Exploring Drivers of Wildfires in Spain

: Wildfires play a dual role in ecosystems by providing ecological benefits while posing catastrophic events; they also inflict non-catastrophic damage and yield long-term effects on biodiversity, soil quality, and air quality, among other factors, including public health. This study analysed the key determinants of wildland fires in Spain using openly available spatial data from 2008 to 2021, including fire perimeters, bioclimatic variables, topography, and socioeconomic datasets, at a resolution of 1 km 2 . Our methodology combined principal component analysis (PCA), linear regression analysis, and one-way analysis of variance (ANOVA). Our findings show that scrub/herbaceous vegetation (average 63 ± 1.45% SE) and forests (average 19 ± 0.76% SE) have been highly susceptible to wildfires. The population density exhibited a robust positive correlation with wildfire frequency (R 2 = 0.88, p < 0.0001). Although the study provides insights into some fire-related climatic drivers over Spain, it includes only temperature-and precipitation-based variables and does not explicitly consider fuel dynamics. Therefore, a more advanced methodology should be applied in the future to understand the local specifics of regional wildfire dynamics. Our study identified that scrub/herbaceous areas and forests near densely populated regions should be prioritised for wildfire management in Spain, particularly under changing climate conditions.


Introduction
Wildland fires are considered a natural phenomenon worldwide [1] and are one of the major disasters that have altered the environmental conditions of the southern part of Europe [2].Wildland fires provide many ecological functions, including maintaining ecosystem health, helping forest regeneration, and controlling insect outbreaks and disease damage [3].However, despite their ecological benefits, these fires can lead to catastrophic events that result in the loss of biodiversity and forest resources.Moreover, they pose significant socioeconomic challenges by causing infrastructure destruction and endangering human life and health, particularly in densely populated regions [4,5].In Euro-Mediterranean (EU-MED) countries such as France, Greece, Italy, Portugal, and Spain, over 95% of fire incidents were caused by anthropogenic activities (such as purposefully, arsonistically, or deliberately), and the remaining 5% are naturally caused by lightning and other unknown factors [6,7].However, over 60,000 fires occur in the EU-MED region each year, consuming more than 500,000 hectares of land [8].This indicates the enormous number of land cover distributions lost to wildfires, altering their overall ecosystem functioning.
On a European basis, EU-MED countries have the highest frequency of fires and the greatest number of burned areas because of extreme wildfire episodes [9][10][11], whereas Spain had the second-highest number of wildfire episodes after Portugal.Moreover, the impact of climate change has altered the climatic conditions of the Mediterranean Basin, with temperature increases in summers, a reduction in precipitation in springs and autumns, and wetter and milder climates in winters [12][13][14][15].As mentioned earlier, these circumstances exacerbate high heatwave episodes, which in turn generate wildland fire ignitions, resulting in severe fire activity and fatalities [16].Nevertheless, studies have discovered that several factors, such as meteorological variables, topographical characteristics, and factors related to socioeconomic conditions, have an important role in increasing the frequency and extent of wildland fires during wildfire seasons [17][18][19].Hence, the annual burned areas in Mediterranean countries of Europe have seen an increase over the past two decades, attributed to intensified climate change coupled with other environmental factors [20].
Importantly, understanding the distribution of land use/cover (LUC) in the Mediterranean region is critical because it could provide fire experts with a clear picture of how wildfires might affect communities based on climate and human-induced factors influencing fire dynamics [21,22].Over the decades, wildland fires have been changing the states of LUC types in the Mediterranean basin [6].Therefore, to gain insights into the trends of wildfire occurrences, we analysed CORINE land-cover (CLC) maps of 2006, 2012, and 2018.These maps provide valuable information about the historical, present, and potential future patterns of wildfires.Hence, accurate data can assist in understanding the current situation and distribution of various land-cover types [21,22].However, it is crucial to note that not all land cover types are equally flammable, as [22] discovered that forests and agricultural lands burn more than shrublands and grasslands due to the accumulation of fuel loads, sources of fire ignition, and climatic circumstances.At the country level, Spain has witnessed significant LUC changes as a result of human activity in recent years [23].
Machine learning techniques, such as principal component analysis (PCA), have become crucial for understanding the underlying factors of wildfires [24,25].In the realm of wildfire research, these approaches are gradually gaining recognition and relevance.Numerous studies have been conducted using PCA to determine the underlying factors influencing fires [25,26].Indeed, among the many algorithms available, PCA is becoming a powerful tool for predicting the driving factors of wildland fires, the number of fires, and burned areas at all levels [25,[27][28][29][30][31].Additionally, some authors have demonstrated that the PCA model produces better results as compared to other approaches [29,30,32,33].Consequently, PCA models have been widely employed in identifying wildfire drivers at global and continental scales; however, there has been minimal research at the national and provincial levels.Concisely, the motivation behind this research is to contribute to the body of knowledge of leveraging artificial intelligence methods such as PCA to build appropriate fire propagation measures and management strategies that would help mitigate the danger of wildland fires in Mediterranean-type regions.
This paper presents a temporal, spatial, and statistical analysis of factors driving wildfires across various LUC types and regions in Spain.The study explores the use of PCA combined with linear regression and analysis of variance (ANOVA) to identify the primary factors influencing wildfires within the area of study.The research sets three objectives: (i) to analyse the spatial and temporal patterns of fire incidents and the extent of burned areas, (ii) to determine which LUC types (agricultural areas, forests, others, scrub/herbaceous vegetation and wetlands) are most susceptible to wildfires, and (iii) to pinpoint the main factors driving wildfires at both national and provincial levels.The Materials and Methods section provides detailed descriptions and information on the datasets used.Consequently, this research aids in understanding the factors that drive wildland fires, thereby enabling predictions of their behaviour and spread patterns for specific periods and locations.This study offers insights that can inform the development of effective fire suppression strategies, the establishment of emergency warning systems, and the strategic allocation of resources to combat and suppress future wildfires.

