Protected Areas Show Substantial and Increasing Risk of Wildfire Globally

Abstract

Protected area coverage is set to expand in response to climate change and the biodiversity crisis, but we lack assessments of wildfire incidence in protected areas. Here, we quantify biogeographical variation in global patterns of burned area in protected areas. During the twenty-first century, wildfires have burned 2 billion hectares of protected areas—an area the size of Russia and India combined—and, while protected areas only cover 19.2% of semi-natural ecosystems, they concentrate 28.5% of the area burned annually. Wildfire in protected areas increased significantly between 2001 and 2024 (+0.46% yr⁻¹), even after taking into account increases in protected area (+0.27% yr⁻¹), pointing to a disproportional impact of fire on protected areas under increasingly severe fire weather. This pattern showed marked variation across biomes, with the largest disproportionate increases occurring in fire-prone biomes (e.g., Mediterranean and dry tropical forests, tropical grasslands, and xeric shrublands). There were important exceptions to this general trend, and protected area fire was lower than expected in biomes where fire activity is naturally limited by moisture (e.g., tropical rainforests or montane grasslands). Wildfires are important for the health of many ecosystems, and such values of burned area will not always mean a negative outcome. Amidst concerted efforts to expand protected area coverage, such as the Global Biodiversity Framework, our results highlight the need for new management strategies that address the globally increasing impacts of burned area across protected areas under unabated climate change.

1. Introduction

Anthropogenic activities are catalyzing a global polycrisis where multiple stressors interact synergistically and aggravate environmental impacts [1]. Climate change, for instance, is exacerbating the biodiversity crisis by degrading habitats and driving large-scale species adaptation, migration, or—in the worst case scenario—extinction [2]. A less frequently considered feedback between biodiversity and climate change lies in the surge in wildfire activity experienced in recent years, where the frequency and severity of extreme fires has increased (despite declines in overall area burned globally) [3]. Protected areas (PAs) are considered a cornerstone for biodiversity conservation [4], and ensuring effective nature protection under increasing wildfire incidence is a key challenge for the 21st century.

There is substantial uncertainty as to whether protected areas show a proportionally higher or lower fire activity relative to unprotected areas [5]. This is because assessments of burned area within PAs have so far been performed locally or regionally, and the results vary widely across regions [6,7]. We still lack a robust global quantification of how fire impacts Pas, as, under the Kunming-Montreal Global Biodiversity Framework (GBF), PA coverage should increase up to 30% of the land by 2030. Quantifying whether protected area fire is changing will be crucial to the successful implementation of the GBF and other conservation efforts.

Here we quantify, for the first time, variation in the biogeographic patterns of burned area within PAs across biomes and how they have changed between 2001 and 2024. Also, given that PAs’ coverage has increased steadily during the 21st century, we addressed the critical question of whether the incidence of protected area fire has been increasing proportionally at the same pace as the expansion of PAs, or whether the increase in PAs has been accompanied by disproportionate increases (overproportion) or decreases in burned area within PAs. The goal of this short note was not to address whether the establishment of forest reserves has led to changes in the fire regime but to quantify fire incidence in protected areas globally.

2. Materials and Methods

2.1. Global Maps of Protected Areas

To quantify the incidence of wildfire inside PAs, we focused our analysis on natural and semi-natural “burnable” vegetation (forests, shrublands, grasslands, and savannas), and we overlaid annual burned area from the Moderate Resolution Imaging Spectroradiometer (MODIS) Burned Area product (MCD64A1) collection 6.1, along with maps of PA coverage collated from the World Database of Protected Areas [8] and other sources, and the definition of biomes from Dinerstein et al. [9].

We used the WDPA database (January 2025 version) [8] as the main data source on global protected areas. We calculated PA coverage following the guidelines provided by the UN Environment Programme World Conservation Monitoring Centre (UNEP-WCMC). We only considered terrestrial PAs and the terrestrial part of marine protected areas. Following current practice [8,10,11], PAs designated as “Proposed” and “Not reported” were excluded as their implementation is not yet finalized. UNESCO Man and Biosphere (MAB) Reserves were similarly excluded following [8] because MAB core areas are already accounted for in the WDPA.

