 km2 when measured just upstream of the main urban area (the densely urbanized part of the catchment was excluded because of its low fire risk). Urban pressure in the Paillon valley is high due to its proximity to Nice and the spatial confinement induced by local topography and the nearby sea, so it has undergone significant land cover change over time. Minimum and maximum altitudes are 0 m and 1495 m, respectively, with mean and median altitudes of 560 m and 528 m,

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

3

Fig. 1. Reference catchment limits and hydrological networks overlain on color air photos of the southern portion of the Alpes-Maritimes department. The catchment on the left is the Loup and to the right is the Paillon (just upstream of Nice).

respectively. Mean and median slopes based on a 5 m DEM are 46.2% and 45.3%, respectively. Geologically, the catchment is underlain by alternating layers of marls and limestone and some areas of sandstone and conglomerates with frequent bedrock surfaces on steeper slopes and at higher altitudes. The main channel, tributaries, and associated alluvial plains tend to be encased in narrow valleys except near the coast where the city of Nice is located. Dominant tree species are the following: Aleppo pine (Pinus halepensis Mill.), Evergreen oak (Quercus ilex Willd.), downy oak (Quercus pubescens Willd.); and Scots pine (Pinus sylvestris L.) at higher altitudes. The Loup catchment is located about 15 km west of the Paillon where it flows into the sea between the major urban centers of Nice and Antibes (Fig. 1). Urban density is low and is dominated by small villages and isolated housing. Situated at the periphery of the 20 km2 Sophia Antipolis high tech business park and the city of Grasse, the catchment has seen a sharp increase in the number of villas built in forested areas. Except for a few hectares at the outlet, the entire 289 km2 of the catchment were included in the study. Minimum and maximum altitudes are 0 m and 1776 m, respectively, with mean and median altitudes of 834 m and 960 m, respectively. Mean and median slopes based on a 5 m DEM are 30.5% and 25.2%, respectively. The geological substrate is dominated by limestone with some marl; bedrock surfaces are frequent here too in the upper part of the catchment. The relatively large alluvial plain near the outlet is only partially developed due to flood hazards, so there are no major urban centers in the catchment. Dominant tree species are similar to the Paillon but with more Scots pine (P. sylvestris L.) due to the greater area at higher altitudes. Land cover mapping data and fire risk estimation Digital data and mapping buildings A vector polygon database (BD TOPO) containing all buildings in both catchments was obtained from the national geographic serographique National or IGN). In the BD TOPO, all vice (Institut Ge buildings are digitized by the IGN for 2008, and these data were updated manually for the reference catchments based on 2009, 20 cm resolution, color and infrared ortho-rectified air photos. Each catchment was overlain with a 4 km  4 km grid and missing buildings were identified by superimposing the vector layer on the air photos; once identified, they were digitized and added to the

vector layer. In addition, the occasional false building was deleted to provide a complete corrected reference layer for 2009. Ortho-rectified, 50 cm resolution, panchromatic orthorectified air photos for 1964 and 1978 were acquired from the IGN for the Paillon catchment. Similar air photos for 1956 and 1983 were acquired for the Loup. Although the dates and time intervals are not identical for the two catchments, the first air photos represent the earliest good quality images available and the second series represent the most appropriate mid-interval images available between first and last (2009) images. In addition, a 5 m DEM (2010) was acquired from the IGN for both catchments. For the intermediate time period, the 2009 building layer was overlain on the air photos and buildings not yet built in 1978 (Paillon) and 1983 (Loup) were deleted from the layer. Buildings that existed and were subsequently demolished were added to the intermediate vector building layers. The procedure was then repeated between the intermediate and early images to construct polygon layers showing buildings for each time period and catchment. After initial layer creation, all dates were checked for errors by an undergraduate intern over an interval of a month. Defining the WUI type and associated fire ignition risk The method for defining the different types of WUI, summarized briefly here, was derived from the work of Lampin-Maillet (2009) and Lampin-Maillet et al. (2010) who calculated the number of fires per unit area in the WUI interface for selected sites in S and SE France based on building density and vegetation continuity. Using ArcGIS, building density within coalesced 100 m buffers around building polygons was used to define four WUI types: Isolated, Scattered, Dense and Very Dense; polygon WUI layers were then converted to raster format at a resolution of 5 m. A vegetation layer could be created for 2009 from the 20 cm IR ortho-photos described above. However, there was no realistic means of creating vegetation layers for the earlier panchromatic images. Therefore, a single vegetation layer was created using the NDVI from the IR images in 2009 and considered constant over the entire time period. Based on these images, the NDVI was calculated, cell size was converted from 20 cm to 5 m (to harmonize with the 5 m DEM) using pixel aggregation, and the resulting image was run through a low pass median filter. NDVI values were classified into three categories using the following thresholds: less than 0.01 for Sparse, from 0.01 to less than 0.15 for Discontinuous, and greater than 0.15 for