Study Area
The study areas comprise all autonomous communities of Spain with the exclusion of the Canary Islands territory (UTM projection of ETRS89, latitude of 40.4637 • N, and longitude of 3.7492 • W).Therefore, the remaining total land area of the study site was approximately 498,497 km 2 .Figure 1 depicts the study area and land cover distribution map of Spain and the percentage of each land cover type across the country.The climate of the Iberian Peninsula of Spain varies considerably.It is characterised by moderate winter cycles, extremely drier and unbearably hot in the summer period, and wetter and cooler in the mountain areas.Annually, the average temperature ranges from 0 • C to 18 • C, annual precipitation ranges from 150 mm to 2500 mm, and the average wind speed is 7.9 miles per hour.As of the end of 2020, the demographic population density of Spain is estimated to be 66,229,715 people, with Andalucia having the highest population density of approximately 11,796,105 people residing in the region.Within study periods, our study discovered 3430 fire events, and these events damaged more than 867,500 hectares caused by wildfires between 2008 to 2021.Notably, 266 wildfires were reported in 2012, resulting in the loss of 178,504 hectares of land.In addition, when considering land cover types, about 257 fires occurred in 2017, destroying around 112,892 hectares of scrub/herbaceous vegetation.

Data Sources 2.2.1. Bioclimatic, Socioeconomic, and Topographic Variables
Several studies have identified temperature and precipitation as critical factors affecting moisture availability and fire propagation [25,34], with fuel moisture found to be crucial to assessing fire danger [35] due to the positive response of fire activity, such as fire frequency and spread, to increasing forest fuel dryness [36,37].Therefore, this study incorporates 19 bioclimatic variables (bio-1 to bio-19) from the WorldClim database [38,39], with 11 variables describing temperature-related metrics and 8 variables related to precipitation.From our observations, the line graph shows that July and August experience the highest peak temperatures, averaging 23 • C. Additionally, Shuttle Radar Topography Mission (SRTM V.2.1) [40] elevation data at ~30 s (1 km 2 ) spatial resolution of the Universal Transverse Mercator (UTM) resolution grid was downloaded from WorldClim [38] and processed in ArcGIS version 10.7 to obtain topographic features such as elevation and distance from the streams in the study area.Table 1 shows the explanatory codes of the 19 bioclimatic variables, including elevation and streams.In terms of topographic variables, hydrological analysis was performed on the elevation data to obtain the stream order and distances of the fire points around the streams across the study area.The elevation data were filled to raise the cells with shallow pixel values and prevent broken streamlines.Subsequently, a flow-direction raster was created using the filled elevation data, and areas with higher accumulations (areas greater than 1% of the highest accumulation) corresponding to stream channels were obtained using the flow-direction raster.Euclidean distances around the streamlines were obtained, and pixel values were extracted into the fire database to obtain the distance between each fire point and the nearest stream channel.
Regarding socioeconomic variables, the population density data were acquired from the Socioeconomic Data and Application Centre database (SEDAC-https://sedac.ciesin.columbia.edu/,accessed on 19 June 2023), a data centre in NASA's Earth Observing System Data and Information System (EOSDIS).The data provide a gridded population density raster at a 1 km 2 resolution.The population ranges from 2010 to 2020, with a 5-year gap.The data consists of the world's population density derived from the country's census counts and adjusted to match the total population of the United Nations country totals [41].The raster files were masked to Spain, and each cell value corresponding to the population density of each year was extracted into a point feature of the burned areas to obtain the corresponding population density.Compared to raster data, vector data, such as point features, allow further analysis and comparison of population densities across the years and use the figures to understand the relationships between wildland fires and population.Each population year within the fire period was selected as the corresponding period.For example, the 2010 population was used for all fires between 2008 and 2011.This allowed for an objective representation of the population density across wildland fire occurrences over the study period.In addition, life expectancy, human development index (HDI), gross national income (GNI), and income index were obtained from Global Data Lab (https://globaldatalab.org/, accessed on 19 June 2023).These variables are within the study period spanning from 2008 to 2021.