When PA establishment dates were missing, we followed previously published approaches [10,11,12] by randomly selecting a year (with replacement) from all PAs within the same country with a known establishment date. For countries with fewer than five protected areas with known establishment dates, we randomly selected a year from all terrestrial protected areas with a known date. The random assignment was repeated 1000 times to identify the median year of establishment, which was then assigned to each protected area with unknown establishment dates.

Some countries apply data restrictions to the WDPA, and data could not be accessed for China, India, Turkey, Eritrea, Western Sahara, and the regions of Jammu and Kashmir and Azad Kashmir. Regarding China, we used additional data from the China Nature Reserve Specimen Resource Sharing Platform http://www.papc.cn/html/folder/946895-1.htm (accessed on 1 June 2024). The total coverage was 1,721,008 km² (~18% of total land area), which coincides with PA coverage according to other studies [13]. Data on protected areas for India was obtained from the Wildlife Institute of India https://wii.gov.in (accessed on 1 June 2024), and it was adapted to the WDPA structure. We searched the spatial information of Indian PAs with Open Street Map using an OverPass query “https://overpass-turbo.eu”, (accessed on 1 June 2024). We obtained a total of 691 PAs, covering an area of 149,443 km², also in line with other publications. We acknowledge that heterogeneity in data sources is undesirable but also unavoidable.

2.2. Burned Area and Land Cover Maps

Regarding spatial burned area estimates, we used the MODIS Burned Area product (MCD64A1) Collection 6.1 in geographic coordinates format. MODIS BA provides global information on burned area in a 500 m grid and monthly time steps [14]. We created monthly binary layers by setting a value of 1 to burned pixels and aggregating them at yearly steps. Aggregated rasters were then multiplied for a cell-size raster layer calculated in spherical coordinates. This corrected for pixels burning more than once in a year.

We used the MODIS land cover (LC) type product (MCD12Q1) collection 6.1 [15] to filter burned areas and protected areas to specific LC types, following the IGBP scheme. More specifically, we only studied fire in semi-natural vegetation, which includes forests (evergreen needleleaf forests, evergreen broadleaf forests, deciduous needleleaf forests, and deciduous broadleaf forests), shrublands (open and closed), savannas (woody savannas and savannas), and grasslands. This mask was necessary to avoid artifacts related to a higher presence of non-flammable crops outside protected areas, which could artificially increase the impact of fires within protected areas. This mask was applied considering the conditions of the year before a given fire. That is, the analysis of a fire occurring in 2020 would be based on the value of the mask released at the end of 2019.

We used MODIS LC layers as a template grid to rasterize vector data, as in previous publications [16]. Country and continent boundary layers were converted to binary rasters at MODIS resolution (~500 m) after applying the rule of the 50% cell cover. PA polygons were similarly dissolved by year. Each pixel was then associated with a specific continent/region and characterized as either protected or unprotected following previous publications [16], and the LC mask was then applied. MODIS resolution was then aggregated at a 0.25° × 0.25° grid for spatial analyses at a global scale.

2.3. Statistical Analyses

Our primary response variables were the cover of protected areas across biomes (in absolute terms and also in percent) and protected area fire (the percent of burned area within each protected area biome, %PAF). The absolute cover of PAs was estimated by adding the area of all PAs within a biome after applying the land use mask. The percent cover of PAs (%PA) within a biome was calculated from the ratio between the absolute cover of PAs and the total area (protected and unprotected) of that biome (after applying the land mask):

$$\%PA= {\displaystyle \frac{{PA}_C}{{PA}_C+ {noPA}_C}}100$$

(1)

where PA_C and noPA_C represent the cover of protected and unprotected areas, respectively. The %PAF was estimated from the fraction of the total burned area that occurred within PAs (after applying the land cover mask):

$$\%PAF= {\displaystyle \frac{{BA}_{PA}}{{BA}_{PA}+ {BA}_{noPA}}}100$$

(2)

where BA_PA and BA_noPA indicate the sum of burned area across protected and unprotected areas, respectively.

Temporal trends across the response variables during the study period were estimated using the Mann–Kendall (MK) test and the Sen’s slope, which are common techniques in time series analyses [17]. The former is a non-parametric test, based on ranks, that is used to detect the presence of a monotonic tendency in a time series [18,19]. The latter quantifies the overall slope based on the median of all slopes calculated between each pair of points in the time series.

