Wildfire Prediction Research Articles

Remotely sensed dry matter productivity and soil moisture content as potential predictors of arid rangeland wildfires: A case study of Kgalagadi District, Botswana

Fire is a critical tool for managing rangeland ecosystems; however, the increasing wildfire occurrence poses a considerable danger to rangeland ecosystem continuity. Predicting fire occurrence and mapping wildfire danger is critical in managing highly flammable rangelands. To identify potential remotely sensed variables for wildfire prediction, this study employed a Random Forest (RF) classifier using selected environmental variables to assess their possible use for wildfire prediction in Kgalagadi District, Botswana. The study used 107,883 active fire points from the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor from 2015 to 2021. Datasets of remotely sensed Dry Matter Productivity (DMP), Soil Moisture (SM), Land Surface Temperature (LST), Live Fuel Moisture Content (LFMC), and Dead Fuel Moisture Content (DMFC) were analysed in ArcMap 10.7 Esri©. The RF model developed gave an Out of Bag (OOB) error of 9.91% and an overall accuracy of 90.15% for classifying fires and non-fire points using the test dataset. The results also showed a Kappa coefficient of 0.803, with 88.25% and 91.76% producer and user accuracies, respectively, for classifying fire points. The DMP was the most significant variable with Mean Decrease Accuracy (MDA)= 1,055.20 and Mean Decrease Gini (MDG)= 9.328.62), followed by SM (MDA= 828.39 and MDG= 15,745). The LFMC and DMFC were found to be weak in detecting fires. It is recommended that field studies be carried out in the study area to calibrate these to improve their contribution to accurate fire prediction, as most literature shows that they are significant in fire prediction.

Forest Fire Prediction using Deep Learning

Wildfire prediction is an essential component of wildfire management. Accurate fire prediction is critical to mitigating its effects. It plays an important role in resource allocation, mitigation, and recovery. Wildfires rob the natural world of life and destroy the land they live on. Building forecast models will allow authorities to estimate the long-term effects of climate change on local forest distribution. Human motives include clearing and other agricultural activities, grassland care for livestock, extraction of non-timber forest products, industrial development, resettlement, hunting, neglect, and arson. Lightning strikes are the main cause of fires only in very remote areas of Canada and the Russian Federation. These wildfires are often man-made or caused by mother nature through varying weather conditions and wind. Wildfires economic and environmental impacts of wildfires are significant, and predictions can prevent long-term damage to sensitive forest ecosystems.

Modeling Wildfire Spread with an Irregular Graph Network

The wildfire prediction model is crucial for accurate rescue and rapid evacuation. Existing models mainly adopt regular grids or fire perimeters to describe the wildfire landscape. However, these models have difficulty in explicitly demonstrating the local spread details, especially in a complex landscape. In this paper, we propose a wildfire spread model with an irregular graph network (IGN). This model implemented an IGN generation algorithm to characterize the wildland landscape with a variable scale, adaptively encoding complex regions with dense nodes and simple regions with sparse nodes. Then, a deep learning-based spread model is designed to calculate the spread duration of each graph edge under variable environmental conditions. Comparative experiments between the IGN model and widely used fire simulation models were conducted on a real wildfire in Getty, California, USA. The results show that the IGN model can accurately and explicitly describe the spatiotemporal characteristics of the wildfire spread in a novel graph form while maintaining competitive simulation refinement and computational efficiency (Jaccard: 0.587, SM: 0.740, OA: 0.800).

Spatial pattern prediction of forest wildfire susceptibility in Central Yunnan Province, China based on multivariate data