4

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

Continuous. Thresholds were determined by comparing the classified image to the 20 cm resolution color and infrared orthorectified air photos and through field visits. Since vegetation decreases as housing density increases, the vegetation layer probably underestimates vegetation cover in proximity to the 2009 Dense and Very Dense categories for earlier dates. Overlaying the housing and vegetation layers provided a WUI layer with 12 categories: 4 housing*3 vegetation classes. Based on statistical analyses of forest fire frequency (ignition points) and burned area, each WUI*vegetation category was associated with a fire risk (Lampin-Maillet, 2009; Lampin-Maillet et al., 2010). The indices shown in Table 1 were adapted from ignition density and burned area values published in Lampin-Maillet, 2009 and Lampin-Maillet et al., 2010 to represent relative fire risk. Based on their data, the method used here was the following: 1) ignition frequency per category expressed as number of ignitions per 1000 ha was divided by the lowest ignition rate category (base value of lowest risk category ¼ 1); 2) the same procedure was repeated for burned area ratio per category; 3) frequency and burned area ratio indices for each category were summed and rounded off to the nearest 0.5 value. Since some commercial and industrial buildings were integrated in the very dense housing category, these indices were reduced, so the index value would be less than for the dense housing category. According to Table 1, fire risk is considered greatest for isolated housing and then for scattered buildings, and it then decreases for Dense and Very Dense classes. The overall WUI risk for any time period was considered to be the sum of the ‘area per category (km2)*risk value associated with each category’ in the study catchments. Fire risk for natural areas within the catchment but not in a WUI category were ignored. Historical temperature and rainfall trends and fire risk Monthly temperature and rainfall recordings are available for the WUI mapping time frame described above from 1956 to 2009. As will be described below, the fire record does not extend as far back as the air photo and weather data as it only begins in 1973 the e fire data base (http:// with the creation of the online Prome www.promethee.com/default/incendies), so relating temperature and rainfall to fire frequency and burned area was only possible for this shorter time interval. Local climate data In order to estimate trends in temperature and rainfall, historical data were collected from weather stations near the study sites. A wide range of weather stations within the Alpes-Maritimes department was tested and trends were consistent regardless of the number of stations included. Weather stations retained for the study corresponded to the following criteria: located close to the study catchments, continuous temperature and rainfall recordings since 1956, and no significant urban heat island effects that would cause temperatures to evolve artificially over time. In all, six weather stations were identified; one of these was in the Paillon catchment, mean distance from a reference catchment limit for the other five stations was about 8.2 km, and no station was further than 15 km from a reference catchment. For each station, mean minimum (Tmin) and mean maximum (Tmax) monthly

temperatures were acquired for July and August (peak fire season temperatures). The Tmin and Tmax values represent the means of the daily Tmin and Tmax values for each month. Mean monthly temperature (Tavg) is the average of Tmin and Tmax. Monthly rainfall values for July and August were also obtained for the same time interval. For summer trends corresponding to the air photo periods, Tmin, Tmax and rainfall were averaged for the following intervals: 1956e1970, 1971e1985, 1986e2000, and 2001e2009. For the time scale corresponding to the fire record (1973e2009), average monthly temperatures and rainfall for May to September were calculated for each year between 1973 and 2009. For reasons which will become apparent in the results, two time intervals within this record were distinguished: 1973e1989, and 1990e2009. Differences between these two periods in mean monthly temperature and rainfall (raw and natural log transformed) were tested using analysis of variance with time period as the independent categorical variable and temperature and rainfall variables as dependent terms.

Estimating impacts of temperature and precipitation changes on forest fire ignition risk Forest fire risk was related to monthly temperature and rainfall the e” data base data by culling fire data from the national “Prome (http://www.promethee.com/default/incendies), a state run data collection and storage bank that registers all forest fires greater than 1 ha in Mediterranean France since May, 1973. For each fire, area burned, municipality, and fire cause (when known) are recorded. Forest fires greater than 10 ha occurring in May to September in land cover conditions similar to the study catchments between May 1973 and September 2009 (37 years) were identified and related to the nearest weather station in order to record Tmax, Tmin, and rainfall for the month of the fire and preceding months going back as far as May. Fire variables are therefore the following: fires per month, fires per year, area burned per fire, area burned per month, area burned per year and square root transformed values of fire frequency and natural log transformed values of burned area. After initial descriptive statistical analyses of fire and weather variables, fire variables for the month of August were isolated to relate to both spring and summer weather. August was selected because it represents the peak fire season in the region in terms of both frequency and area, as will be shown, and fires in August depend on weather conditions that can evolve over several weeks and impact soil and vegetation water contents. Hence, August fires may depend partly on spring and summer rainfall and temperatures in July and August. Pearson productemoment correlations were used on raw and normalized data to test for significant paired associations. Multiple linear regressions and analyses of covariance were used to test for relationships between August fire frequency (and square root of fire frequency) and the natural log of total burned area to spring and summer rainfall and temperature variables. In addition, fire variables were tested for differences between the earlier (1973e1989) and latter (1990e2009) periods.

Table 1 Fire risk values for each WUI*Vegetation category. The third row shows mean fire risk values by housing type. Isolated sparse

Isolated discont.

Isolated contin.

Scattered sparse

Scattered discont.

Scattered contin.

Dense sparse

Dense discont.

Dense contin.

Very dense sparse

Very dense discont.

Very dense contin.

5.0 7.5

10.0

7.5

3.5 4.8

4.5

6.5

2.0 2.7

3.0

3.0

1.5 2.2

2.0

3.0

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

Results WUI and forest fire ignition risk evolution The evolution in forest fire risk with land cover change is presented below by catchment. For each catchment, fire risk evolution is preceded by a concise description of building and WUI changes over time. Paillon housing density dynamics and changes in WUI forest fire risk The number of buildings in the Paillon catchment grew from 6356 in 1964 to 11,977 in 1978 and finally to 17,921 in 2009. Hence, the number of buildings in the catchment almost tripled in the 45 year period, increasing by about 88% and 50% in the two time periods, respectively (182% overall). This growth corresponds to mean rates of 401.5 and 191.7 buildings per year in the earlier (1964e1978) and latter (1978e2009) periods, respectively, and an overall growth rate of 257 buildings per year in the 45 year interval. The corresponding changes in WUI area are shown in Fig. 2a. In the earliest period (1964), Isolated housing dominates and surface area decreases as housing density increases. As housing density increases over time, the opposite trend is found in 2009 where area increases with housing density. Total catchment area included in the four housing categories was 59.3 km2 in 1964, 82.5 km2 in 1978, and 94.7 km2 in 2009. This corresponds to increases of about 39% and 15% in the two time periods, respectively, and an overall 1964e2009 increase in area of about 60%. More specifically, these changes correspond to increases of about 1.7 km2 y1 in 1964e1978, 0.4 km2 y1 in 1978e2009, and 0.8 km2 y1 overall. Fig. 2b shows the rate of change in area per housing type in ha y1 for the intervening time periods (initial, latter, overall). Isolated housing area declines sharply in 1964e1978 and then more slowly in 1978e2009, so the overall trend (1964e2009) is clearly negative. Scattered housing increases significantly in the earlier period but loses area in the latter period with a net overall trend of a slight increase. The dense category shows the greatest change of all classes and time intervals in 1964e1978 with a strong increase in area. This growth is less important in 1978e2009. Finally, the very dense category grew significantly in both periods and has the greatest mean change of all categories. Fig. 3aec maps the housing-vegetation overlays and show the spatial component of changes in housing density. Most changes occur in proximity to the main river channels and major roads and two or three large areas of change can easily be distinguished. The