Fire Dataset Acquisition and Processing
The spatial wildfire data used for this analysis on the European scale were obtained from the European Forest Fire Information System (EFFIS database-https://effis.jrc.ec.europa.eu/,accessed on 19 June 2023) as provided by [42].The collection and documentation of forest fires across European countries date back to the 1970s.Still, the lack of harmonisation of these datasets has limited a collaborative approach to regional wildfire prevention [43].As a result, the EFFIS was developed as a collaborative effort by the European Commission services (Directorate General Environment and the Joint Research Centre) and the relevant fire services across the countries (forest fires and civil protection services) to provide data needed by agencies such as the European Parliament and the European Commission Services [2].The fire dataset in the EFFIS database is a comprehensive analysis that covers the complete forest management cycle, including fire prevention and post-fire damage analysis, and receives comprehensive forest fire information from countries across European and Mediterranean regions [6].The study period for fire record information is 14 years (2008-2021), which comprises burned area per hectare (ha), sources of ignition, start and end dates, size of the fire occurrence, and location of each wildfire.The UTM resolution was readjusted to 1 km 2 resolution to be consistent with the other datasets used in this project.The limitation in the use of this dataset is the exclusion of small-size fires as the data do not capture fires with burned areas below 30 Ha.However, while only a fraction of the total number of fires is mapped in the EFFIS dataset, the burned areas of size larger than 30 ha represent 75% to 80% of the total EU burned areas (http://effis.jrc.ec.europa.eu/about-effis/data-license/,accessed on 19 June 2023), thus, making it suitable to assess country-level fire behaviour.

Land Cover Classification
The land cover maps for 2006, 2012, and 2018 were obtained from the Copernicus website (https://land.copernicus.eu/pan-european/corine-land-cover,accessed on 19 June 2023).The CORINE Land Cover (CLC) database is a European Programme coordinated by the European Environment Agency (EEA) that provides consistent information on land cover and land cover changes across Europe.Initially, the original data were masked to Spain and reclassified from 15 classes at level 2 and 44 land-cover types at levels 3 to 7 using map algebra with incorporated Python function code in ArcGIS version 10.7.
Furthermore, individual classes were dissolved to obtain five reclassified land cover types.The polygon was rasterised to extract cell values corresponding to the land-cover type in the fire database.To this end, we reclassified land use and land cover types into five distinct classes: agricultural areas, forests, scrub/herbaceous vegetation, wetland, and others (see Table 2 for complete details and information).

Methods
We developed a methodology to examine the temporal and spatial patterns of wildfires in Spain, focusing on the frequency of fires and the extent of burned areas.A flowchart of the stepwise methodology is presented in detail in the diagram below (Figure 2).Initially, the process involved consolidating all datasets into a unified database to define criteria for fire size, enabling the evaluation of burned areas and fire counts.We conducted spatial and statistical analyses on historical bioclimatic variables, wildfire perimeters from 2008 to 2021, CORINE land cover data for 2006, 2012, and 2018 at six-year intervals, population density figures for 2010, 2015, and 2020 at five-year intervals, and digital elevation model (DEM) data.Linear regression and one-way ANOVA were employed for all datasets to investigate variable correlations.Then, PCA was applied in the second phase to establish connections between dependent and independent variables.Data processing, analysis, and visualisation were performed using ArcGIS version 10.7 and R software version 4.3.1.

