Modeling the Spatial Distribution of Wildfire Risk in Chile Under Current and Future Climate Scenarios

Bastarrika A., Rodriguez-Montellano A., Roteta E., Hantson S., Franquesa M., Torre L., Gonzalez-Ibarzabal J., Artano K., Martinez-Blanco P., Mesanza A., Anaya J.A., Chuvieco E.

Iban M.C., Aksu O.

Wildfire susceptibility maps play a crucial role in preemptively identifying regions at risk of future fires and informing decisions related to wildfire management, thereby aiding in mitigating the risks and potential damage posed by wildfires. This study employs eXplainable Artificial Intelligence (XAI) techniques, particularly SHapley Additive exPlanations (SHAP), to map wildfire susceptibility in Izmir Province, Türkiye. Incorporating fifteen conditioning factors spanning topography, climate, anthropogenic influences, and vegetation characteristics, machine learning (ML) models (Random Forest, XGBoost, LightGBM) were used to predict wildfire-prone areas using freely available active fire pixel data (MODIS Active Fire Collection 6 MCD14ML product). The evaluation of the trained ML models showed that the Random Forest (RF) model outperformed XGBoost and LightGBM, achieving the highest test accuracy (95.6%). All of the classifiers demonstrated a strong predictive performance, but RF excelled in sensitivity, specificity, precision, and F-1 score, making it the preferred model for generating a wildfire susceptibility map and conducting a SHAP analysis. Unlike prevailing approaches focusing solely on global feature importance, this study fills a critical gap by employing a SHAP summary and dependence plots to comprehensively assess each factor’s contribution, enhancing the explainability and reliability of the results. The analysis reveals clear associations between factors such as wind speed, temperature, NDVI, slope, and distance to villages with increased fire susceptibility, while rainfall and distance to streams exhibit nuanced effects. The spatial distribution of the wildfire susceptibility classes highlights critical areas, particularly in flat and coastal regions near settlements and agricultural lands, emphasizing the need for enhanced awareness and preventive measures. These insights inform targeted fire management strategies, highlighting the importance of tailored interventions like firebreaks and vegetation management. However, challenges remain, including ensuring the selected factors’ adequacy across diverse regions, addressing potential biases from resampling spatially varied data, and refining the model for broader applicability.

Alkan Akinci H., Akinci H., Zeybek M.

Antalya is one of the provinces with the highest number of forest fires in Türkiye. In 2021, 278 forest fires occurred within the administrative boundaries of Antalya Regional Directorate of Forestry. The main objective of this study is to produce forest fire susceptibility (FFS) maps of Antalya province using machine learning (ML) models. In addition to forest fire inventory data, 16 factors, including topographic, environmental, meteorological, and human-driven, were used in the study. Inventory data included 2166 fire ignition points from the General Directorate of Forestry. 70% of the inventory dataset was used to train the ML models and 30% to validate the models. Overall accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC) approaches were considered as validation metrics. FFS maps of Antalya were produced using stand-alone ML algorithms, K-Nearest Neighbors, and Support Vector Machines, as well as tree-based Conditional Inference Trees (CTREE), Random Forest (RF), Gradient Boosting Machines (GBM), and Extreme Gradient Boosting (XGBoost) algorithms. To the best of our knowledge, this is the first study using the CTREE algorithm for forest fire susceptibility mapping. Therefore, this study is important for the related literature. The validation results revealed that the XGBoost model outperformed other models. It is thought that the FFS map produced using the XGBoost model will guide forest engineers, wildland firefighting teams, and firefighters to minimize damage and control forest fires.

Ochoa C., Bar-Massada A., Chuvieco E.