5

first is in the southern part of the catchment close to the city of Nice. The second is near the center of the catchment and a third smaller patch is located in the western central portion of the catchment. Overlaying these layers on a slope layer of the catchment shows they all have in common a relatively gentle slope in addition to being located near major roads. Changes in building numbers and housing density have repercussions on forest fire risk in the WUI. The WUI indices (calculated as the sum of area per WUI category in Fig. 3 (excluding natural areas) * risk value per category in Table 1) for 1964, 1978, and 2009 are 329, 386, and 387, respectively. These values correspond to increases of about 17.2% for 1964e1978 and 0.4% during 1978 and 2009, respectively; the net 1964e2009 change in WUI fire risk is an increase of 17.7%. Loup housing density dynamics and changes in WUI forest fire risk There were far fewer buildings in the initial period (1956) in the Loup than in the Paillon with only 2111 buildings. Growth was rapid to reach 10,033 and 14,865 buildings in 1983 and 2009, respectively. Hence, the number of buildings increased more than six fold in the 53 year interval. Grow rates during the two time intervals (1956e1983 and 1983e2009) were about 375% and 48%, respectively, with an overall 1956e2009 change of 604%. In terms of buildings per year, this corresponds to values of about 293 and 186, with a net growth rate of 241 buildings per year in 1956e2009. This figure is similar to the value for the Paillon (257 bldgs. y1). Fig. 4a and b shows the evolution in housing category area over time; y scale limits are identical to Fig. 2a and b to facilitate comparison with the Paillon. Similarly to the Paillon, there is a strong tendency for area in dense and very dense categories to increase over time and the trends for these categories are generally quite similar for the two catchments despite a slight decrease in dense housing in 1983e2009. Trends, however, for the isolated and scattered categories are quite different. Contrary to the Paillon, Isolated housing remained more or less constant over time with only a small decrease in area between 1956 and 2009. Scattered housing progressed regularly over time at a nearly constant rate. Total WUI area increased from 27.0 km2 in 1956 to 63.5 km2 in 1983, and finally to 71.5 km2 in 2009. This corresponds to increases of about 1.4 km2 y1 in 1964e1978, 0.3 km2 y1 in 1978e2009, and 0.8 km2 y1 overall. These values are also quite similar to those of the Paillon. Fig. 5aec shows the evolution in the housing-vegetation overlays. Practically all the change occurs in the southern half of the catchment. As for the Paillon, slope inclination plays a major role in

Fig. 2. (a) Housing area by type and date in the Paillon. (b) Rate of change in housing area by type and time interval in the Paillon.

6

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

Fig. 3. (a) Map of 1964 housing type * vegetation categories in the Paillon. (b) Map of 1978 housing type * vegetation categories in the Paillon. (c): Map of 2009 housing type * vegetation categories in the Paillon.

defining the spatial component of housing density. The role of distance from the coast remains a major constraint here, and large areas of relatively gently sloping land in the northern half of the catchment remain unoccupied. This is not the case for the Paillon, where distance to the sea is less of a constraint due to a more efficient road network and shorter distances. The WUI forest fire risk indices for 1956, 1983, and 2009 are 167, 280.5, and 299.7, respectively. These values correspond to increases of 68.0% and 6.8% for the 1956e1983 and 1983e2009 periods, respectively, and an overall 1956e2009 increase of 79.5%, which is substantially greater than for the Paillon (17.7%).

Temperature and rainfall changes and forest fire ignition Fire data As described in the Methods, the analysis carried out below deals with fires that burned 10 ha or more, and the statistics of fire frequency and burned area correspond to this particular set of fires. In the 35 year fire record, there were 174 fires in the months of May to September which burned a total area of just over 26,500 ha. Of these, 35 fires (20% of all fires) were located in municipalities with land in the Loup catchment accounting for 2569.6 ha of burned area (about 10% of total burned area); 59 fires (about 34%) were in the Paillon for 10,216.8 ha of burned area (or about 39% of total). Hence, up to about 54% of fires and 49% of burned area of the AlpesMaritimes department are located in municipalities with land in

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

7

Fig. 4. (a) Housing area by type and date in the Loup. (b) Rate of change in housing area by type and time interval in the Loup.

the study catchments. The fires will first be considered according to monthly frequency and burned area and then by year. Monthly distribution of fires. As can be seen in Fig. 6, August corresponds to peak fire season with 46.6% of annual fires occurring in that month. About 22.4% occur in July, 17.2% in September, and 13.8% in May and June combined. Therefore, almost 70% of fires take

place during the summer months of July and August combined. The trend is even greater for larger fires. About 58.8% of fires greater than 100 ha occur in August and 19.6% in July, so almost 80% of larger fires take place in the summer. The remaining large fires can be found principally in September (19.6%) and a few in May and June (2.0%).