Statistical Analysis
For statistical analysis, we used PCA to determine the association between the dependent variable (fire perimeter data) and the independent variables (bioclimatic variables, CORINE land cover, elevation, distance to stream, life expectancy, HDI, GNI, income index, and gridded population data).In this study, PCA was employed to gain insights into and identify the drivers of wildland fires in Spain.Hence, PCA is a useful method that allows the interpretation of variance within the original data to effectively lower the dimensionality of the datasets.Additionally, we used a linear regression model to make predictions among the predictors and explanatory variables to test for statistical relationships.These statistical analyses and PCA tests were performed using R software version 4.3.1 [44] using the following library packages: ggrepel, ggplot2, factoextra, FactoMineR, janitor, lubridate, magrittr, pcaMethods, raster, and tidyverse.

Trends in Number of Fires and Burned Areas
The number of fires and the size of burned areas in Spain due to wildfires varied considerably across the study period, both within different land-cover types and regions.Generally, the total number of fires increased significantly within the study period from 32 fire events in 2008 to 850 in 2021 (Figure 3 left, Appendix A: Table A2).Specifically, scrub/herbaceous vegetation accounts for the highest number of fires across the various land-cover types with a mean of 153 ± 37 (1 standard error of the mean), as shown in Appendix A: Table A2).Forests have the second highest number of wildfires, with a mean of 47 ± 12, while agriculture stands at a mean of 31 ± 8 fires (Appendix A: Table A2).The highest number of wildfires in a single year per land-cover class occurred in scrub/herbaceous vegetation, with a total of 528, accounting for ~62% of the total wildfires in 2021 (Appendix A: Table A2).Relatedly, scrub/herbaceous vegetation has the highest burned areas through the study period with a mean of 36,552 ± 7239 ha, with forest and agriculture areas having mean values of 11,397 ± 2381 ha and 9456 ± 2608 ha, respectively (Appendix A: Table A3), having second and third highest burned areas within the study period.The highest total burned area was in 2012, with a total area of 178,504 ha (Appendix A: Table A3, Figure 3 right).The total burned areas significantly differ on the scrub/herbaceous land compared to other land cover classes, with other land cover classes showing interrelatedness in the extent of yearly burned areas.However, when the proportion is considered, scrubland significantly differs from all the land cover classes, while the proportion of forest burned is not different from agricultural lands but differs significantly from other land cover classes, as revealed by the corresponding letter in the boxplots.Similar to the highest number of fires per year per land-cover type, the highest burned area per year also occurred in the scrub/herbaceous vegetation with a total of 101,727 ha, making ~57% of total burned areas in 2012 (Appendix A: Table A3).Expectedly, others (land cover class as described in Table 2 and wetland land category are the land cover types with the lowest burned areas (see Appendix A: Table A3, for complete details and information) and number of fires (Appendix A: Table A2), respectively.Additionally, during the study period (2008-2021), the year 2017 had the second-highest total burned areas of 129,434 ha.Notably, both 2012 and 2017 correspond to the period of El Nino events, which resulted in the increase in temperatures across Spain (Figure 3 right).Overall, ~63% of total wildfires in Spain occurred in scrub/herbaceous vegetation; forest areas are responsible for ~19%, while agricultural areas accounted for ~13% of total wildfires in Spain (Appendix A: Table A2).
The number of fires per month shows two distinct periods of high wildfires, specifically during spring (February to April) and summer (July to October), as shown in Figure 4, left.This depicts that major fire events are not limited to the summer period when there is less rainfall and intense heat but in the late winter and early spring when the temperature is relatively low.This can also be attributed to the increasing winter temperature, which favours fire events across different land-cover classes in Spain.Furthermore, while March has the highest number of wildfire events within the study period, with a total of 648, the distribution across the study period is positively skewed, as shown in Figure 5, with more fires, on average, occurring during the summer periods.The monthly burned areas show the concentration of large fires towards the summer periods (from July to October).Unlike the number of fires that show two distinct periods, the monthly burned areas' distribution shows a higher number of burned areas in the summer, which peaks in August (Figure 5, right).Therefore, while the summer and spring seasons drive the number of wildfires, the size of fires is more favoured during the summer.
The number of fires and the extent of burned areas were not evenly distributed across all regions in Spain (Figure 6).Hence, some areas experienced a higher number of fire events than others.Moreover, weather patterns across different regions of Spain also vary significantly, which can further complicate the management and prevention of fires.Interestingly, autonomous communities such as Principado de Asturias, Galicia, and Castilla y León have the highest number of fire events, with 754, 732, and 576 fires, respectively (Appendix A: Table A4, Figures 7 and 8).However, provinces like Galicia, Castilla y León, and Andalucía have the highest burned areas with a total of 170,370 ha, 168,845 ha, and 115,384 ha, respectively (Figure 9).The total burned areas per region did not reveal the complete information about burned areas, as some autonomous regions are considerably more extensive in size compared to others.The larger burned areas may be due to the size of the region; therefore, the percentage of region burned areas across the study period (2008-2021) was calculated using Equation (1).The result shows Cantabria, Principado de Asturias, and Galicia as having the highest burned areas relative to the region's size (Figure 9).This is important, especially when compared to the total burned area; Cantabria has a total burned area of 34,450, less than Galicia, Castilla y León, and Andalucía (regions with the highest total burned area), but higher in proportion.This is due to the size, which is considerably smaller.While we observed differences in burned proportion across Spain regions, further studies could explore the difference in burned proportion based on the available wildland and compare the results with ours to understand variations.Similarly, regions like Castilla y León and Andalucía have a higher total burned areas; however, the proportion, relative to the size, is lower (Figure 9; Table A4).While further exploring the relationship between burned areas and the number of fires in each region, we found a moderate relationship, which shows that the extent of burned areas is not necessarily dependent on the number of fire events within a particular region.For example, 51,595 ha of land cover was burned in Aragon, corresponding to 49 fire events, while Cantabria, with a total of 450 fires, had just 34,450 ha of burned areas.Additionally, regression analysis between the number of fires and burned areas per region (Figure 10 shows a significant relationship, with the number of fires explaining 74% of variations in burned areas with an R 2 of 0.74 and a p-value of 0.001.However, when the number of fires is examined while factoring the land cover types, it explains 77% of the variation in yearly burned areas within the regions with an R 2 of 0.77 and p-value < 0.001 (Figure 11) and 85% variation in monthly burned areas with an R 2 of 0.85 and p-value < 0.001 (Figure 12).We calculated the percentage of burned areas in each municipality per year, as shown in Table A4 and Figure 9.
where %BA = percentage of burned area in each region; BA ry = Burned areas in each region per year; A r = Area of a region in hectares.