Analysing wildfire initiation patterns and identifying their primary drivers is essential for the development of more efficient fire prevention strategies. However, such analyses have traditionally been conducted at local or national scales, hindering cross-border comparisons and the formulation of broad-scale policy initiatives. In this study, we present an analysis of the spatial variability of wildfire initiations across Europe, focusing specifically on moderate to large fires (> 100 ha), and examining the influence of both human and climatic factors on initiation areas. We estimated drivers of fire initiation using machine learning algorithms, specifically Random Forest (RF), covering the majority of the European territory (referred to as the “ET scale”). The models were trained using data on fire initiations extracted from a satellite burned area product, comprising fires occurring from 2001 to 2019. We developed six RF models: three considering all fires larger than 100 ha, and three focused solely on the largest events (> 1000 ha). Models were developed using climatic and human predictors separately, as well as both types of predictors mixed together. We found that both climatic and mixed models demonstrated moderate predictive capacity, with AUC values ranging from 79 % to 81 %; while models based only on human variables have had poor predictive capacity (AUC of 60 %). Feature importance analysis, using Shapley Additive Explanations (SHAP), allowed us to assess the primary drivers of wildfire initiations across the European Territory. Aridity and evapotranspiration had the strongest effect on fire initiation. Among human variables, population density and aging had considerable effects on fire initiation, the former with a strong effect in mixed models estimating large fires, while the latter had a more important role in the prediction of very large fires. Distance to roads and forest-agriculture interfaces were also relevant in some initiation models. A better understanding of drivers of main fire events should help designing European forest fire management strategies, particularly in the light of growing importance of climate change, as it would affect both fire severity and areas at risk. Factors of fire initiation should also be part of a comprehensive approach for fire risk assessment, reduction and adaption, contributing to more effective wildfire management and mitigation across the continent.

Cordero R.R., Feron S., Damiani A., Carrasco J., Karas C., Wang C., Kraamwinkel C.T., Beaulieu A.

AbstractA string of fierce fires broke out in Chile in the austral summer 2023, just six years after the record-breaking 2017 fire season. Favored by extreme weather conditions, fire activity has dramatically risen in recent years in this Andean country. A total of 1.7 million ha. burned during the last decade, tripling figures of the prior decade. Six of the seven most destructive fire seasons on record occurred since 2014. Here, we analyze the progression during the last two decades of the weather conditions associated with increased fire risk in Central Chile (30°–39° S). Fire weather conditions (including high temperatures, low humidity, dryness, and strong winds) increase the potential for wildfires, once ignited, to rapidly spread. We show that the concurrence of El Niño and climate-fueled droughts and heatwaves boost the local fire risk and have decisively contributed to the intense fire activity recently seen in Central Chile. Our results also suggest that the tropical eastern Pacific Ocean variability modulates the seasonal fire weather in the country, driving in turn the interannual fire activity. The signature of the warm anomalies in the Niño 1 + 2 region (0°–10° S, 90° W–80° W) is apparent on the burned area records seen in Central Chile in 2017 and 2023.

Ghaffarian S., Taghikhah F.R., Maier H.R.

Disasters can have devastating impacts on communities and economies, underscoring the urgent need for effective strategic disaster risk management (DRM). Although Artificial Intelligence (AI) holds the potential to enhance DRM through improved decision-making processes, its inherent complexity and "black box" nature have led to a growing demand for Explainable AI (XAI) techniques. These techniques facilitate the interpretation and understanding of decisions made by AI models, promoting transparency and trust. However, the current state of XAI applications in DRM, their achievements, and the challenges they face remain underexplored. In this systematic literature review, we delve into the burgeoning domain of XAI-DRM, extracting 195 publications from the Scopus and ISI Web of Knowledge databases, and selecting 68 for detailed analysis based on predefined exclusion criteria. Our study addresses pertinent research questions, identifies various hazard and disaster types, risk components, and AI and XAI methods, uncovers the inherent challenges and limitations of these approaches, and provides synthesized insights to enhance their explainability and effectiveness in disaster decision-making. Notably, we observed a significant increase in the use of XAI techniques for DRM in 2022 and 2023, emphasizing the growing need for transparency and interpretability. Through a rigorous methodology, we offer key research directions that can serve as a guide for future studies. Our recommendations highlight the importance of multi-hazard risk analysis, the integration of XAI in early warning systems and digital twins, and the incorporation of causal inference methods to enhance DRM strategy planning and effectiveness. This study serves as a beacon for researchers and practitioners alike, illuminating the intricate interplay between XAI and DRM, and revealing the profound potential of AI solutions in revolutionizing disaster risk management.

Shmuel A., Heifetz E.

Accurate predictions of daily wildfire growth rates are crucial, as extreme wildfires have become increasingly frequent in recent years. The factors which determine wildfire growth rates are complex and depend on numerous meteorological factors, topography, and fuel loads. In this paper, we have built upon previous studies that have mapped daily burned areas at the individual fire level around the globe. We applied several Machine Learning (ML) algorithms including XGBoost, Random Forest, and Multilayer Perceptron to predict daily fire growth rate based on meteorological factors, topography, and fuel loads. Our best model on the entire dataset obtained a 1.15 km2 MAE. The ML model obtained a 90% accuracy when predicting whether a fire’s growth rate will increase or decrease the following day, compared to 61% using a logistic regression. We discuss the central factors that determine wildfire growth rate. To the best of our knowledge, this study is the first to perform such analyses on a global dataset.