Fig. 5. (a) Map of 1956 housing type*vegetation categories in the Loup. (b): Map of 1983 housing type*vegetation categories in the Loup. (c): Map of 2009 housing type*vegetation categories in the Loup.

8

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

Fig. 6. Number of fires (line) and burned area (bars) per month in the Alpes-Maritimes department, 1973e2009.

Burned area distribution (Fig. 6) generally reflects fire frequency trends. The importance of large fires is clearly portrayed in Fig. 6. For August, 46.6% of fires account for 60.2% of the burned area, and in July, 22.4% of fires account for 27.2% of burned area. Hence, almost 90% (87.4%) of fire affected area is burned in the summer. The remaining burned area is shared between September (10.1%) and May and June combined (2.5%). Yearly distribution of fires and burned area. The number of fires per year is highly variable, ranging from 0 to almost 20 fires per year (Fig. 7). In general, the period before about 1990 tends to concentrate a greater number of fires. Almost three-quarters (73.6%) of all fires occur in the first 17 years of the fire record (1973e1989), so

fewer than a quarter (26.4%) are found in the last 20 years (1990e2009). Similarly, all years with 10 or more fires per year are concentrated in the initial 17 year period, during which 5 of the 17 years (29.4%) are years with 10 or more fires. None of the years afterward has as many as 10 fires per year. The yearly trend therefore shows a sharp decline in the number of fires with time, and the transition tends to occur around the end of the 1980s after several years of frequent fires. This observation is confirmed by an analysis of variance where both fire frequency and square root of fire frequency are significantly greater in 1973e1989 compared to 1990e2009 (R2 values of 0.22 and 0.25, respectively). The correlation between frequency and burned area is 0.67, and between square root of frequency and log of burned area it is 0.91. As could be expected, burned area distribution is highly skewed with a large number of relatively small fires (about 75% of fires burned < 120 ha) and few large fires (only 5% of fires burned >600 ha). This is also shown through the mean and median burned area values which are 152.4 ha and 28.0 ha, respectively. Within the yearly record (Fig. 7), 1986 stands out as a particularly devastating year with 10,622 ha burned in one season. For comparison, the second largest year (1974) had 2547 ha burned. In 1986, there were four fires greater than 1000 ha (two of these were in the Paillon) and the three largest fires on record can all be found in that year. Three of the largest five fires occurred in the Paillon and the greatest of all fires on record (3402 ha in 1986) took place in the Loup. Just as for fire frequency, the initial 17 year period shows more years with high surface areas burned. In 1973e1989, five years have cumulated burned areas greater than 2000 ha. In the latter 20 year period (1990e2009), only 2003 has a burned area greater than 2000 ha. The trends in burned area are therefore logically consistent with fire frequency. By combining frequency and area trends, the following years stand out as exceptionally high fire years: 1974, 1978, 1979, 1985, 1986, and 2003. The years with no summer fires equal to or greater than 10 ha are the following: 1992, 1995, 1996, 1997, 2002, 2008 and 2009. However, the difference between the two periods is statistically significant for neither area nor log of burned area due to the very high dispersion about the means.

Temperature and rainfall trends As described in the Methods, trends in temperature and rainfall are considered at two time scales. In the first, summer temperature and rainfall data in proximity to the potential near-coastal WUI area were analyzed to identify trends between 1956 and 2009. These dates correspond more or less to the mapped WUI interval. In the second, monthly temperatures and rainfall during the fire record period (1973e2009) were analyzed. Table 2 Trends for the mean of July and August temperatures and the sum of July and August rainfall in 1956e2009.

Fig. 7. Number of fires (line) and burned area (bars) per year in the Alpes-Maritimes department, 1973e2009. The worst fire year is 1986.

Variable

1956e1970 (15 yrs)

1971e1985 (15 yrs)

1986e2000 (15 yrs)

2001e2009 (9 yrs)

Mean Summer Tmin (
C) Median Summer Tmin (
C) Mean Summer Tmax (
C) Median Summer Tmax (
C) Mean Summer rain (mm) Median Summer rain (mm)

17.0

17.4

18.6

18.8

17.5

18.2

19.4

19.6

26.9

26.8

28.2

28.3

26.8

26.5

27.5

28.2

60.1

72.8

54.1

47.9

52.8

63.8

33.6

33.3

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

9

Weather trends in 1956e2009. In the 1956e2009 interval, summer Tmin and Tmax values show a clear warming trend over time (Table 2). Although the rate of warming is not constant, due certainly to naturally high variability in the Mediterranean climate and partly perhaps because it remains sensitive to start/stop dates of the intervals and of their duration, there is a clear trend for warmer temperatures with time in SE France. Note that the last time period is shorter than the 3 preceding intervals. Mean and median Tmin increased on average by 1.7
C and 2.0
C, respectively; and mean and median Tmax increased by 1.5
C and 1.4
C, respectively. Summer rainfall (Table 2) appears to have decreased over the same 1956e2009 period. Rainfall is typically highly variable in Mediterranean climates and there is an increase in precipitation in 1971e1985 over 1956e1970, but following trends are negative with net losses in mean and median summer rain between first and last periods of 6.2 mm and 8.8 mm, respectively, corresponding to decreases in rainfall in the order of 20e30%. Monthly and yearly temperature and rainfall trends during the 1973e2009 fire record period. As expected, mean temperature peaks in the summer with very little difference between July and August (Fig. 8), and it then drops off in September to about the same level as June. Lowest rainfall values occur in July with similar values for both June and August. The major autumn rainy season is initiated in September and values increase significantly during this period (Fig. 8). Although temperature varies little between July and August and rainfall is lowest in July, there are far more fires in August than in July as described above and in Fig. 6; this demonstrates the effect of prolonged heat on soil and vegetation water contents and biomass flammability. Yearly trends in summer average temperature and spring and summer rainfall are shown in Fig. 9. Readers can refer to Fig. 7 to see corresponding trends in fire frequency and burned area. Despite significant annual fluctuations, temperatures show a clear tendency for warming starting about the mid-70's; 7 of the 10 hottest years in 1973e2009 occur after 1993, and 4 of the 5 hottest years in the 1973e2009 period occur after 2002. In addition, analyses of variance show that despite very low R2 values (avg. 0.15), mean temperatures

Fig. 8. Mean monthly temperature (line) and rainfall (bars) per month.