Wildfire Variability
Andalucía, Cataluña, and Comunidad de Madrid are the regions with the highest population, with a mean population of 11,884,485, 11,365,403, and 9,834,728, respectively.While this does not directly relate to wildfires, population within points of fire occurrences will better describe the influence on wildfires.Therefore, the total population within 1 km 2 grids (spatial resolution of the gridded population data) of fire points was used to analyse the relationship between the number of fires, burned areas, and population.A strong association was found between the total population and the number of fires, showing population as a significant driver of wildfires in Spain.Examining the relationship between population and the total number of fires per year through linear regression, an R 2 of 0.88 and a p-value of <0.001 show a significant positive relationship with population, explaining 88% variations in yearly wildfires in Spain (Figure 13), showing the importance of increasing population to fire frequency.Based on our findings across the different land cover classes, a high relationship was observed between burned areas and the number of fires across each period.The number of fires was found to significantly influence burned areas across each land cover type, with a p-value < 0.001 (Appendix A: Table A7).Furthermore, two years, corresponding to 2012 and 2017, witnessed a significant increase in burned areas during the period of study, with p-values of 0.001 and 0.03, respectively (Appendix A: Table A7 highlighted).This is due to the significant increase in temperature and drought events within these two years.This reveals the importance of temperature change in driving the extent of burned areas across Spain, with increasing temperature significantly increasing the burned areas.
On a monthly basis, there was an observed variability in the extent of burned areas and the number of fires (refer to Figure 5).The relationship between both, across different land cover classes, shows that the monthly fire ignition across different land cover classes significantly influenced burned areas, with a p-value < 0.001 (Appendix A: Table A8).While a strong relationship was observed between monthly number of fires and burned areas across different land cover classes with an R 2 of 0.74, June, July, and August are months with larger burned areas, with p-values of 0.03, 0.01, and 0.009, respectively (Appendix A: Table A8 highlighted).This further reinforces the importance of the summer periods, characterised by rising temperatures and decreased rainfall, in influencing the spread of fires post-ignition.