Alkhatib R., Sahwan W., Alkhatieb A., Schütt B.

Due to the harm forest fires cause to the environment and the economy as they occur more frequently around the world, early fire prediction and detection are necessary. To anticipate and discover forest fires, several technologies and techniques were put forth. To forecast the likelihood of forest fires and evaluate the risk of forest fire-induced damage, artificial intelligence techniques are a crucial enabling technology. In current times, there has been a lot of interest in machine learning techniques. The machine learning methods that are used to identify and forecast forest fires are reviewed in this article. Selecting the best forecasting model is a constant gamble because each ML algorithm has advantages and disadvantages. Our main goal is to discover the research gaps and recent studies that use machine learning techniques to study forest fires. By choosing the best ML techniques based on particular forest characteristics, the current research results boost prediction power.

Turco M., Abatzoglou J.T., Herrera S., Zhuang Y., Jerez S., Lucas D.D., AghaKouchak A., Cvijanovic I.

Record-breaking summer forest fires have become a regular occurrence in California. Observations indicate a fivefold increase in summer burned area (BA) in forests in northern and central California during 1996 to 2021 relative to 1971 to 1995. While the higher temperature and increased dryness have been suggested to be the leading causes of increased BA, the extent to which BA changes are due to natural variability or anthropogenic climate change remains unresolved. Here, we develop a climate-driven model of summer BA evolution in California and combine it with natural-only and historical climate simulations to assess the importance of anthropogenic climate change on increased BA. Our results indicate that nearly all the observed increase in BA is due to anthropogenic climate change as historical model simulations accounting for anthropogenic forcing yield 172% (range 84 to 310%) more area burned than simulations with natural forcing only. We detect the signal of combined historical forcing on the observed BA emerging in 2001 with no detectable influence of the natural forcing alone. In addition, even when considering fuel limitations from fire-fuel feedbacks, a 3 to 52% increase in BA relative to the last decades is expected in the next decades (2031 to 2050), highlighting the need for proactive adaptations.

Yue W., Ren C., Liang Y., Liang J., Lin X., Yin A., Wei Z.

The frequent occurrence and spread of wildfires pose a serious threat to the ecological environment and urban development. Therefore, assessing regional wildfire susceptibility is crucial for the early prevention of wildfires and formulation of disaster management decisions. However, current research on wildfire susceptibility primarily focuses on improving the accuracy of models, while lacking in-depth study of the causes and mechanisms of wildfires, as well as the impact and losses they cause to the ecological environment and urban development. This situation not only increases the uncertainty of model predictions but also greatly reduces the specificity and practical significance of the models. We propose a comprehensive evaluation framework to analyze the spatial distribution of wildfire susceptibility and the effects of influencing factors, while assessing the risks of wildfire damage to the local ecological environment and urban development. In this study, we used wildfire information from the period 2013–2022 and data from 17 susceptibility factors in the city of Guilin as the basis, and utilized eight machine learning algorithms, namely logistic regression (LR), artificial neural network (ANN), K-nearest neighbor (KNN), support vector regression (SVR), random forest (RF), gradient boosting decision tree (GBDT), light gradient boosting machine (LGBM), and eXtreme gradient boosting (XGBoost), to assess wildfire susceptibility. By evaluating multiple indicators, we obtained the optimal model and used the Shapley Additive Explanations (SHAP) method to explain the effects of the factors and the decision-making mechanism of the model. In addition, we collected and calculated corresponding indicators, with the Remote Sensing Ecological Index (RSEI) representing ecological vulnerability and the Night-Time Lights Index (NTLI) representing urban development vulnerability. The coupling results of the two represent the comprehensive vulnerability of the ecology and city. Finally, by integrating wildfire susceptibility and vulnerability information, we assessed the risk of wildfire disasters in Guilin to reveal the overall distribution characteristics of wildfire disaster risk in Guilin. The results show that the AUC values of the eight models range from 0.809 to 0.927, with accuracy values ranging from 0.735 to 0.863 and RMSE values ranging from 0.327 to 0.423. Taking into account all the performance indicators, the XGBoost model provides the best results, with AUC, accuracy, and RMSE values of 0.927, 0.863, and 0.327, respectively. This indicates that the XGBoost model has the best predictive performance. The high-susceptibility areas are located in the central, northeast, south, and southwest regions of the study area. The factors of temperature, soil type, land use, distance to roads, and slope have the most significant impact on wildfire susceptibility. Based on the results of the ecological vulnerability and urban development vulnerability assessments, potential wildfire risk areas can be identified and assessed comprehensively and reasonably. The research results of this article not only can improve the specificity and practical significance of wildfire prediction models but also provide important reference for the prevention and response of wildfires.