Fig. 9. Mean summer temperatures (line) and spring and summer rainfall (bars) per year.

for the months of May to August are all statistically (a  0.05) significantly warmer in the 1990e2009 period than in 1973e1989. Rainfall trends over time are less clear and are marked especially by high inter-annual variation. The coefficient of variation for mean summer temperature is 0.03 while it is 0.73 and 0.75 for spring and summer rainfall, respectively. Despite a tendency for less rain in the latter period, the difference is not statistically significant for either the raw or log transformed rainfall values due to this high variability. Relationship between fires and meteorological conditions From the preceding figures and tables, it is clear that as the temperatures have increased between 1973 and 2009, there have been fewer and smaller fires. This paradoxical situation will be dealt with fully in the discussion, but it is already clear there are no significant positive correlations (nor negative) between August fire frequency or burned area (or transformed versions of these variables) and temperature and rainfall (or transformed versions of these variables). Ordering data and using Spearman rank order correlations does not improve correlations, though rank of frequency is correlated with rank of burned area with a Spearman value of 0.95. In an analysis of covariance, where Period (1 ¼ 1973e1989, 2 ¼ 1990e2009) is used as a categorical variable, and mean summer temperature, log of the sum of rainfall in May and June (Spring rain) and in July and August (Summer rain) are integrated as continuous variables, only Period is statistically significant (P < 0.05) for all dependent variables: fire frequency, square root of frequency, burned area and natural log of burned area. There is therefore no statistical relationship between weather and fire occurrence. For the 1973e1989 period, fire frequency is negatively correlated with rain for both July and August (same r value of 0.51), the sum of rain in July and August (r ¼ 0.58), and the natural log of the sum of rain in May and June (r ¼ 0.52). (All r values are similar for the square root of fire frequency.) The natural log of burned area is negatively correlated with rainfall in July (0.57). Neither frequency nor area (or transformed versions of these variables) are correlated with temperature. In addition, the big fire years (frequency or area) of 1974, 1978, 1979, 1985, and 1986 are not the

10

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

hottest/driest years of the period. Hence, there is no obvious relationship between fire frequency or burned area and monthly weather conditions in the first period. For 1990e2009, fire frequency is positively correlated with mean August (r ¼ 0.70) and summer (r ¼ 0.62) temperatures (values are similar but slightly lower for square root of frequency). Similarly, burned area is correlated with August and summer temperatures (r ¼ 0.78 and 0.65, respectively), and correlations with log of burned area are 0.83 and 0.76 for these same temperature variables, respectively. The square root of fire frequency is negatively correlated with the sum of May and June rainfall (r ¼ 0.52). During this period, the two exceptional fire years (1994 and 2003) figure among the three hottest years of the period, and 2003 had exceptionally low rainfall as well. There is therefore a relationship between extreme weather conditions and forest fires within this period, and extremely hot dry summers led to more fires and greater burned area. Discussion An underlying hypothesis of the study is that the nature of the relationship between WUI and fire risk is relatively constant over time. This is untrue since a number of fire risk factors evolve over time. Among these, we can cite the awareness and behavior of individuals living in the WUI. It can be argued that public awareness about natural risks is greater than ever before, just as others can advance that generations living closer to the land and having to count on themselves more in the face of disasters were more careful. A number of other measures have also evolved, including electrical standards and bush clearing and vegetation burning regulations which tend toward lower fire risks. Similarly, vegetation was assumed constant over time due to the impossibility of quantifying vegetation cover in the early and intermediate images, yet forest species and stand structure are important factors in determining fire risk (Fernandes, 2013). There is no commercial logging in the Alpes-Maritimes department, so much of the forest is left to evolve with little management outside local parks. Controlled burns are done systematically, at the request of shepherds mainly, but these are mostly outside the WUI zone. Within the WUI, forest clearing (elimination of underbrush and creation of open space between tree crowns) within a radius of 50 m around homes became compulsory in 2001. To our knowledge, the impacts of these measures on fire risk have not yet been quantified and they are difficult to estimate, especially as the measure is not always respected. The law stipulates that home owners must clear the forest at their own expense, even if the forest is not on their land, so many resent the financial cost. Finally, some fire breaks are maintained by the state with mixed results on fire propagation (Perchat & Rigolot, 2005). Quantifying the cumulative impacts of changes in WUI and social practices is impossible at this stage, but we consider the increase of about 18% in WUI fire risk for the Paillon represents a slight increase in risk over about 50 years (roughly 1960e2009), and the near 80% increase for the Loup corresponds to a clear increase in the same time interval, despite the potential mitigating factors described above. For the Alpes-Maritimes departmental WUI as a whole, this would represent an increase of about 40e50% in the 50 year interval. WUI fire risk between the intermediate dates (1978 and 1983) and 2009, the period corresponding roughly to the fire record, has shown a slight but less dramatic increase in fire risk since part of the urban growth contributed to increase urban density. Syphard, Clarke, and Franklin (2006) also noted a substantial increase in urban density over time in California with little or no change in WUI area. In our case, WUI area increased substantially leading to a general increase in fire risk. However, if