Factors That Influence Wildfires
The biplot of PCA results showed different driving factors in wildland fires in various regions of Spain.Generally, the first three principal components account for approximately 80% of the total variability in the wildfire dataset (Appendix A: Table A7).Our results identified key drivers of wildfires based on land cover (Figure 14a), burned areas (Figure 14b), and number of fires (Figure 14c).To begin, on the one hand, the land cover distribution shows a difference in fire drivers.Fires in wetlands are driven by temperature-related variables (annual mean temperature, mean diurnal range, max temperature of warmest month, mean temperature of wettest quarter, mean temperature of driest quarter, mean temperature of warmest quarter, and mean temperature of coldest quarter).On the other hand, wildfires in scrub/herbaceous vegetation were mainly driven by elevation and temperature (i.e., temperature seasonality and temperature annual range).Population density, elevation, and stream variables influence fires in all the land cover types except wetlands.In agricultural areas, temperature-related variables are primarily the main drivers of fires.Specifically, the analysis shows a temperature-driven increasing burned areas, with elevation contributing to the rising burned areas as seen in Figure 14a.Small areas respond to all bioclimatic variables, with growing burned areas favoured by temperature variables (temperature seasonality, temperature annual range, minimum temperature of coldest month, and mean temperature of coldest quarter) and elevation (Figure 14b).To this end, the number of fires shows differences in drivers for increasing reoccurrence of wildfire events.To obtain the number of wildfire recurrences, the fire dataset was spatially joined to know the count of overlapping fire polygons, thus examining the number of times a polygon was burned within the study period.From the analysis, small recurrent fires (1-3 times) are influenced by all variables (Figure 14c).However, Precipitation-related variables were observed to influence the increasing recurrence of fires within the study period.

