1. Introduction
Wildfire refers to unplanned and uncontrolled fires that occur in natural environments such as forests, cultivated lands, and grasslands [
1]. Wildfire disasters pose significant risks to the ecological environment [
2,
3]. They not only incinerate vegetation, damaging the roots and underground parts of plants and leading to plant mortality and reduced vegetation, but also have a severe impact on biodiversity [
4]. Moreover, wildfires can burn organic matter in the soil, impairing soil fertility and structure and resulting in soil erosion, loss of topsoil, and water source contamination [
5,
6,
7]. Furthermore, wildfires can trigger natural disasters such as landslides and floods, exacerbating the destruction of the ecological environment [
8,
9]. The harm caused to the ecological environment by wildfire disasters is comprehensive and far-reaching, posing a significant threat to the stability and sustainability of ecosystems [
10].
China is one of the countries with the highest frequencies of forest fires, with a level of devastation surpassing the global average [
11,
12]. Guangxi Zhuang Autonomous Region, as the most serious area of wildfire disasters in the country, had 774 recorded forest fires in 2019 alone, and the affected area and number of casualties were among the highest in the country, posing a great threat to the ecological environment and the safety of human life and property. Nanning, being a high-risk region for wildfire disasters in Guangxi, has suffered severe losses to its ecological environment and social economy due to frequent wildfire occurrences in recent years. In November 2021, a significant wildfire disaster took place in Wutang Town, Nanning, scorching an estimated area of 500 acres, devastating the surrounding vegetation, and causing substantial economic losses. Additionally, in April 2022, a wildfire incident in Xiaolu Village, Nanning, posed a grave threat to the safety of nearby nature reserves, reservoirs, and villages. Hence, as a vital component of wildfire monitoring and prediction, it is of utmost importance to assess the risk posed by wildfire disasters in terms of their destructive impact and losses to the ecological environment [
13]. Additionally, government agencies and industries such as tourism and transportation can utilize wildfire risk assessment results to develop effective wildfire prevention plans, thereby mitigating the losses caused by wildfire disasters to the ecological environment [
14].
Wildfire risk refers to the probability of wildfire occurrence and spread (danger) and the potential damage to the environment and human society (vulnerability) [
15,
16,
17]. In terms of the ecological environment, wildfire risk describes the potential risk of wildfire damage to the ecosystem. Specifically, wildfire danger refers to the likelihood of wildfire occurrence, with areas of high danger being more prone to wildfires [
18,
19]. Ecological vulnerability characterizes the degree to which an ecosystem is susceptible to wildfire damage, with high vulnerability indicating that the ecosystem is more likely to be harmed and destroyed by wildfires [
20,
21].
Wildfire danger is an integral component of wildfire risk assessment. Its principle is based on historical wildfire data and involves analyzing the responses of wildfires to factors such as topography, vegetation, climate, and human activities to predict the probability of wildfire occurrence [
22,
23,
24,
25]. In recent years, advancements in computer science, geographic information systems, and remote sensing technologies have enabled the use of multiple data sources, such as remote sensing imagery, meteorological data, and geospatial information data, to build wildfire danger models [
26,
27,
28]. This development has laid the foundation for obtaining more accurate and larger-scale assessments of wildfire danger. In particular, the widespread adoption of machine learning (ML) and artificial intelligence algorithms has greatly improved the speed and accuracy of wildfire danger research [
29]. Algorithms such as logistic regression [
30], artificial neural networks [
31], support vector machines [
32], random forest [
33], and deep neural networks [
34] can effectively handle the complex nonlinear relationships between wildfire occurrence and various factors, leading to more precise predictions of wildfire danger [
35,
36]. Furthermore, a notable advancement in wildfire danger research is the amalgamation of ML algorithms with the SHapley Additive exPlanations (SHAP) method [
37,
38,
39]. In contrast to the single ML models that provide predictions without in-depth insights into feature importance, the ML-SHAP models can offer a unique advantage in comprehending the contributions of different factors to wildfire danger predictions. SHAP enables quantification of the impact of each input feature on the model’s output, thereby facilitating the interpretation and understanding of underlying factors influencing wildfire danger [
40,
41]. This transition to SHAP-based approaches enhances the transparency and interpretability of wildfire danger assessments, enabling stakeholders to make well-informed decisions [
39,
42].