urban growth rates continue to slow or stagnate and urban density continues to increase as the impacts of more recent urban density and fire prevention zoning laws are felt, WUI fire risk may level off or even decrease with time. The increased temperatures and lower rainfall noted here are similar to the findings of Wastl, Schunka, Leuchnera, Pezzatti, and Menzel (2012) in the Southern Alps over the same time period, but relating fire risk to monthly temperature and rainfall data is difficult for several reasons. Most of the research work in this field uses a more appropriate daily weather scale in the place of the coarse monthly data available to us. Fire susceptible meteorological conditions can vary substantially within a short time period, so a monthly weather time scale is not the most appropriate except for extreme conditions. In addition, wind characteristics and relative humidity are extremely important in fire ignition and propagation (Baeza et al., 2002; Holsten et al., 2013), and these data were neither available nor appropriate at the monthly scale. Nonetheless, there is a trend to hotter and perhaps drier summers with time, and this should also contribute to increasing fire risk, as the extreme conditions of 1994 and 2003 show in the second temporal period. Pausas (2004) observed temperature increases in the Eastern Iberian Peninsula similar to the values in this study, so trends observed here may be consistent with much of the Mediterranean area. The original hypothesis was that forest fire frequency and burned area would evolve with changes in the WUI interface and summer temperatures, as has been reported for Portugal and other Mediterranean areas (Amatulli, Camia, & San-Miguel-Ayanz, 2013; Collins, de Neufville, Claro, Oliveira, & Pacheco, 2013; Moriondo et al., 2006; Pausas, 2004; Wastl et al., 2012). However, the results indicate that neither WUI characteristics nor summer weather are predominant in fire frequency and burned area. Looking at fire records dating from 1904 till 2006, Zumbrunnen et al. (2009) found that temperature and precipitation became less important as the fire regime shifted to more frequent human causes. Wastl et al. (2012) note fewer fires in Southern Germany despite an increase in forest fire risk due to human related factors, and Pezzatti et al. (2013) attribute a decrease in forest fires in Switzerland to a change in fire-fighting policy and prevention legislation. With this in mind, the head officer of the departmental (Alpes-Maritimes) fire-fighting service and some of his officers were interviewed, and much of the information below is drawn from this personal communication. the e database records fires that have reached a The Prome minimum size of 1 ha. Therefore, it does not represent the actual number of fire ignitions over time since all fires extinguished before reaching a size of 1 ha are not recorded. Therefore, although data show that the number of fires has decreased since the end of the 1980s, the actual number of fire ignitions may have increased but never reached the 1 ha threshold that would have put them in the e (or the 10 ha that would have included them in this Prome study), but there are no data available to verify this. After the major fires of the 1980s, especially 1986, the service adopted an official strategy of extinguishing forest fires in their initial stage. The evolution included a range of technical and strategic measures put into place around the same time. One of the more important technical adaptations was the acquisition of three firefighting helicopters in the late 1980s. They have a faster call-to-site reactivity time than airplanes and can fill and dump more frequently. This was accompanied by a major investment in radio communication technology that enabled better communication during surveillance and fire-fighting. Strategically, firefighters were sent out into the field in larger numbers with more equipment during high risk periods. Being closer to fire ignition points enabled a faster response. More recently (since 2003 in the Alpes-Maritimes), public access to forests during high fire risk periods is legally

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

forbidden, thereby reducing potential ignitions. In addition to better official surveillance, the widespread use of cell phones means that fire ignition calls from the public arrive sooner and in greater numbers. The strategy has clearly been effective and despite an increase in fire risk due to temperature and WUI changes in the Alpes-Maritimes department, there are fewer fires than before, even for the extreme conditions of 2003, the hottest/driest spring and summer on record. This is contrary to the findings of Penman et al. (2013) whose modeling results showed that weather factors, especially extreme conditions, strongly influenced fire size while fire suppression had only a minor impact in Australia. Our results indicate that fire suppression plays a major role, even in extreme weather conditions, in the Alpes-Maritimes WUI where distance to water sources for fire-fighting helicopters and planes is relatively short and vegetation continuity is considerably less extensive than in Australia. A consistent decrease in burned area due to fire suppression without significant land management (fuel load control) inevitably increases biomass which sooner or later may lead to a mega fire in extreme temperature, moisture and wind conditions (Collins et al., 2013; Fernandes, 2013; Rocca, Brown, MacDonald, & Carrico, 2014; Veblen, Kitzberger, & Donnegan, 2000). Several factors, however, distinguish the Mediterranean from the North American context. Due to high population densities, there are only rare situations where summer fires can be allowed to burn without intervention, as can be done in the U.S.A for example (Rocca et al., 2014), so suppression is the only politically justifiable response. According to modeling by Collins et al. (2013), the initial effectiveness of fire suppression, before biomass accumulation becomes a threat, lasts about 30e40 years, and statistical and modeling results by Curt et al. (2013) in Mediterranean France suggest that fire risk increases with biomass accumulation up to about 20e30 years and then flattens out or decreases; hence, fuel buildup may be less of a threat in Mediterranean forests. In the context of southern France, size, shape and connectivity of fuel all have a strong impact on fire recurrence (Curt et al., 2013). In addition, recent legislation on forest fire prevention now forbids construction in most forested areas of Mediterranean France and this will reinforce the trend for greater housing density and progressively less isolated or scattered buildings as noted above. Firefighting priorities are oriented to protecting lives first, then property, and finally the forest. Denser housing communities are easier to protect, so although fuel buildup may be a problem, this may not necessarily increase the risk to human lives or property in the current and future housing density contexts. Despite these arguments, the question of the limits of fire suppression in the face of more frequent extreme weather conditions and fuel load accumulation remains critical (Flannigan et al., 2013). Conclusion Both WUI and weather factors affect forest fire risk. In the AlpesMaritimes department of SE France, changes in the WUI characteristics and summer temperatures indicate a progressive increase in fire risk over time (about 1960e2009), but the number of fires and burned area have both decreased since the end of the 1980s. This paradox is explained by a change in fire-fighting strategy at that time which has since emphasized a rapid and massive response to fire ignitions in their early stage. Acknowledgments ne ral of the The authors express their gratitude to the Conseil Ge Alpes-Maritimes and the CSI fund of the University of Nice Sophia Antipolis for funding the acquisition of historical aerial

11

photographs. The PACA-CRIGE provided the 2009 photos free of ac cost, for which we are also grateful. We thank Colonel Bauthe (director of the Alpes-Maritimes firefighting brigade) and his colleagues for agreeing to meet with us and discuss the findings of this study. Finally, we would like to express our recognition to the anonymous reviewers for their detailed review and useful recommendations.