In order to quantify whether PAs were burning out of proportion, we compared %PA with %PAF. If protected areas burn preferentially, then we should observe that %PAF is significantly larger than %PA. For example, if protected areas occupy 10% of the forest area within a biome (%PA = 10%), but 50% of the burned area occurs in protected areas (%PAF = 50%), then fires would be disproportionately affecting protected areas (%PAF > %PA). If there is no effect of protection on the burned area, we expect that %PA would not be significantly different from %PAF (%PAF = %PA). Finally, if protected areas are less affected by fire than the rest of the landscape, then protected area fire should be smaller than the percent of land occupied by protected areas (%PAF < %PA). We assessed for statistical differences between annual values of %PAF and %PA using the MK test and Sen’s slope with D (the difference between slopes). The first test informs on whether both slopes are significantly different, and the second one quantifies the increment, or decrement, of the difference D (if MK’s p-value < 0.05).

All analyses were performed using the R software version 4.4.2, using the base packages as well as terra for raster data manipulation, sf [20] and lwgeom [21] for vector data manipulation and geometries correction, rmapshaper [22] for polygon simplification, doParallel [23] and foreach [24] for parallel computing, and the trend [25] package for Mann–Kendall test and Sen’s slope calculations.

3. Results

By the end of 2024, PAs covered 1.7 billion hectares of semi-natural vegetation, but during the 21st century, fire has so far burned 2 billion hectares of PAs (Figure 1a–d). That is, wildfires during 2001–2024 burned the equivalent of 117% of the area occupied by PAs by 2024. This does not mean that wildfires have affected the entirety of PAs, as some areas burned more than once (wildfire impacts have concentrated over 23.7% of global PA coverage, while 76.3% remains unburned). In absolute terms, the PAs most affected by fire are those in tropical and subtropical grasslands (primarily from Africa), where 1.6 billion hectares of PAs have burned, although PAs only occupy 324 million hectares (Mha) in that biome, followed by flooded grasslands, where 134.3 Mha of PAs have burned (although PAs only occupy 28.4 Mha) (Figure 1c,d). Within forest biomes, most of the protected area fire concentrates in tropical moist broadleaf forests (88.4 Mha, equivalent to 19.3% of the land covered by PAs) and in tropical dry broadleaf forests (34.2 Mha, equivalent to 82% of the PAs). The PAs least affected by fire in absolute terms were those occurring in mangroves (1.5 Mha, or 28.9% of the total area covered by PAs) and tundra (1.9 Mha, or 1.4% of the total area covered by PAs).

The absolute magnitude of burned area across the different biomes, as reported above, largely reflects differences in the size and fire regimes of the different biomes. The vast majority of burned area globally occurs in tropical grasslands and savannas (especially in Africa), where fires naturally show very short rotation periods (1–5 years) [26]. It is thus to be expected that an overwhelming proportion of the global burned area within PAs will also be concentrated in that biome. The short rotation period in tropical savanna fires [27] additionally explains why the area burned by fire is 5-fold larger than that occupied by PAs in that biome (these naturally fire-prone ecosystems were expected to burn several times over our 24-year study period).

In order to better quantify how wildfires are differentially impacting PAs across biomes, we examined the proportion of burned area occurring within protected areas (% protected area fire, or %PAF hereafter) (Figure 1e,f) and compared it against the percent of semi-natural vegetation that is protected (Figure 1g,h). We observed that %PAF globally increased from 18.0% in 2001 to 28.5% in 2024 (Sen’s slope annual trend size = +0.46% yr⁻¹, Mann–Kendall test, p < 0.0001) (Figure 2a). The increase was due partly, but not completely, to the increase in PA coverage over the same time period (from 12.6% in 2001 to 19.2% in 2024; Sen’s slope annual trend size = +0.27% yr⁻¹, Mann–Kendall test, p < 0.0001). This result is pointing towards an overproportion in the incidence of wildfires within PAs globally, which is substantiated by the fact that %PAF increases at a significantly faster pace than the cover of PAs (Sen’s D (difference across slopes) = +0.19% yr⁻¹, Mann–Kendall test, p < 0.0001).