Currently, wildfire danger research primarily focuses on annual-scale analysis and often overlooks seasonal differences in wildfires [
20]. However, the occurrence and behavior of wildfires can significantly vary across different seasons [
43,
44]. Merely considering average annual conditions may not fully reveal the true level of wildfire danger [
12]. In reality, climate conditions and vegetation statuses in different seasons have a substantial impact on wildfire danger [
45,
46]. Therefore, accounting for seasonal changes in factors such as vegetation and climate is crucial for a more comprehensive and accurate evaluation of wildfire danger. Integrating the SHAP method into seasonal wildfire danger analysis allows for the assessment of varying effects of driving factors of wildfires in different seasons. Through the integration of seasonal perspectives with the SHAP method, this strengthened spatiotemporal analysis enables a deeper understanding of the driving forces behind wildfires in different seasons, thereby promoting the development of precise and effective wildfire prevention and management strategies.
Additionally, more attention needs to be given to the potential risks that wildfire disasters pose to the ecological environment. Wildfires not only cause vegetation destruction and soil erosion, but also contribute to the loss of biodiversity, water source contamination, and climate change issues [
47,
48]. However, in many wildfire danger studies, the evaluation of the potential risks of wildfire damage to the ecological environment is often overlooked or limited to qualitative descriptions [
20,
49]. This situation restricts our in-depth understanding of the impacts of wildfire disasters and hinders the development of effective wildfire prevention, control, and ecological protection strategies. Therefore, to better understand the seasonal differences of wildfires and the potential risks of wildfire disasters to the ecological environment, it is necessary to consider the seasonal changes in factors such as vegetation and climate. This will strengthen the spatiotemporal analysis of wildfire danger research and allow for a more in-depth assessment of the potential impacts of wildfires on ecosystems during different seasons. Such research can provide more accurate and comprehensive wildfire risk assessments, which will serve as a more precise and effective scientific basis for wildfire prevention and management, as well as the protection of the ecological environment [
50].
In view of this, this study proposes a methodology and process for seasonal wildfire risk assessment that is integrated with the ecological environment. Firstly, historical wildfire samples in Nanning were taken as the research objects, taking into account the spatial distribution of wildfire samples to select the appropriate seasonal and nonseasonal factors. An ML algorithm was utilized to construct wildfire danger models for different seasons, and the SHAP method was utilized to assess the differences in the impact of wildfire driving factors during different periods. Secondly, ecological vulnerability models for each season were developed by integrating various ecological environmental factors. Finally, by integrating the danger and vulnerability information within the same period, wildfire disaster risk models for each season were constructed. The specific objectives were as follows: (1) Summarizing the seasonal spatial distribution patterns of wildfire-prone areas, exploring the predictive performance advantages of a seasonal-based wildfire danger assessment method, and analyzing the contribution differences of factors to wildfire occurrence in different seasons; (2) analyzing ecological vulnerability and the contribution of ecological variables to vulnerability during different seasons; (3) evaluating the wildfire risk in Nanning during different seasons and summarizing the spatial distribution patterns of different risk levels. In summary, this research advances the field of wildfire risk assessment by considering seasonal differences and ecological factors, contributing to a more comprehensive understanding of wildfire occurrence patterns and supporting informed decision-making for effective wildfire management and prevention strategies.
3. Methods
The methodology employed in this research consisted of three stages.
In the first stage, we aimed to construct wildfire danger models for different seasons in Nanning using an ML algorithm. We explored the prediction performance differences of wildfire danger based on annual and quarterly scales, analyzed the spatial distribution characteristics of wildfire danger in different seasons, and assessed the variation in the degree of contribution of each factor on a seasonal scale. The main procedures are described as follows. Initially, we established a spatial database to acquire and categorize wildfire samples from five distinct periods: all seasons, spring, summer, autumn, and winter. Additionally, we gathered wildfire driving factors corresponding to each of these periods. Subsequently, we conducted multicollinearity analysis on the factors for different periods to ensure their independence. Second, we randomly divided the modeling datasets for different periods into training and testing datasets according to a ratio of 7:3. Based on the training set, we used the light gradient boosting machine (LGBM) algorithm to construct wildfire danger models for different periods and generated wildfire danger maps for each season to assess their spatial distribution. Based on the test set, multiple evaluation metrics were used to evaluate the zonal rationality and predictive performance of each model. Third, we employed the SHapley Additive exPlanations (SHAP) method to assess the importance variation in wildfire driving factors across different seasons and focused on analyzing the differences in the influences of seasonal factors on wildfire danger prediction results. The overall technology roadmap is depicted in
Figure 6.