References Amatulli, G., Camia, A., & San-Miguel-Ayanz, J. (2013). Estimating future burned areas under changing climate in the EU-Mediterranean countries. Science of the Total Environment, 450e451, 209e222. http://dx.doi.org/10.1016/ j.scitotenv.2013.02.014. Badia, A., Serra, P., & Modugno, S. (2011). Identifying dynamics of fire ignition probabilities in two representative Mediterranean wildland-urban interface areas. Applied Geography, 31, 930e940. http://dx.doi.org/10.1016/j.apgeog.2011.01.016. Baeza, M. J., De Luis, M., Raventos, J., & Escarre, A. (2002). Factors influencing fire behaviour in shrublands of different stand ages and the implications for using prescribed burning to reduce wildfire risk. Journal of Environmental Management, 65, 199e208. http://dx.doi.org/10.1006/jema.2002.0545. Bar-Massada, A., Stewart, S. I., Hammerc, R. B., Mockrin, M. H., & Radeloff, V. C. (2013). Using structure locations as a basis for mapping the wildland urban interface. Journal of Environmental Management, 128(2013), 540e547. http:// dx.doi.org/10.1016/j.jenvman.2013.06.021. Benoit, G., & Comeau, A. (Eds.). (2005). A sustainable future for the Mediterranean: the blue plan's environment and development outlook. U.K.: Earthscan. ISBN-13 978-1-84407-259-0; ISBN-10 1-84407-259-2. Brunetta, G., & Rotondi, G. (1996). Migratory flows from southern to northern Mediterranean borders. Social Geography, 133, 65e80. Catry, F. X., Rego, F. C., Baç~ ao, F., & Moreira, F. (2009). Modeling and mapping wildfire ignition risk in Portugal. International Journal of Wildland Fire, 18, 921e931. http://dx.doi.org/10.1071/WF07123. Chas-Amil, M. L., Touza, J., & García-Martínez, E. (2013). Forest fires in the wildlandurban interface: a spatial analysis of forest fragmentation and human impacts. Applied Geography, 43(2013), 127e137. http://dx.doi.org/10.1016/ j.apgeog.2013.06.010. Collins, R. D., de Neufville, R., Claro, J., Oliveira, T., & Pacheco, A. P. (2013). Forest fire management to avoid unintended consequences: a case study of Portugal using system dynamics. Journal of Environmental Management, 130, 1e9. http:// dx.doi.org/10.1016/j.jenvman.2013.08.033. Curt, T., Borgniet, T. L., & Bouillon, C. (2013). Wildfire frequency varies with the size and shape of fuel types in southeastern France: implications for environmental management. Journal of Environmental Management, 117(2013), 150e161. http:// dx.doi.org/10.1016/j.jenvman.2012.12.006. ESPON. (2013). Territorial dynamics in Europe: Natural hazards and climate change in European regions. Territorial Observation No. 7, ISBN 978-2-919777-25-9. Fernandes, P. M. (2013). Fire-smart management of forest landscapes in the Mediterranean basin under global change. Landscape and Urban Planning, 110, 175e182. http://dx.doi.org/10.1016/j.landurbplan.2012.10.014. Flannigan, M., Cantin, A. S., de Groot, W. J., Wotton, M., Newbery, A., & Gowman, L. M. (2013). Global wildland fire season severity in the 21st century. Forest Ecology and Management, 294, 54e61. http://dx.doi.org/10.1016/ j.foreco.2012.10.022. Ganteaume, A., & Jappiot, M. (2013). What causes large fires in Southern France. Forest Ecology and Management, 294, 76e85. http://dx.doi.org/10.1016/ j.foreco.2012.06.055. Holsten, A., Dominic, A. R., Costa, L., & Kropp, J. P. (2013). Evaluation of the performance of meteorological forest fire indices for German federal states. Forest Ecology and Management, 287, 123e131. http://dx.doi.org/10.1016/ j.foreco.2012.08.03. Lampin-Maillet, C. (2009). Caract erisation de la relation spatiale entre organisation spatiale d'un territoire et risque d'incendie: Le cas des interfaces habitat-for^ ets du
se de doctorat de l'universite  Aix-Marseille, mention Letsud de la France. The ographie- Structures et dynamiques spatiales), 321 tres et Science humaines (Ge pages þ annexes. Lampin-Maillet, C., Jappiot, M., Long, M., Bouillon, C., Morge, D., & Ferrier, J.-P. (2010). Mapping wildland-urban interfaces at large scales integrating housing density and vegetation aggregation for fire prevention in the South of France. Journal of Environmental Management, 91(3), 732e741. http://dx.doi.org/ 10.1016/j.jenvman.2009.10.001. Martinez, J., Vega-Garcia, C., & Chuvieco, E. (2009). Human-caused wildfire risk rating for prevention planning in Spain. Journal of Environmental Management, 90, 1241e1252. http://dx.doi.org/10.1016/j.jenvman.2008.07.005. €lders, N. (2010). Comparison of Canadian Forest fire danger rating system and Mo national fire danger rating system fire indices derived from Weather Research and Forecasting (WRF) model data for the June 2005 Interior Alaska wildfires. Atmospheric Research, 95, 290e306. http://dx.doi.org/10.1016/ j.atmosres.2009.03.010. Moreira, F., Viedma, O., Arianoutsou, M., Curt, T., Koutsias, N., Rigolot, E., et al. (2011). Landscape e wildfire interactions in southern Europe: Implications for