Discussion
A combination of climatic, socioeconomic, and topographic variables has been significantly influencing the occurrence of fires and the extent of burned areas in Spain [23,25,32,[45][46][47] and other Mediterranean countries such as Greece [48], the south-eastern part of France [30], Italy [31], and Portugal [29].Regions in Spain show a large variability in the wildfire patterns and the extent of burned areas across different land cover types.Our study focused on the analysis of wildfire variability across different land cover types in Spain within the study period (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021), where autonomous communities of Castilla y León, Galicia, and Andalucía experienced the most burned areas, with 17,0162 ha, 168,845 ha, and 115,379 ha, respectively.Based on the analysis of the land cover dynamics in relation to wildfire activity, we found that scrub/herbaceous vegetation (encompassing grassland, sclerophyllous vegetation, and transitional woodland-shrubs after reclassifying land cover types) and forests are susceptible to most wildfires in Spain.This is in line with the study on wildfires in the European Mediterranean region [6] showing that forests experienced a more frequent occurrence of fires than other LUC types.This finding also closely aligns with a recent study conducted by [49] in Italy, which suggests that shrubland and forests, particularly coniferous forests, are prone to burning within the Mediterranean biome.The increased risk is attributed to such factors as species composition, fuel accumulations, and the impact of drier and hotter climate conditions during the fire seasons.Furthermore, our study demonstrated that agricultural-related land cover types, i.e., agricultural landscapes, were not affected by large fires across the regions in Spain.This is evident when comparing the landcover types of regions with the highest burned areas, which reveals an inverse relationship between burned areas and the percentage of agricultural areas across three regions (Galicia, Castilla y León, and Andalucía).This result supports the study [45], which asserted that agricultural-related fires might be becoming less significant in Spain.
Population density is identified as one of the key socioeconomic factors impacting wildfires in the framework of our study, which is in agreement with previous studies conducted in Spain [50].We found a positive correlation between yearly dynamics of population and the number of fires, as well as monthly population and burned areas, confirming the role of human factors as drivers of fire dynamics [17,50].
In our study, we used Principal Component Analysis (PCA).Recently, PCA has been applied to investigate the diverse drivers of wildfires, encompassing climatic, socioeconomic, and topographical factors on different eco-biome types around the world [25,29,30,49].Many studies have compared the patterns of association between wildfires and environmental [17,51] or human-caused [23,52] drivers of wildfires in the EU-MED region.For example, Montoya et al. [25] used PCA to determine the drivers of wildfires in different eco-biome types in Mexico, and results showed that grasslands, hydrophilic vegetation, and temperate forests had the highest burned areas, accounting for 42% of the total land cover areas.PCA was used to identify specific fire regime zones in south-eastern France and found that wildfire drivers in the research area are based on fuel loads, population density, topography, and inadequate fire-fighting capacity [30].Similarly, Jiménez-Ruano et al. [28] identified regions of fire activity in Spain using PCA and Ward's hierarchy clustering models based on the spatial-temporal framework of their key fire characteristics.PCA was applied in Portugal [29] to identify the most vulnerable locations for wildfires, revealing that wildfires are more likely to occur in the central and southernmost areas of Portugal in the future.However, while many machine learning algorithm methods like PCA coupled with different mathematical models have been commonly employed to identify wildfire drivers or predict burned areas using climate, socioeconomic, and topographic factors (some of which are listed in this study), few studies have considered these factors combined [53,54].
In our modelling work, we considered the temperature-and precipitation-related bioclimatic drivers of fire events.The summer months in Spain over the last decades have witnessed a major increase in extreme heat waves and drought conditions [55,56].Our analysis showed that fires are driven by population, elevation, distance to streams, and extreme precipitation-related variables such as the precipitation of the wettest month, the precipitation of the driest month, the precipitation of the wettest quarter, and the precipitation of the coldest quarter.More specific analysis is needed to understand the impacts of precipitation since fuel dynamics was not explicitly analysed in this study.On the one hand, decreasing precipitation in combination with other weather variables could lead to lower fuel moisture.On the other hand, the increasing precipitation favours forest growth and biomass accumulation through increased photosynthesis [57], leading to a higher fuel load in the forest landscapes.Therefore, we need to track precipitation dynamics in high temporal frequency.Fuel availability and dry conditions make the forest susceptible to fire ignition during wildfire seasons, especially in the presence of a large population density, which has been found to relate positively to the number of fires.Temperature-related variables, especially extreme events such as temperature seasonality, temperature annual range, minimum temperature of the month, mean temperature of the coldest quarter, and maximum temperature of the warmest month, were found significant in explaining the extent of burned areas across different land covers.It is evident that Spain experienced a notable temperature increase during the summer months (between June and September), with an average temperature of 23 • C (Appendix A: Figure A1).
In order to fully estimate the impacts of weather dynamics on wildfire, a more sophisticated methodology should be applied, including an assessment of evapotranspiration rates [58] based on a number of climatic variables, e.g., temperature, precipitation, wind speed, and relative humidity [59].Although our study used simplified methods, based only on the temperature-and precipitation-based bioclimatic indicators, it still captures climate impacts on the persistently high summer burned areas in Spain over the study period.Obviously, weather drives fire-favourable conditions being combined with human ignitions, which result in extensive burned areas during fire season [60].The years 2012 and 2017 recorded the highest burned areas.The extreme drought conditions of 2012 [61] and the intense summer heat of 2017 in Southern Europe [62,63] resulted in a significantly higher burned area in these two periods as compared to the average burned areas over the study period.Furthermore, the yearly number of fires showed an abrupt increase in 2019.Prior to 2019, the number of fires and burned areas (except the years with extreme temperature and drought events) has reduced, and this can be attributed to efficiency in emergency response [64] and successful risk prevention in some regions within Spain [65].However, several studies [2,53] have predicted an increase in wildfire activities resulting from climate change and an increase in fuel load due to socio-economic activities.To explain the sudden increase in the number of fires and burned areas within Spain in recent years, more factors should be analysed, including human activities and fuel dynamics.
In future work, our study would use a more advanced methodology, such as a processed-based wildfire model [66,67], to take into account interactions of several firerelated variables.Moreover, deep analysis of local dynamics using expert knowledge across Spain should be performed, while our study proposed a large-scale simplified framework based on openly available datasets for weather, land cover, and population density.

Conclusions
Wildfire activity in Spain is driven by both biophysical factors and human activities.Notably, the frequency and intensity of these wildfires vary depending on the type of land cover in different regions of the country.Our findings show that scrub/herbaceous vegetation (average 63 ± 1.45% SE) and forests (average 19 ± 0.76% SE) have been highly susceptible to wildfires.The population density exhibited a robust positive correlation with fire frequency (R 2 = 0.88, p < 0.0001).Our study identified that scrub/herbaceous areas and forests near densely populated regions should be prioritised for wildfire management in Spain, particularly under changing climate conditions.
1.In the abstract section, please change "only temperature and precipitation-based" to "only temperature-and precipitation-based". 2. In Section 2.2.1, the last paragraph should read: "Table 1 shows the explanatory codes of the 19 bioclimatic, topographic, and socio-economic variables."3. On the Table 1 header, please add a comma after "bioclimatic," so it should read: "bioclimatic, topographic, and socio-economic variables."4. In section 2.2.3, please add wetland as part of the land cover types "agricultural areas, forests, scrub/herbaceous vegetation, wetland, and others." 5. Kindly add this Figure A1:

Figure 1 .
Figure 1.Study area.The percentage of each landcover type per municipality is indicated in the pie chart.