In the second stage, a vulnerability assessment of the ecological environment was conducted for different seasons, as shown in
Figure 7. Firstly, Landsat imagery data were collected and divided according to the seasons. The greenness, heat, wetness, and dryness indicators were calculated and obtained for each season, and then normalized. Subsequently, principal component analysis (PCA) was applied to integrate the four indicators in order to calculate the remote sensing ecological index (RSEI) for each season. Finally, based on the spatial distribution of the RSEI, the vulnerability of the ecological environment in Nanning was assessed for each season.
In the third stage, the wildfire risk assessment was conducted for different seasons in Nanning. Following the disaster risk assessment method of the United Nations Department of Humanitarian Affairs (UNDHA), the wildfire danger model for each season was integrated with the ecological vulnerability model to construct the wildfire risk model for the corresponding season. Additionally, the spatial distribution characteristics of wildfire risk were summarized for each season, providing a scientific basis for the prevention and management of wildfire disasters in Nanning.
3.1. Wildfire Danger Assessment Method Based on ML Algorithm
3.1.1. Light Gradient Boosting Machine (LGBM)
The light gradient boosting machine (LGBM) is a ML algorithm based on the gradient boosting decision tree (GBDT) [
77]. LGBM stands out due to its algorithm based on histograms and a leaf-wise growth strategy, which effectively reduces memory usage and computation time while handling sparse data. Moreover, LGBM incorporates various strategies, such as regularization, to prevent overfitting and supports multi-threading and parallel computing, allowing for efficient processing of massive data and high-dimensional features. As a result, LGBM offers faster processing speed, lower memory consumption, and improved accuracy, making it a valuable tool for wildfire danger assessment [
78,
79]. LGBM utilizes a gradient-boosting algorithm to enhance model accuracy, and the calculation formula is as follows:
where
represents the final predicted value,
denotes the learning rate,
represents the prediction value of the
-th decision tree, and
is the initial predicted value.
In addition to using wildfire samples as positive samples, ML models also require negative samples to construct the data foundation for the purpose of building the wildfire danger regression prediction model [
9]. In this study, equal numbers of negative samples were randomly selected in areas with low historical wildfire density, and the distance between samples was greater than 500 m to ensure the randomness of the negative sample region’s characteristics. Specifically, the numbers of negative samples for spring, summer, autumn, and winter were 2345, 561, 1908, and 4522, respectively. Wildfire-positive samples were assigned a value of “1”, while negative samples were assigned a value of “0”. In total, 70% of the samples were randomly selected as the training set, and the remaining 30% were used as the test set. The training set was used for model training and construction, while the test set was used to verify the predictive performance of the danger models for each period. In this study, we employed the grid search method for hyperparameter tuning to optimize the performance of our danger prediction model. K-fold cross-validation with a value of 10 was used to construct a more reliable and accurate model. The hyperparameter tuning results for the models in each period are shown in
Table 2.
3.1.2. Performance Assessment of Danger Models
This study selected multiple indicators in order to comprehensively evaluate the predictive performance of the wildfire danger models from different dimensions. The receiver operating characteristic (ROC) curve is a common method used to describe the classification performance of the model [
80]. The ROC curve plots the true positive rate (TPR) on the vertical axis and the false positive rate (FPR) on the horizontal axis, and the area under the curve (AUC) represents the model’s classification ability. The AUC value ranges from 0.5 to 1, where a value closer to 1 indicates a better classification performance of the model [
81]. Precision represents the proportion of correctly predicted wildfire samples among all predicted wildfire samples. It measures the accuracy of positive predictions [
82]. Recall measures the proportion of correctly predicted wildfire samples among all actual wildfire samples. It captures the ability of the model to identify positive samples [
37]. The F1 score combines both precision and recall into a single metric, allowing for a comprehensive evaluation of the model’s effectiveness in predicting wildfires [
83]. Overall accuracy (OA) refers to the proportion of correctly predicted samples to the total number of samples [
84]. The kappa coefficient (KC) is a statistical measure used to assess the consistency between the model’s predicted results and the true labels. A value closer to 1 indicates a higher level of agreement between the model’s predictions and the true labels [
34]. The root mean squared error (RMSE) is used to measure the average difference between the model’s predicted values and the true labels. A smaller RMSE indicates a smaller error between the model’s predicted results and the true values, indicating a better predictive capability of the model [
18]. The above indicators are calculated as follows:
In Equations (2)–(5), (true positive) indicates the number of wildfire samples that were correctly classified as wildfire. (true negative) indicates the number of non-wildfire samples that were correctly classified as non-wildfire. On the other hand, (false positive) is the number of non-wildfire samples that were incorrectly classified as wildfire. (false negative) is the number of wildfire samples that were incorrectly classified as non-wildfire. In Equation (6), represents the number of samples. and respectively, represent the predicted value and the actual value of the -th sample.