Tan C., Feng Z.

Forest fire is a primary disaster that destroys forest resources and the ecological environment, and has a serious negative impact on the safety of human life and property. Predicting the probability of forest fires and drawing forest fire risk maps can provide a reference basis for forest fire control management in Hunan Province. This study selected 19 forest fire impact factors based on satellite monitoring hotspot data, meteorological data, topographic data, vegetation data, and social and human data from 2010–2018. It used random forest, support vector machine, and gradient boosting decision tree models to predict the probability of forest fires in Hunan Province and selected the RF algorithm to create a forest fire risk map of Hunan Province to quantify the potential forest fire risk. The results show that the RF algorithm performs best compared to the SVM and GBDT algorithms with 91.68% accuracy, 91.96% precision, 92.78% recall, 92.37% F1, and 97.2% AUC. The most important drivers of forest fires in Hunan Province are meteorology and vegetation. There are obvious differences in the spatial distribution of seasonal forest fire risks in Hunan Province, and winter and spring are the seasons with high forest fire risks. The medium- and high-risk areas are mostly concentrated in the south of Hunan.

Pang Y., Li Y., Feng Z., Feng Z., Zhao Z., Chen S., Zhang H.

Forest fires may have devastating consequences for the environment and for human lives. The prediction of forest fires is vital for preventing their occurrence. Currently, there are fewer studies on the prediction of forest fires over longer time scales in China. This is due to the difficulty of forecasting forest fires. There are many factors that have an impact on the occurrence of forest fires. The specific contribution of each factor to the occurrence of forest fires is not clear when using conventional analyses. In this study, we leveraged the excellent performance of artificial intelligence algorithms in fusing data from multiple sources (e.g., fire hotspots, meteorological conditions, terrain, vegetation, and socioeconomic data collected from 2003 to 2016). We have tested several algorithms and, finally, four algorithms were selected for formal data processing. There were an artificial neural network, a radial basis function network, a support-vector machine, and a random forest to identify thirteen major drivers of forest fires in China. The models were evaluated using the five performance indicators of accuracy, precision, recall, f1 value, and area under the curve. We obtained the probability of forest fire occurrence in each province of China using the optimal model. Moreover, the spatial distribution of high-to-low forest fire-prone areas was mapped. The results showed that the prediction accuracies of the four forest fire prediction models were between 75.8% and 89.2%, and the area under the curve (AUC) values were between 0.840 and 0.960. The random forest model had the highest accuracy (89.2%) and AUC value (0.96). It was determined as the best performance model in this study. The prediction results indicate that the areas with high incidences of forest fires are mainly concentrated in north-eastern China (Heilongjiang Province and northern Inner Mongolia Autonomous Region) and south-eastern China (including Fujian Province and Jiangxi Province). In areas at high risk of forest fire, management departments should improve forest fire prevention and control by establishing watch towers and using other monitoring equipment. This study helped in understanding the main drivers of forest fires in China over the period between 2003 and 2016, and determined the best performance model. The spatial distribution of high-to-low forest fire-prone areas maps were produced in order to depict the comprehensive views of China’s forest fire risks in each province. They were expected to form a scientific basis for helping the decision-making of China’s forest fire prevention authorities.

Carrasco G., Almeida A.C., Falvey M., Olmedo G.F., Taylor P., Santibañez F., Coops N.C.