12

D.M. Fox et al. / Applied Geography 56 (2015) 1e12

landscape management. Journal of Environmental Management, 92, 2389e2402. http://dx.doi.org/10.1016/j.jenvman.2011.06.028. 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. Climate Research, 31, 85e95. Mouillot, F., Ratte, J.-P., Joffre, R., Mouillot, D., & Rambal, S. (2005). Long-term forest dynamic after land abandonment in a fire prone Mediterranean landscape (central Corsica, France). Landscape Ecology, 20, 101e112. http://dx.doi.org/ 10.1007/s10980-004-1297-5. Nunes, A. N. (2012). Regional variability and driving forces behind forest fires in Portugal an overview of the last three decades (1980e2009). Applied Geography, 34, 576e586. http://dx.doi.org/10.1016/j.apgeog.2012.03.002. Oliveira, S., Oehler, F., San-Miguel-Ayanz, J., Camia, A., & Pereira, J. M. C. (2012). Modeling spatial patterns of fire occurrence in Mediterranean Europe using multiple regression and random forest. Forest Ecology and Management, 275, 117e129. http://dx.doi.org/10.1016/j.foreco.2012.03.003. Pausas, J. (2004). Changes in fire and climate in the Eastern Iberian Peninsula (Mediterranean basin). Climatic Change, 63, 337e350. Penman, T. D., Collins, L., Price, O. F., Bradstock, R. A., Metcalf, S., & Chong, D. M. O. (2013). Examining the relative effects of fire weather, suppression and fuel treatment on fire behaviour e a simulation study. Journal of Environmental Management, 131, 325e333. http://dx.doi.org/10.1016/j.jenvman.2013.10.007. Perchat, S., & Rigolot, E. (2005). Comportement au feu et utilisation par les forces de lutte des coupures de combustible touch ees par les grands incendies de la saison
res: Ed. de la Carde
re. (in French). 2003. Morie Pezzatti, G. B., Zumbrunnen, T., Bürgi, M., Ambrosetti, P., & Conedera, M. (2013). Fire regime shifts as a consequence of fire policy and socio-economic development: an analysis based on the change point approach. Forest Policy and Economics, 29, 7e18. http://dx.doi.org/10.1016/j.forpol.2011.07.002. Rocca, M. E., Brown, P. M., MacDonald, L. H., & Carrico, C. M. (2014). Climate change impacts on fire regimes and key ecosystem services in Rocky Mountain forests. Forest Ecology Management. http://dx.doi.org/10.1016/j.foreco.2014.04.005. Roy, H. G., Fox, D. M., & Emsellem, K. (2014). Spatial dynamics of land cover change in a Euro-Mediterranean catchment (1950e2008). Journal of Land Use Science. http://dx.doi.org/10.1080/1747423X.2014.898105.

San-Miguel-Ayanz, J., Moreno, J. M., & Camia, A. (2013). Analysis of large fires in European Mediterranean landscapes: lessons learned and perspectives. Forest Ecology and Management, 294, 11e22. http://dx.doi.org/10.1016/ j.foreco.2012.10.050. Serra, P., Pons, X., & Sauri, D. (2008). Land-cover and land-use change in a Mediterranean Landscape: a spatial analysis of driving forces integrating biophysical and human factors. Applied Geography, 28, 189e209. http://dx.doi.org/10.1016/ j.apgeog.2008.02.001. Sharples, J. J., McRae, R. H. D., Weber, R. O., & Gill, A. M. (2009). A simple index for assessing fire danger rating. Environmental Modelling & Software, 24, 764e774. http://dx.doi.org/10.1016/j.envsoft.2008.11.004. Stewart, S. I., Radeloff, V. C., Hammer, R. B., & Hawbaker, T. J. (2007). Defining the wildlandeurban interface. Journal of Forestry, 201e207. Syphard, A. D., Clarke, K. C., & Franklin, J. (2006). Simulating fire frequency and urban growth in southern California coastal shrublands, USA. Landscape Ecology, 22, 431e455. http://dx.doi.org/10.1007/s10980-006-9025-y. Syphard, A. D., Radeloff, V. C., Keeley, J. E., Hawbaker, T. J., Clayton, M. K., Stewart, S. I., et al. (2007). Human influence on California fire regimes. Ecological Applications, 17, 1388e1402. Theobald, D. M., & Romme, W. H. (2007). Expansion of the US wildland-urban interface. Landscape and Urban Planning, 83, 340e354. http://dx.doi.org/ 10.1016/j.landurbplan.2007.06.002. Van Eetvelde, V., & Antrop, M. (2004). Analyzing structural and functional changes of traditional landscapesdtwo examples from Southern France. Landscape and Urban Planning, 67, 79e95. http://dx.doi.org/10.1016/S0169-2046(03)00030-6. Veblen, T. T., Kitzberger, T., & Donnegan, J. (2000). Climatic and human influences on fire regimes in ponderosa pine forests in the Colorado Front Range. Ecological Applications, 10, 1178e1195. Wastl, C., Schunka, C., Leuchnera, M., Pezzatti, G. B., & Menzel, A. (2012). Recent climate change: long-term trends in meteorological forest fire danger in the Alps. Agricultural and Forest Meteorology, 162e163(2012), 1e13. http:// dx.doi.org/10.1016/j.agrformet.2012.04.001. Zumbrunnen, T., Bugmann, H., Conedera, M., & Bürgi, M. (2009). Linking forest fire regimes and climatedA historical analysis in a dry inner Alpine Valley. Ecosystems, 12, 73e86. http://dx.doi.org/10.1007/s10021-008-9207-3.