3.1.3. SHapley Additive exPlanations (SHAP) Method
SHapley Additive exPlanations (SHAP) is an interpretable method based on game theory which is used to explain the outputs of any ML model [
85]. Its essence lies in the Shapley value, which is a method of allocating contributions to total consumption based on players’ participation. The interpretability of Shapley values is manifested as an additive feature attribution method, where the model’s predictions are explained as the sum of attribution values for each input feature [
49,
86]. We employed the Tree-SHAP method, which considers tree-based models in conjunction with an input dataset X of size N × M, generating an N × M matrix with Shapley values (where N represents the number of samples and M denotes the number of features), enabling the comprehensive global explanation of wildfire danger models constructed at different time periods. The calculation formula for Shapley is as follows:
where
represents the
-th sample;
denotes one of the features;
represents a feature subset that does not include the feature
;
represents the model’s prediction output when the feature subset
is removed from the model;
represents the model’s prediction output when the feature
is added to the feature subset
; and
is the total number of samples in the dataset. In this study, we utilized the SHAP method to generate global summary plots and single dependence plots. These plots were used to analyze the direction and contribution of each factor in predicting wildfire danger during different periods. The dependence relationship between the attribute values of seasonal factors and the prediction results of danger was evaluated.
3.2. Ecological Vulnerability Assessment Method Based on RSEI
Disaster vulnerability can be defined as the degree of vulnerability to damage that an area or system exhibits in the face of a disaster [
87]. It reflects the resistance and recovery capacity of the subject under disaster conditions. Specifically in the context of the ecological environment, disaster vulnerability refers to the vulnerability of various components in the ecosystem (such as vegetation, soil, water resources, etc.) to disasters. It takes into account the responsiveness of the ecosystem’s structure, functions, and dynamic processes to disasters. In this study, we used RSEI to assess the vulnerability of wildfire disasters in different seasons by integrating greenness, heat, wetness, and dryness in the study area.
In order to mitigate the influence of varying scales and dimensions among different indicators on the evaluation results, normalization was applied to the indicators [
73]. The normalization process is represented by Equation (19).
where
represents the normalized value of each indicator,
represents the value of each indicator at the
-th pixel, and
and
represent the minimum and maximum values of each indicator, respectively.
Subsequently, the principal component analysis (PCA) method was employed to objectively allocate weights based on the contribution rates of NDVI, LST, WET, and NDBSI. This approach helps to reduce biases caused by subjective weighting based on personal experience [
88,
89]. The final outcome is a composite index, RSEI, which is used to assess the ecological environment status [
90]. The calculation formulas are presented as Equations (20) and (21).
where
denotes the initial remote sensing ecological index obtained through PCA;
is the minimum value of
;
is the maximum value of
; and
signifies the standardized remote sensing ecological index, ranging from 0.0 to 1.0. A higher RSEI value indicates a better ecological environment status, and it also suggests that the occurrence of wildfire may result in more severe ecological damage. Conversely, a lower RSEI value suggests that the ecological environment is relatively less susceptible to the impacts and damage caused by wildfires.
3.3. Wildfire Risk Assessment Method
The natural disaster risk arises from the interaction between disaster danger and vulnerability. It represents the potential risk of losses to specific entities resulting from the occurrence of the disaster [
91,
92,
93]. The assessment of disaster risk combines both these factors and quantifies the level of risk by evaluating potential impacts and losses [
94,
95]. The calculation formula for disaster risk is shown in Equation (22).
where
represents the wildfire danger value of the
-th pixel, and
represents the ecological vulnerability value of the
-th pixel. The resulting
represents the risk of wildfire-induced ecological damage for the
-th pixel, ranging from 0 to 1.