Wildfires are an important disturbance factor in forest ecosystems. Assessing the probability of forest wildfires can assist in forest wildfire prevention, control, and supervision. The logistic regression model is widely used to forecast the probability, spatial patterns, and drivers of forest wildfires. This study used logistic regression to establish a spatial prediction model for forest wildfire susceptibility, which was applied to evaluate the risk of forest wildfires in Central Yunnan Province (CYP), China. A forest wildfire risk classification was implemented for CYP using forest burn scar data for 2001 to 2020 and the logistic spatial prediction model for forest wildfire susceptibility. Climate, vegetation, topographical, human activities, and location were selected as forest wildfire prediction variables. The results showed that: (1) The distributions of temperature, vegetation coverage, distance to water bodies, distance to roads, and precipitation were positively correlated with the occurrence of forest wildfires. Elevation, relative humidity, the global vegetation moisture index, wind speed, slope, latitude, and distance to residential areas were negatively correlated with the occurrence of forest wildfires. (2) The results of the logistic spatial prediction model for forest wildfire susceptibility showed a good fit to wildfire data, with an overall simulation probability of 81.6%. The optimal threshold for spatial prediction for forest wildfire susceptibility in CYP was determined to be 0.414. A significance level of a selected model variable of < 0.05 resulted in an area under the receiver operating characteristic curve (AUC) of 0.882–0.890. (3) Forest wildfire prevention efforts should focus on Southwest Yuxi City and southern Qujing City accounted for a high proportion of the areas at high risk of forest wildfires. Other localities should adjust forest wildfire prevention measures according to local conditions and strengthen existing wildfire prevention and emergency resource planning and allocation. (4) Some factors contributing to forest wildfires where different among the different areas. Forest wildfire risk factors had different degrees of impact under different spatial and temporal scales. The spatial relationships between wildfire disasters and influencing factors should be established in areas with heterogeneous environmental conditions for the selection of relevant factors.

Winter North Atlantic SST as a Precursor of Spring Eurasian Wildfire

AbstractBoreal fires considerably influence carbon release and air quality, but its occurrence mechanisms need to be explored. This study investigated the influence of winter North Atlantic sea surface temperature (SST) on spring Eurasian wildfire using single value decomposition method. Two possible pathways link such time‐lag phenomena. The anomalous SST pattern in winter can persist to spring, resulting in a significant high air‐pressure anomaly over southeastern Siberia, which corresponds to less‐than‐normal precipitation and enhanced aridity conditions and the increasing wildfire risk. Meanwhile, there are simultaneously negative snow cover anomalies over southeastern Siberia during winter which is closely related to the winter SSTs. Such snow anomalies persist from winter to spring, leading to negatively anomalous soil moisture. Both pathways provide a favorable environment for wildfire activities. This study implies that the winter North Atlantic SST could be a robust driver and potential precursor for spring wildfire prediction over southeastern Siberia.

Wildfire Prediction Model Based on Spatial and Temporal Characteristics: A Case Study of a Wildfire in Portugal’s Montesinho Natural Park

Wildfires influence the global carbon cycle, and the regularity of wildfires is mostly determined by elements such as meteorological conditions, combustible material states, and human activities. The time series and spatial dispersion of wildfires have been studied by some scholars. Wildfire samples were acquired in a monthly series for the Montesinho Natural Park historical fire site dataset (January 2000 to December 2003), which can be used to assess the possible effects of geographical and temporal variations on forest fires. Based on the above dataset, dynamic wildfire distribution thresholds were examined using a K-means++ clustering technique for each subgroup, and monthly series data were categorized as flammable or non-flammable depending on the thresholds. A five-fold hierarchical cross-validation strategy was used to train four machine learning models: extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and decision tree (DT). Finally, to explore the performance of those we have mentioned, we used accuracy (ACC), F1 score (F1), and the values for the area under the curve (AUC) of the receiver operating characteristics (ROCs). The results depicted that the XGBoost model works best under the evaluation of the three metrics (ACC = 0.8132, F1 = 0.7862, and AUC = 0.8052). The model performance is significantly improved when compared to the approach of classifying wildfires by burned area size (ACC = 72.3%), demonstrating that spatiotemporal heterogeneity has a broad influence on wildfire occurrence. The law of a spatiotemporal distribution connection in wildfires could aid in the prediction and management of wildfires and fire disasters.

