In the French Mediterranean, large fires have significant socioeconomic and
environmental impacts. We used a long-term georeferenced fire time series
(1958–2017) to analyze both spatial and temporal distributions of large
fires (LFs;
It is now unanimously agreed upon that large fires have the most significant socioeconomic and environmental impacts, threatening or damaging infrastructures, ecosystems and even costing human lives, especially in the expanding wildland–urban interfaces (WUIs) (Blanchi et al., 2014; Syphard and Keeley, 2015; Radeloff et al., 2018). However, the definitions of what can be considered a large fire are numerous (Shvidenko and Nilsson, 2000; Stocks et al., 2002; Barbero et al., 2015a; Stavros et al., 2014; Nagy et al., 2018; Tedim et al., 2018), the cutoff being arbitrary or statistically assessed. Usually, large fires represent only a small proportion of the total number of fires but they typically account for the bulk of burned area in many regions throughout the world (Stocks et al., 2002; San Miguel-Ayanz et al., 2013; Stavros et al., 2014; Barbero et al., 2014, 2015a; Ganteaume and Guerra, 2018) and in fact determine the trend and interannual variability in the total burned area.
Large fires and fire severity have increased over the past several decades across parts of the globe (Pausas and Fernández-Muñoz, 2012; Dennison et al., 2014; Stephens et al., 2014), these changes being attributed to a combination of climate change (Westerling et al., 2006; Bradstock et al., 2009; Flannigan et al., 2009; Barbero et al., 2015b; Abatzoglou and Williams, 2016) and past fire suppression (McKenzie et al., 2004; Littell et al., 2009; Miller et al., 2009). However, these upward trends are not universal and some landscapes, mostly in southern Europe, have not experienced such increases in large fires and have even showed a decline since the 1990s (San Miguel-Ayanz et al., 2013; Ruffault and Mouillot, 2015; Ganteaume and Guerra, 2018), albeit conflicting signals were found across parts of Portugal and Spain (Turco et al., 2016). This overall fire reduction has been attributed to an increased effort in fire management after the large fires in the 1980s (Turco et al., 2016; Fréjaville and Curt, 2017).
In Mediterranean systems, bottom-up drivers are generally thought to play a strong role in fire activity. Indeed, ignitions are mainly due to human activities (negligence or arson) as seen in California (Syphard and Keeley, 2015; Kolden and Abatzoglou, 2018) or in southeastern France (Ganteaume and Jappiot, 2013) where very few fires are started by lightning strikes (Ganteaume et al., 2013). Likewise, fuel structure and composition control fire spread and, therefore, the location of the largest fires (Duane et al., 2015; Fernandes et al., 2016). The fuel structure is also subject to human activities (Moreira et al., 2011), with agricultural land abandonment or systematic fire suppression leading to the buildup of a large amount of fuels (Pausas and Fernández-Muñoz, 2012). Additionally, top-down drivers including fire weather conditions can help define areas where large fires are most likely to occur (Moritz et al., 2012; Ruffault et al., 2016) but also provide windows of opportunity for fire spread. Large fires in Mediterranean climate ecosystems are often enabled by episodes of severe fire weather of varying duration that can be generated by dry and hot winds as seen in California (Abatzoglou et al., 2013; Kolden and Abatzoglou, 2018) or by dry but cold wind as seen in southeastern France (Ruffault et al., 2016). Collectively, climatic factors alongside ignition sources, fuels, but also suppression forces are thought to influence fire spread. It is noteworthy that changes in fire suppression policy over the last few decades mentioned above have induced sharp decreases in fires, partially modifying the functional relationships linking fire to climate (Curt and Fréjavill, 2018; Syphard et al., 2017), and thus decreasing fire activity independently of the climate forcing (Hawbaker et al., 2013; Syphard et al., 2007).
We focused here on the French Mediterranean, the most fire-prone region of
France, where the largest fire on record reached 11 580 ha despite a highly
fragmented landscape. This is also a highly populated area characterized by
an extensive WUI and high network density, which are highly impacted by fire
ignitions, especially in the western part (Ganteaume and Long-Fournel, 2015)
with the potential for several consecutive reburns. The region includes
plant communities well adapted to Mediterranean climate conditions that
confer a high fire risk on this area but an increase in fire recurrence and
a shortening of the period between fires were shown to impact vegetation
structure, especially with the decrease in mature tree cover (Ganteaume et
al. 2009), including the loss of resilience of
Additionally, little attention has been devoted to understanding the spatial distribution of large fires along a longitudinal transect. From a bottom-up perspective, fire-prone areas along the Mediterranean coast have been extensively built up in the western part of the region, thereby reducing the availability of fuel while increasing the probability of human-started fires (Ganteaume et al., 2013). From a top-down perspective, climatological annual precipitation (wind speed) is increasing (decreasing) eastwards, gradually lowering the weather-induced fire danger. How these two factors, namely fuel continuity and fire weather, modulate the occurrence of large fires is still unclear.