Figure 3 .
Figure 3. Change in number of fires (for fires with burned area ≥ 30 Ha) per year by land cover (left) and burned areas (right) per year by land cover classes in Spain through the study period (2008-2021).

Figure 4 .
Figure 4. Boxplot of distribution of total burned areas in Spain by land cover types (left) and proportion of land cover burned (right) through the study period (2008-2021).The land cover types whose means are statistically different at α = 0.05 are in different letters, while colour transparency denotes a difference in median values.

Figure 5 .
Figure 5.The seasonality of fire activity of the total number of fires per month (left) from January to December (the line showing two distinct periods of an increasing number of wildfires) and the total burned area per month (right).For each month, we have the distribution of a historical number of fires over 2008-2021 (14 periods per month).The data describe fires with burned area ≥ 30 ha.

Figure 6 .
Figure 6.Distribution of monthly fire frequencies (top) and burned areas (bottom) in Spain (red dots represent observations while outliers are in blue).For each month, we have a distribution of a historical number of fires and burned areas over 2008-2021.The burned areas whose means are statistically different at α = 0.05 are in different letters.

Figure 7 .
Figure 7. Temporal sequence of the total number of fires (fires with burned area ≥ 30 ha) per region during the study period from 2008 to 2021, with each colour denoting the specific fire count for each region.

Figure 8 .
Figure 8. Spatial pattern of the number of fires per region between 2008 and 2021 in the study area.

Figure 9 .
Figure 9. Spatial distribution of total burned area (in percentage) per region between 2008 and 2021 in the study area.

Figure 10 .
Figure 10.Regression relationship between the number of fires and burned areas per region (point shows the total burned areas within each region) and regression formulae and scale has been log transformed to address heteroskedasticity.The blue line shows the regression line while the shaded grey area represents the standard error.

Figure 11 .
Figure 11.Regression relationship between the yearly burned areas and the number of fires factoring different land cover types (points and colours show the land cover types).The blue line shows the regression line while the shaded grey area represents the standard error.

Figure 12 .
Figure 12.Regression relationship between the monthly burned areas and the number of fires factoring different land cover types (points and colours show the land cover types; the blue line shows the regression line while the shaded grey area represents the standard error).

Figure 13 .
Figure 13.Regression relationship between total population and number of fires per year (point shows the total number of fires in each year).The number of fires refers to fires with a burned area ≥30 ha.The blue line shows the regression line while the shaded grey area represents the standard error.

Figure 14 .
Figure 14.Bi plot of PCA result.(a) Land cover type, (b) the mean of burned areas in hectares, (c) the number of fires for the study area for the period of 14 years (2008-2021).

Figure A1 :
Figure A1: Seasonal trend in average monthly temperature.Changes during the summer period correspond to increasing burned areas in the summer period (source: World Bank Climate Knowledge Bank).

Figure A1 .
Figure A1.Seasonal trend in average monthly temperature.Changes during the summer period correspond to increasing burned areas in the summer period (source: World Bank Climate Knowledge Bank).

Table 1 .
Explanatory code for bioclimatic, topographic, and socio-economic variables.

Table 2 .
CORINE and reclassified land-cover types.

Table A2 .
Number of fires across different land-cover types in Spain between 2008 and 2021.Land cover types whose means are statistically different at (α = 0.05) are in different letters.

Table A3 .
Burned areas (Ha) across different land-cover types in Spain between 2008 and 2021.Land cover types whose means are statistically different at (α = 0.05) are in different letters.

Table A4 .
Burned area per region in hectares.

Table A5 .
Yearly number of fires per region.

Table A6 .
Relationship between burned areas and number of fires across different land cover, factoring the different years (highlighted years have a significant relationship between burned areas and number of fires at (α = 0.05)).The strength of statistical significance at α = 0.05 is represented by the * symbol where * represent weak significance and *** represent strong significance.

Table A7 .
Relationship between burned areas and number of fires across the different land cover, factoring the different months (highlighted months have a significant relationship between burned areas and the number of fires at (α = 0.05)).The strength of statistical significance at α = 0.05 is shown by the * symbol where * represent weak significance and *** represent strong significance.

Table A8 .
Principal component proportion of variables.