Sentimental wildfire: a social-physics machine learning model for wildfire nowcasting

The intensity of wildfires and wildfire season length is increasing due to climate change, causing a greater threat to the local population. Much of this population are increasingly adopting social media, and sites like Twitter are increasingly being used as a real-time human-sensor network during natural disasters; detecting, tracking and documenting events. The human-sensor concept is currently largely omitted by wildfire models, representing a potential loss of information. By including Twitter data as a source in our models, we aim to help disaster managers make more informed, socially driven decisions, by detecting and monitoring online social media sentiment over the course of a wildfire event. This paper implements machine learning in a wildfire prediction model, using social media and geophysical data sources with Sentiment Analysis to predict wildfire characteristics with high accuracy. We also use wildfire-specific attributes to predict online social dynamics, as this has been shown to be indicative of localised disaster severity. This may be useful for disaster management teams in identifying areas of immediate danger. We combine geophysical satellite data from the Global Fire Atlas with social data provided by Twitter. We perform data collection and subsequent analysis & visualisation, and compare regional differences in online social sentiment expression. Following this, we compare and contrast different machine learning models for predicting wildfire attributes. We demonstrate social media is a predictor of wildfire activity, and present models which accurately model wildfire attributes. This work develops the concept of the human sensor in the context of wildfires, using users’ Tweets as noisy subjective sentimental accounts of current localised conditions. This work contributes to the development of more socially conscious wildfire models, by incorporating social media data into wildfire prediction and modelling.

A machine-learning approach for identifying dense-fires and assessing atmospheric emissions on the Indochina Peninsula, 2010–2020

Persistent and intensive wildland dense-fires (DFs) release substantial amounts of airborne pollutants, resulting in a sharp increase in emissions and leading to serious impacts on the environment and human health over extensive geographical areas. It is challenging to thoroughly investigate patterns of fire occurrence and fire distribution for predicting wildfire behaviour, and it is especially difficult to distinguish the characteristics of human-caused and climate-driven fires. Here, we identify and assess dense-fire (DF) from the perspective of spatiotemporally integrated processes using a machine-learning method based on a density-based clustering algorithm with noise constraint ratio. DFs represent collections of fires with homogenous behaviour and therefore allow the study of their internal features, which can reveal fixed patterns of fire occurrence and distribution as well as the evolution of fires over time. We estimated and labelled thousands of fire clusters on the Indochina Peninsula between 2010 and 2020, most of which occurred between December and May. For large-scale DFs, the number of fires contained and amount of atmospheric pollutants emitted were accounted for throughout most of the region, and the time, location and scale of their occurrence each year were relatively stable and predictable. Furthermore, the results of a secondary cluster analysis of fire interactions over the past decade showed two extreme fire events, labelled”north” and”south” groups, whose activities significantly impacted the atmospheric environment of the Indochina Peninsula. Additionally, we predicted their start/end dates and daily emissions. The study also found that the recurrence of high-density fires and the correlation between the DF edge and administrative border suggested a positive anthropogenic influence. To the authors' knowledge, this study is the first to analyze fires in a spatiotemporal Euclidean space by using density-based clustering, with high-density fires as independent subjects to study fire behaviour. The method proposed in this study can provide a reference for wildfire prediction and emission forecasting and fire control work.

Prediction of Wildfires Based on the Spatio-Temporal Variability of Fire Danger Factors

Most methods in the field of wildfire prevention are based on expert assessment of fire danger factors. However, their weights are usually assumed constant for the entire application area despite the geographical and seasonal changes of factors. This study aimed to develop a wildfire prevention method based on partial and general fire danger ratings taking into account their spatio-temporal variability. The study was conducted for Krasnoyarsk territory, Orenburg region and the Meschera lowland as the most forest, steppe and peat fire dangerous regions of Russia respectively. Surface temperature, moisture, vegetation structure, anthropogenic load, topography and their variation over subzones and in time were used as fire danger factors. They were evaluated by measuring parameters such as radiobrightness temperature, Normalized Difference Water Index (NDWI), Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), distance to settlements and roads, elevation, slope and aspect. Materials from the Terra/Aqua, Sentinel-3, Landsat-8, Sentinel-2 satellites, ASTER Global Digital Elevation Model and Open Street Maps vector layers were used in the study. Correlation between these parameters and the actual fires in 2016-2018 was analyzed. Linear relationships were established, and correlation coefficients, equations of partial ratings and prevention 90%-threshold values were identified. On their basis, the parameter weights were computed to integrate them into the general fire danger rating. The developed method was validated using data over 2019. The results showed 67% confidence and 61% reliability of fire prevention along with the spatio-temporal patterns of fire danger factors. The method is recommended for preventing wildfires within the study areas and can be extend to similar regions.