Previous works in the French Mediterranean were based on gridded fire data commencing from the mid-1970s (e.g., Ruffault et al., 2016; Fréjaville and Curt, 2017; Ganteaume and Guerra, 2018; Lahaye et al., 2018). Here, we used, for the first time, longer time series of georeferenced fires extending back to 1958 and sought to examine both spatial and temporal distributions of large fires across the French Mediterranean. More specifically, this paper has a threefold objective. First, we sought to identify the locations associated with large fire recurrence and quantify the spatial extent of the region with reburns. Second, we sought to establish the mean fire extent and the fire return level along a longitudinal transect spanning the French Mediterranean and identify the possible role of climate conditions and fuel continuity in shaping this longitudinal gradient. This exploratory analysis may provide some insights into a fire aspect that was overlooked in previous studies. Finally, building on previous research, we sought to re-estimate trends in large fires across the region taking advantage of a fire record spanning almost 6 decades.
The study area (total surface area of 11 157 km
The two parts of the study area (Fig. 1), located on a west–east gradient of
the Mediterranean, share most climate characteristics albeit annual
precipitation (wind speed) increases (decreases) eastwards (Ruffault et al.,
2017). These areas also differ in the structure of landscapes; forested
massifs are larger in the eastern zone while the proportion of WUIs (15 %
vs. 7 %; Ganteaume and Long-Fournel, 2015, and Ganteaume, unpublished
data, respectively) and urbanization are higher in the western area (396 vs.
176 inhabitants km
Map of the study area. The boundary between the western and the eastern regions is also indicated. Forested systems in green were extracted
from the “BD Forêt 2014” of the National Geographic Institute
(
Large fires in the French Mediterranean have already been studied in previous works using shorter time series based on the gridded regional fire database Prométhée that has recorded fires since 1973 (Fréjaville and Curt, 2015; Ruffault and Mouillot, 2017; Ruffault et al., 2018). However, these gridded data provide neither the fire perimeter needed to assess reburns nor the temporal length needed to assess return periods in large fires. Here, we used the georeferenced fire perimeter database compiled by the Office National des Forêts (ONF) and Directions Départementales des Territoires et de la Mer (DDTM Bouches du Rhône and Var) available from 1961 to 2017 in the western part and from 1958 to 2016 in the eastern part of the study area. Fire perimeters were derived from aerial photography and remote sensing (the latter since 2016) and confirmed by ground truth targeting mostly fires larger than 10 ha in the earliest period. Approximate perimeters of older fire events (i.e., before 1990) have been corrected using aerial photos and Landsat satellite images when available (i.e., a more accurate delineation of fire perimeter adjustment was performed) (Faivre, 2011).
We focused on large fires
We computed the daily Fire Weather Index (FWI) from the Canadian Forest Fire Weather Index system using daily surface meteorological variables at an 8 km spatial resolution from the quality-controlled SAFRAN dataset providing maximum temperature, minimum relative humidity, precipitation and wind speed over France from 1959 to 2017 (Vidal et al., 2010a, b, 2012). The FWI computation usually requires noon observations. However, given that SAFRAN is a daily meteorological database, we calculated FWI using maximum temperature and minimum relative humidity as surrogates of noon observations following prior analyses (e.g., Jolly et al., 2015; Abatzoglou et al., 2018). Although the FWI was empirically calibrated for estimating whether atmospheric conditions and fuel moisture content are prone to wildfire development in Canada (Van Wagner, 1987), the FWI has already proven useful to track large fires in Mediterranean regions (Dimitrakopoulos et al., 2011) including the French Mediterranean (Barbero et al., 2019). Grid cells of the FWI lying within the study area were first averaged across the June–September season and then averaged across all latitudes spanning the region of interest to form a longitudinal cross section of mean summer FWI conditions.
We extracted fuel cover data from the “BD Forêt 2014” of the National
Geographic Institute (
Based on a sequence of 58 layers of annual large fire scars covering the 1958–2017 period, the following fire attributes were extracted: (i) fire recurrence or the number of fires that occurred at the same location over the period studied and (ii) time since the last fire.