5. Discussion
5.1. SHAP Dependence Analysis of Nonseasonal Factors
The research findings revealed that among all nonseasonal factors, land use, slope, and distance to rivers played a significant role in predicting wildfire danger across different seasons in Nanning. Utilizing the SHAP single dependency analysis method, we gained detailed insights into the impact of these factors on wildfire danger within various attribute intervals (
Figure 17).
In terms of land use, we observed different trends in different seasons. In spring, the Shapley values for cultivated land, forest land, and grassland were generally greater than 0, indicating that these land use types are all susceptible to wildfire disaster. In contrast, the Shapley values of forest land and grassland remained high in summer, autumn, and winter, while cultivated land showed high Shapley values in only some samples. It is evident that forest land and grassland were more susceptible to wildfire impacts compared to cultivated land during these seasons.
In terms of slope, we discovered that in any season, the samples exhibited Shapley values greater than 0 when the slope exceeded 5° but remained below 30°. This suggests that steeper slopes contributed to rugged terrain, accelerating the spread of wildfires, and fuel accumulation. Consequently, areas with higher slopes were more prone to wildfire occurrences.
In terms of distance to rivers, we observed that when the distance was greater than 500 m but less than 2000 m, the samples exhibited positive Shapley values, indicating a higher danger of wildfires in these areas. It is evident that regions adjacent to rivers were less prone to fire incidents, while areas farther away from rivers but still maintaining a certain distance were more susceptible to wildfires. These areas likely harbored a greater abundance of flammable vegetation and combustible materials, thereby increasing the risk of wildfire spread and propagation.
On the whole, the Shapley value trends of slope and distance to rivers in different seasons are relatively consistent, indicating that their influence on wildfire danger remains relatively stable across seasons. However, for the same land use type, such as cultivated land, there are variations in the distribution of Shapley values among different seasons, suggesting that the impact of land use on wildfire danger varies across seasons. It is evident that seasonal factors play an important moderating role in explaining the influence of the other factors on wildfire danger in different seasons. This variability reminds us of the need to adopt appropriate prevention and management strategies in different seasons to address the potential risks of wildfires.
5.2. Comparison with Previous Studies
Wildfire danger assessment is a complex task involving multiple influencing factors and non-linear relationships, and traditional machine learning algorithms often struggle to provide explanations for model predictions [
96]. The SHAP method, by calculating the Shapley values of features, quantifies the contribution of each feature to the prediction results, thus aiding in the interpretation of the model’s prediction process and outcomes [
40]. In our study, we observed that slope, as an important terrain factor influencing wildfires, received extensive support in previous research [
38,
39,
49]. Additionally, many studies emphasized the positive role of vegetation coverage in promoting wildfire occurrences, which aligns with our findings [
38,
49]. However, there are also studies suggesting a weaker impact of vegetation coverage, and even a negative relationship with wildfire occurrences [
39]. Moreover, meteorological factors, such as temperature, have shown a significant positive impact on wildfire occurrences in some studies [
37,
49], while others have demonstrated an inverse result, with temperature having a minor negative effect on wildfires [
39,
97]. These discrepancies in wildfire danger assessment arise from differences in climate and vegetation coverage across regions, leading to variations in the assessment results. Furthermore, within the same region, the frequency of wildfire occurrences may change over time due to temporal variations in climate and vegetation coverage. Therefore, exploring the temporal and spatial patterns of wildfire danger based on seasons and understanding the complexity of influencing factors are of vital importance.
As of now, numerous scholars have conducted extensive research on wildfire danger within different seasons [
12,
20,
50]. Their findings consistently show distinct spatial distributions of wildfire danger across different seasons in the same region, demonstrating that considering seasonal dimensions provides a more comprehensive identification of potentially high-danger areas. However, variations exist in the assessment results of seasonal wildfire danger among different regions. In our study, the distribution of high-danger wildfire areas was more widespread in spring and winter, with fewer areas in summer and autumn, which is in line with a previous study [
12]. However, another study showed higher danger in spring and autumn [
50]. Furthermore, some regions even exhibit a divide between winter wildfire states in the north and summer wildfire states in the south [
20]. These observations indicate significant regional differences in wildfire danger influenced by seasonal changes. Different climate conditions, vegetation coverage, and terrain factors in distinct regions may result in varying patterns of wildfire danger across seasons. Therefore, to understand the wildfire occurrence patterns in different regions, it is essential to fully consider seasonal factors and to explore the reasons behind the seasonal differences, thereby enhancing the accuracy and effectiveness of wildfire prediction and response.