This overproportion was consistent across PAs in forest and non-forest biomes overall (Figure 2a,b), but it was not universal, and major differences arose across biomes (Figure 2c–p). Among forest biomes, the %PAF averaged over the study period was significantly higher than the cover of PAs in Mediterranean (+12.5%, Wilcoxon test p > 0.0001) and tropical dry (+7.4%, Wilcoxon test p > 0.0001) forests, and it was marginally higher in boreal (+2.0%, Wilcoxon test p = 0.09) and temperate (+1.8%, Wilcoxon test p = 0.08) conifer forests. In contrast, in tropical moist broadleaf and conifer forests, %PAF was significantly lower (−8%, Wilcoxon test p < 0.0001; and −1.8%, Wilcoxon test p = 0.02, respectively) than the cover of PAs, indicating that PAs in these tropical biomes suffer a disproportionately lower impact of wildfires than that expected from their coverage. Finally, there were no differences between %PA cover and %PAF in temperate broadleaf forests. The disproportionate impact of wildfires over protected areas is thus mostly limited to drier and hotter forests and only marginally apparent in non-tropical conifer forests.

In non-forest biomes, we similarly observed diverging impacts of wildfire in PAs. An overproportion of protected area fire has occurred in tropical (+8.7%, Wilcoxon test p < 0.0001) and flooded grasslands (+1.2%, Wilcoxon test p = 0.031), and also in desert and xeric shrublands (+7.8%, Wilcoxon test p < 0.0001). Conversely, PAs burned at significantly lower proportions than expected in montane grasslands (−12.7%; Wilcoxon test, p < 0.0001) and mangroves (−4.7%, Wilcoxon test p < 0.0001). In the remaining biomes (temperate grasslands and tundra), %PAF was consistent with the cover of PAs. The disproportionate effect of %PAF within non-forest ecosystems seems to be limited to those biomes that experience seasonal droughts.

4. Discussion

Here we quantified global variation in protected area fire and observed distinct relationships across biomes. The observation that an overproportion of protected area fire tends to occur in seasonally dry, fire-prone biomes (e.g., tropical grasslands or Mediterranean forests), while the opposite pattern (decreases in protected area fire) is more prevalent where fires are putatively rare (e.g., tropical moist broadleaf forests [27]), deserves further attention. This pattern could arise from two different processes: the first is that the establishment of protected areas impacts the fire regime. Flammability traits of native vegetation have been positively correlated with habitat fire-proneness [28] and, in turn, the management of PAs tends to foster the role of “natural” processes. Our data therefore supports the hypothesis, based on ecological principles, that increasing “naturalness” in PAs leads to higher wildfire activity in fire-prone biomes (where fire activity is naturally high) and to lower fire activity where fires are naturally scarce. However, our results do not necessarily indicate that burned area increases after protection: it is also possible that areas that are now protected already had higher fire activity before protection was established.

Our results could additionally reflect the impacts of the exclusion of Indigenous and cultural burns from “wilderness” areas [29,30]. The goal of this study was to assess and quantify the biogeographical pattern of fire activity on PAs, as no study had yet attempted this globally and previous regional studies revealed differing results. Future research efforts should thus address these hypotheses to better understand the processes underlying the patterns we document here, given their importance for a successful GBF implementation. Regardless of the mechanism, our results do highlight the need to consider fire management in PAs.

The overproportion of fire in PAs from fire-prone biomes should not be necessarily considered a negative outcome, as wildfires play a key role in the health of those biomes [31]. However, extremely large wildfires have become more common over the last few years [32]. These extreme events may hinder recovery if increases in fire intensity, frequency, severity, or area burned push the fire regime beyond the species realized niches [5,33].

Beyond ecological impacts, wildfires in PAs can also affect infrastructure and people, and, in those cases, land management for human safety (including ignition control and fuel management) should take priority. If the current disproportion between %PAF and PA cover continues (e.g., 28.5% PAF vs. 19.2% PA cover in 2024), we can expect that 44.5% of the area burned globally will be concentrated within PAs if PA coverage reaches GBF’s 30% target in semi-natural ecosystems. We thus encourage that PA management include active fire prevention plans [34]. Under an increasing climate emergency, we can expect further increases in the disproportionate incidence of wildfires in PAs, potentially compromising the goals of the Global Biodiversity Framework and other conservation efforts in some of the world’s most ecologically valuable ecosystems.