Time since the last LF (cat_age in years).
Comparisons of means in burned areas due to LFs were performed using a
nonparametric Mann–Whitney test, and a
Monotonic trends in LF frequency and in burned area due to LFs were assessed using the nonparametric Mann–Kendall test (Kendall, 1975), and a change point detection test (standard normal homogeneity test, SNHT; Alexandersson and Moberg, 1997) was used to identify potential abrupt changes in the time series.
We estimated annual maximum burned area (AMBA) return levels in the eastern
and western parts of the study area using the so-called block (here 1 year)
maxima approach. We extracted the AMBA in both areas and selected the type
of distribution that best fitted both series using the Akaike information
criteria (AIC). In both areas, the gamma distribution was found to best
describe the AMBA series. Using this distribution, the inverse cumulative
distribution was calculated, allowing the determination of the theoretical
quantiles from which we derived the return levels (AMBA) associated with
different return periods ranging from 5 to 100 years. Asymmetric confidence
intervals were calculated using a resampling approach. This approach
consists in creating new sub-samples from the original sample (75 % of the
original sample is extracted at random) using a bootstrapping process with
replacement and then estimating a return level for each of the resampled
data (
Fire recurrence in the 1961–2017 and 1958–2016 periods in the western and eastern parts, respectively.
In total, 353 LFs were recorded in the region (194 in
the western part and 159 in the eastern part; Chi2
Statistics on fires (
Regarding the LF age distribution (Fig. 2), the most frequent LFs belonged to
the 31–40-year class resulting in the most LF-prone decade (ANOVA,
A total
surface area of 311 447 was burned during the period studied, of which
21 % occurred on a surface that already burned in the past (Fig. 3), due
to multiple overlaps in burned areas by recurrent fires (i.e., LF occurrence
on the same surface). LF reburns occurred up to six times in the east but
represented only a small part of the recurrence (0.3 %; Table 2). One to
two reburns were the most frequent patterns in the western part of the study
area (39.4 % and 39.9 % of the recurrence, respectively; Table 2) while in
the east, most reburns occurred only once (46.3 %). The surface impacted
by only one LF represented 74.5 % and 71.2 % of the total area burned by
LFs in the west and the east, respectively (Table 2), and the resulting median LF extent was significantly larger in the east than in the west (668.5 vs. 346.8 ha,
respectively; comparison of medians,
Percentages of burned area (relative to the total burned area)
affected by recurrent LFs and percentages of recurrence relative to the LF
frequency (when number
The mean LF extent varied along a longitudinal gradient, increasing from the west to the east (Fig. 4a, left axis). This signal contrasts with the mean summer FWI gradient decreasing towards the east but is consistent with the sharp increase in biomass towards the east (Fig. 4b). When normalizing by the biomass area (Fig. 4a, right axis), the mean LF extent remains stable throughout the longitudinal gradient, except at the eastern edge of the study area, where a sharp increase in LF extent per biomass surface unit was evident. This may be indicative of a stronger fuel connectivity at the eastern edge, regardless of absolute biomass surface available to burn. Overall, these results suggest that LF spread is not limited by climate conditions across the region but strongly fuel-limited in the west, due to landscape fragmentation and the high proportion of WUI (15 %). Indeed, the landscape has undergone substantial transformation with time in the western part, contributing to reduce fuel cover and thereby the potential for fire spread. This highlights the role of fuel quantity and continuity in fire spread as shown in previous research (Hargrove et al., 2000; Finney et al., 2007) and the need to include fuel cover in future projections of fire activity.
A significant decline in annual LF frequency alongside area burned by LF was
found across the region according to a Man–Kendall test (Fig. 5). This
overall decline is consistent with a significant change point in both LF
metrics in 1991 as shown in previous findings (Fox et al., 2015; Ruffault
and Mouillot, 2015). This signal was especially evident in the eastern part
(Fig. 5c) while neither a change point nor a significant trend
(
Figure 6 shows the AMBA in each part of the study area (panel a) as well as
the gamma distribution models that were found as the best fit to the data
(panels c, d). Estimates of AMBA return intervals show that a
LF
Improving our understanding of large fire activity is of utmost importance to fire prevention and management to mitigate their impacts. Here, we presented a comprehensive analysis of spatial and temporal patterns of LFs in the French Mediterranean. To our knowledge, the fire database compiled and analyzed in this framework provides, for the first time, a detailed description of LFs recorded on georeferenced long time series.