Dynamic prediction of global monthly burned area with hybrid deep neural networks.

Wildfires not only severely damage the natural environment and global ecological balance but also cause substantial losses to global forest resources and human lives and property. Unprecedented fire events such as Australia's bushfires have alerted us to the fact that wildfire prediction is a critical scientific problem for fire management. Therefore, robust, long-lead models and dynamic predictions of wildfire are valuable for global fire prevention. However, despite decades of effort, the dynamic, effective, and accurate prediction of wildfire remains problematic. There is great uncertainty in predicting the future based on historical and existing spatiotemporal sequence data, but with advances in deep learning algorithms, solutions to prediction problems are being developed. Here, we present a dynamic prediction model of global burned area of wildfire employing a deep neural network (DNN) approach that produces effective wildfire forecasts based on historical time series predictors and satellite-based burned area products. A hybrid DNN that combines long short-term memory and a two-dimensional convolutional neural network (CNN2D-LSTM) was proposed, and CNN2D-LSTM model candidates with four different architectures were designed and compared to construct the optimal architecture for fire prediction. The proposed model was also shown to outperform convolutional neural networks (CNNs) and the fully connected long short-term memory (FcLSTM) approach using the refined index of agreement and evaluation metrics. We produced monthly global burned area spatiotemporal prediction maps and adequately reflected the seasonal peak in fire activity and highly fire-prone areas. Our combined CNN2D-LSTM approach can effectively predict the global burned area of wildfires 1 month in advance and can be generalized to provide seasonal estimates of global fire risk.

Data‐driven spatio‐temporal analysis of wildfire risk to power systems operation

Wildfires are natural or man-made disasters that continuously threaten portions of the transmission and distribution grid, and thus the stability of the electric grid. This paper presents a two-stage framework for assessing power system-wildfire risk using a data-driven wildfire prediction model. The first stage of the framework estimates the spatio-temporal probability of potential wildfire ignition and propagation using a deep neural network in combination with the wildfire physical spread model. Analysis reveals similar spatial and temporal patterns between the model-predicted wildfire ignition potential and actual wildfire ignition. Motivated by these observations, the second stage assesses the wildfire risk in the power grid operation in terms of potential loss of load by de-energisation, through combining geospatial information system data of the power grid topology and the stochastic spatio-temporal wildfire model developed in the first stage. The electric power utility applications introduced by the proposed framework are twofold: 1) a spatio-temporal risk model for proactive de-energisation against potential power system failure-induced wildfire, and 2) a spatio-temporal spreading model for optimal grid operations against exogenous wildfire. The proposed model, based on real-world dataset, is demonstrated on the IEEE 24-bus test system mapped to a study area in Northern California, while the results illustrate the proposed model can achieve the best performance in potential wildfire ignition detection (AUC of 0.995) compared to other baselines, as well as demonstrates the risk-aware operation of the power system enabled by the proposed framework.

High-resolution mapping of wildfire drivers in California based on machine learning

Wildfires are important natural disturbances of ecosystems; however, they threaten the sustainability of ecosystems, climate and humans worldwide. It is vital to quantify and map the controlling drivers of wildfires for effective wildfire prediction and risk management. However, high-resolution mapping of wildfire drivers remains challenging. Here we established machine-learning (Random Forests) models using 23 climate and land surface variables as model inputs to reconstruct the spatial variability and seasonality of wildfire occurrence and extent in California. The importance of individual drivers was then quantified based on the Shapley value method. Thus, we provided spatially resolved maps of wildfire drivers at high resolutions up to 0.004° × 0.004°. The results indicated that precipitation and soil moisture are the major drivers dominating 37% of the total burnt area for large and extreme wildfires in summer and 63% in autumn, while elevation plays a major role for 15–58% of burnt areas in small wildfires in all seasons. Winds are also an important contributor to summer wildfires, accounting for 41% of large and extreme burnt areas. This study enhanced our knowledge of spatial variability of wildfire drivers across diverse landscapes in a fine-scale mapping, providing valuable perspectives and case studies for other regions of the world with frequently occurred wildfire.