5.3. Significance of Different Dimensions of Wildfire Riskiness Assessment
Given the seasonal distribution patterns of wildfire incidents in Nanning and the influence of seasonal factors, conducting wildfire risk assessments for different seasons is of great significance. Assessing wildfire risks in different seasons can help us to gain a comprehensive understanding of fire risk conditions. Climate conditions and vegetation states vary across seasons, necessitating considerations of these factors’ fluctuations in wildfire risk assessments. In the study area, the frequency of wildfire incidents is higher in spring and winter, with a wider coverage of high-risk areas. Summer and autumn exhibit higher ecological conditions and vulnerability, making them more susceptible to wildfire impacts. Thus, notable differences exist in the danger and vulnerability assessments among different seasons.
In conclusion, due to the suddenness, destructiveness, and uncontrollability of wildfire incidents, as well as the influence of seasonal differences, conducting wildfire risk assessments for different seasons is necessary. By integrating assessments of wildfire danger and ecological vulnerability across different seasons, we can better understand the potential risks of wildfire-induced ecological losses in different seasons [
20]. This understanding enables us to take the corresponding preventive measures, strengthen monitoring and emergency responses, reduce the impact of wildfire incidents on the ecological environment and society, and ensure the safety and quality of people’s lives.
5.4. Limitations and Future Works
This study has several limitations. Firstly, only four common seasonal factors, namely, NDVI, temperature, precipitation, and wind speed, were considered in this work. However, there are other factors that could be included in the wildfire danger assessment, such as solar radiation [
98,
99], soil moisture [
12,
100], and air humidity [
44,
101]. Secondly, the historical wildfire data provided by government agencies mainly focus on forest areas, and there is incomplete information about wildfire incidents in cultivated land and transitional vegetation areas. Therefore, this study advocates for the use of active fire data products [
37]. Although the wildfire sample dataset has undergone rigorous quality control and screening, it may still need to be compared with imagery from sources such as Landsat 8, Sentinel, and other imagery data to further enhance the reliability of the wildfire samples.
Furthermore, it should be noted that the lightning activity factor was not included in our investigation of seasonal wildfire danger, which was a limitation of this study. Lightning activity plays a crucial role as a major natural ignition source in the formation and propagation of wildfires [
102]. The process of lightning-induced fires is highly complex and closely related to factors such as vegetation type, fuel conditions, weather, and topography, with many aspects still remaining unknown [
103]. Lightning-caused fires in specific regions often exhibit distinct spatiotemporal distribution patterns [
104], and factors such as meteorology [
105,
106] (especially rainfall), combustible materials [
107] (vegetation cover, vegetation type), and topography [
108] significantly influence the frequency and intensity of wildfires caused by lightning activity [
109]. In this study, based on consideration of seasonal factors such as topography, climate, and vegetation cover, we explored the spatiotemporal variation patterns of wildfire danger. Among the seasonal influencing factors, climate factors such as temperature, wind speed, and precipitation directly affect the probability of wildfire occurrence, while vegetation cover determines the fuel supply for wildfires [
102]. The combined effects of these seasonal factors lead to considerable variations in wildfire danger across different seasons and regions. Therefore, considering the critical role of lightning activity in wildfire occurrence, future research could focus on studying the significant impact of lightning activity on wildfires and incorporate it into group of the factors for assessing wildfire danger. By collecting and compiling wildfire disaster data caused by lightning activity and integrating lightning activity factors with key factors such as topography, climate, and vegetation cover, we may enhance the predictive accuracy of models and gain a better understanding of the spatial distribution and seasonal variation of wildfires. Moreover, exploring the interaction between lightning activity and climate factors can further elucidate the relationship between wildfire danger and climate change. Through in-depth research on the impact of climate factors and lightning activity on wildfire danger, we will come to better understand the potential influence of climate change on wildfire risk, providing scientific evidence for wildfire disaster management and mitigation.