In total, 21 % of the burned area occurred on a surface that already
burned in the past due to multiple overlaps in burned areas by recurrent
fires (up to six times in the east). These areas of higher recurrence could
induce a loss of resilience of the forest types such as
We found that the return level was higher in the eastern part of the study
area although LFs were more frequent in the west. These contrasted regional
return levels may provide critical and useful information for risk
assessment and local decision-making. Indeed, LF
Some recent studies across Euro-Mediterranean countries emphasized that large fire preferentially occurred under specific synoptic patterns associated with high temperature (Pereira et al., 2005; Trigo et al., 2016; Hernandez et al., 2015). In southern France, large fires were also facilitated by wind events blowing from northwest (Ruffault and Mouillot, 2015, 2017). The shapes of LFs which were more elongated in the wind direction in the western part support the results of Ruffault et al. (2018), pinpointing that the main wind-driven large fires that had occurred in 2016 were located in the western part while the main heat-driven large fires that occurred in 2003 were located in the east of the area. Taking into account other metrics describing the LF patch complexity (e.g., azimuthal angle or shape index) as in Laurent et al. (2018) could allow the derivation of additional information on the role of wind in their geometry or in the fraction of LFs driven by wind.
The overall reduction in both LF frequency and burned area observed over the last 6 decades is in agreement with previous works that highlighted a decrease in fire activity across parts of southern Europe in response to an increased effort in fire suppression (Turco et al., 2016), taking place in the early 1990s in the French Mediterranean (Ruffault and Mouillot, 2015; Fox et al., 2015; Curt and Fréjaville, 2018). Indeed, the region was highly impacted by fires during the 1970–1990 period and developed a thorough fire suppression and prevention system in the beginning of the 1990s, allocating more means for fire management that allowed faster reactivity in case of fire start (the strategy became extinguishing the fires at their initial stage by massive attack to prevent their spread). The decrease in both LF frequency and burned area since 1991, especially evident in the eastern part of the region, is likely due to this change in firefighting policy and fire prevention regulations (fire suppression could be more intense in the east as fires were historically larger in that region).
Climate projections suggest that atmospheric conditions conducive to large fire will increase in the future. Indeed, the warming and drying trends projected in southern Europe are expected to facilitate fire spread (Turco et al., 2018), at least where fuel and ignitions are not limiting. This trend towards more extreme fire weather conditions is likely to overcome prevention efforts in the French Mediterranean (Lahaye et al., 2018), a region where expanding forests (Abadie et al., 2018) are increasing fuel loading and may offer opportunities for future fire spread.
This work, based on long-term georeferenced fire time series (1958–2017), analyzed both spatial and temporal variations in LFs throughout one of the most impacted areas of the French Mediterranean. On the whole, 21 % of the total area burned by LFs occurred on a surface that had already burned in the past, the region being impacted in some locations up to six times by recurrent LFs (coastal areas of the eastern part of the study area). LFs were less frequent in the eastern part but larger than LFs occurring in the west. This longitudinal gradient in LF extent, featuring a shorter time of occurrence between LFs in the east with respect to the west, contrasts with what we would expect from mean fire weather conditions strongly decreasing eastwards but is consistent with larger fuel cover in the east. Indeed, fuel continuity in the east allows fire to grow large and to reach on average 4000 ha every 7 years, a spatial extent in burned area observed only every 50 years in the west.
An abrupt decline in LF was evident across the eastern part in the early 1990s, mostly due to a change in fire management policy thereby contributing to the weakening of the climate–fire relationship. However, despite large means allocated to fire suppression, large fire outbreak is still possible in the French Mediterranean (such as in 2003 or 2016), as extreme weather conditions can overwhelm fire suppression efforts (Fernandes et al., 2016; Lahaye et al., 2018). A better knowledge of LF drivers is necessary to strengthen fire prevention by providing valuable information on priority areas where LFs are more likely to occur.
The data are property of the State: Directions
Départementale des Territoires et de la Mer du Var
(2019,
The authors declare that they have no conflict of interest.
This article is part of the special issue “Spatial and temporal patterns of wildfires: models, theory, and reality”. It is a result of the conference EGU 2017, Vienna, Austria, 23–28 April 2017.
The authors wish to thank Adeline Bellet and Denis Morge for preprocessing the data with ArcGIS. The authors also sincerely thank Aimee Mac Cormack for English revision of a previous version of the paper. The authors also thank the two anonymous reviewers for their constructive comments and suggestions that helped improved the quality of the paper.
This paper was edited by Ricardo Trigo and reviewed by two anonymous referees.