Forest plantations in Chile occupy more than 2.2 million ha and are responsible for 2.1% of the GDP of the country's economy. The ability to accurately predictions of plantations productivity under current and future climate has an impact can enhance on forest management and industrial wood production. The use of process-based models to predict forest growth has been instrumental in improving the understanding and quantifying the effects of climate variability, climate change, and the impact of atmospheric CO2 concentration and management practices on forest growth. This study uses the 3-PG model to predict future forest productivity Eucalyptus globulus and Pinus radiata. The study integrates climate data from global circulation models used in CMIP5 for scenarios RCP26 and RCP85, digital soil maps for physical and chemical variables. Temporal and spatial tree growth inventories were used to compare with the 3-PG predictions. The results indicated that forest productivity is predicted to potentially increase stand volume (SV) over the next 50 years by 26% and 24% for the RCP26 scenario and between 73% and 62% for the RCP85 scenario for E. globulus and P. radiata, respectively. The predicted increases can be explained by a combination of higher level of atmospheric CO2, air temperatures closer to optimum than current, and increases in tree water use efficiency. If the effect of CO2 is not considered, the predicted differences of SV for 2070 are 16% and 14% for the RCP26 scenario and 22% and 14% for RCP85 for the two species. While shifts in climate and increasing CO2 are likely to benefit promote higher productivity, other factors such as lack insufficient availability of soil nutrients, events such as increasing frequency and duration of droughts, longer periods of extreme temperatures, competing vegetation, and occurrence of new pests and diseases may compromise these potential gains.

Shmuel A., Heifetz E.

Wildfires are a major natural hazard that lead to deforestation, carbon emissions, and loss of human and animal lives every year. Effective predictions of wildfire occurrence and burned areas are essential to forest management and firefighting. In this paper we apply various machine learning (ML) methods on a 0.25° monthly resolution global dataset of wildfires. We test the prediction accuracies of four different fire occurrence classifiers: random forest (RF), eXtreme Gradient Boosting (XGBoost), multilayer perceptron (MLP) neural network, and a logistic regression. Our best ML model predicts wildfire occurrence with over 90% accuracy, compared to approximately 70% using a logistic regression. We then train ML regression models to predict the size of burned areas and obtain an MAE score of 3.13 km2, compared to 7.48 km2 using a linear regression. To the best of our knowledge, this is the first study to be conducted in such resolution on a global dataset. We use the developed models to shed light on the influence of various factors on wildfire occurrence and burned areas. We suggest building upon these results to create ML-based fire weather indices.

Iban M.C., Sekertekin A.

In recent years, the number of wildfires has increased all over the world. Therefore, mapping wildfire susceptibility is crucial for prevention, early detection, and supporting wildfire management decisions. This study aims to generate Machine Learning (ML) based wildfire susceptibility maps for Adana and Mersin provinces, which are located in the Mediterranean Region of Turkey. To generate a wildfire inventory, this study uses active fire pixels derived from MODIS monthly MCD14ML composites. Furthermore, as a sub aim, the performance of seven ML approaches, namely, stand-alone Logistic Regression (LR), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and ensemble algorithms, namely Random Forest (RF), Gradient Boosting (GB), eXtreme Gradient Boosting (XGB), and AdaBoost (AB), was evaluated based on wildfire susceptibility mapping. The capabilities of the corresponding ML methods were assessed using thirteen wildfire conditioning factors, which can be grouped into four main categories: topographical, meteorological, vegetation, and anthropogenic factors . The Information Gain (IG) approach was used to assess their importance scores. A multicollinearity analysis was also performed to assess the relationship between conditioning factors. To compare the predictive performances of ML algorithms, five performance metrics, namely average accuracy, precision, recall, F1 score, and area under the curve, were used. To test the significance of the generated wildfire susceptibility maps and to detect similarities and differences among the output of these ML algorithms, McNemar's test was implemented. In the end, the ML-based models were locally interpreted using the Shapley Additive exPlanations (SHAP) technique. The AUC values of seven methods varied from 0.817 to 0.879, and the accuracy scores ranged between 0.734 and 0.812. The results showed that the RF model provided the best results considering all performance metrics. The accuracy score and AUC values of the RF model were equal to 0.812 and 0.879, respectively. On the other hand, stand-alone algorithms (LDA, SVM, and LR) represented lower performance than tree-based ensemble methods. Both the IG and SHAP analyses showed that elevation, temperature, and slope factors were the most contributing factors. The RF model classifier found that 7.20% of the study area has very high wildfire susceptibility, and the majority of the wildfire samples (68.84%) correspond to the very high susceptible areas in the RF model. The outcomes of this study are likely to provide decision-makers with a better understanding of wildfires in the Eastern Mediterranean Region of Turkey. • ML based wildfire susceptibility maps are generated. • MCD14ML product is used for wildfire inventory. • The performance of RF, GB, XGB, LR, SVM, AB, and LDA is investigated. • RF model provided the best results considering all performance metrics.