Estimating Next Day’s Forest Fire Risk via a Complete Machine Learning Methodology

Next day wildfire prediction is an open research problem with significant environmental, social, and economic impact since it can produce methods and tools directly exploitable by fire services, assisting, thus, in the prevention of fire occurrences or the mitigation of their effects. It consists in accurately predicting which areas of a territory are at higher risk of fire occurrence each next day, exploiting solely information obtained up until the previous day. The task’s requirements in spatial granularity and scale of predictions, as well as the extreme imbalance of the data distribution render it a rather demanding and difficult to accurately solve the problem. This is reflected in the current literature, where most existing works handle a simplified or limited version of the problem. Taking into account the above problem specificities, in this paper, we present a machine learning methodology that effectively (sensitivity &gt; 90%, specificity &gt; 65%) and efficiently performs next day fire prediction, in rather high spatial granularity and in the scale of a country. The key points of the proposed approach are summarized in: (a) the utilization of an extended set of fire driving factors (features), including topography-related, meteorology-related and Earth Observation (EO)-related features, as well as historical information of areas’ proneness to fire occurrence; (b) the deployment of a set of state-of-the-art classification algorithms that are properly tuned/optimized on the setting; (c) two alternative cross-validation schemes along with custom validation measures that allow the optimal and sound training of classification models, as well as the selection of different models, in relation to the desired trade-off between sensitivity (ratio of correctly identified fire areas) and specificity (ratio of correctly identified non-fire areas). In parallel, we discuss pitfalls, intuitions, best practices, and directions for further investigation derived from our analysis and experimental evaluation.

Characterization of global wildfire burned area spatiotemporal patterns and underlying climatic causes

Wildfires are widespread disasters and are concurrently influenced by global climatic drivers. Due to the widespread and far-reaching influence of climatic drivers, separate regional wildfires may have similar climatic cause mechanisms. Determining a suite of global climatic drivers that explain most of the variations in different homogeneous wildfire regions will be of great significance for wildfire management, wildfire prediction, and global wildfire climatology. Therefore, this study first identified spatiotemporally homogeneous regions of burned area worldwide during 2001–2019 using a distinct empirical orthogonal function. Eight patterns with different spatiotemporal characteristics were identified. Then, the relationships between major burned area patterns and sixteen global climatic drivers were quantified based on wavelet analysis. The most significant global climatic drivers that strongly impacted each of the eight major wildfire patterns were identified. The most significant combinations of hotspots and climatic drivers were Atlantic multidecadal Oscillation-East Pacific/North Pacific Oscillation (EP/NP)-Pacific North American Pattern (PNA) with the pattern around Ukraine and Kazakhstan, El Niño/Southern Oscillation-Arctic Oscillation (AO)-East Atlantic/Western Russia Pattern (EA/WR) with the pattern in Australia, and PNA-AO-Polar/Eurasia Pattern-EA/WR with the pattern in Brazil. Overall, these results provide a reference for predicting wildfire and understanding wildfire homogeneity.

A wildfire growth prediction and evaluation approach using Landsat and MODIS data

The increasing wildfire occurrence due to global climate changes urged the improvement of present wildfire growth prediction and evaluation methods. This study aimed to propose novel solutions to their two primary limitations, including the lack of robust fuel classification method and the low spatial resolution of wildfire growth accuracy assessment while ensuring wide applicability using open data satellite missions and software. The first objective was to create a robust two-step fuel model classification method consisted of the supervised machine learning classification of generalized land cover classes in the 1st level and their individual unsupervised classification to vegetation subtypes in the 2nd level. The second objective was creating a wildfire prediction accuracy assessment method using MODIS 250 m images, which overcome the limitations of low spatial resolution while preserving sub-daily temporal resolution. The wildfire on the Korčula island in Croatia was analyzed in the study, being specific for its long duration from 18 to 24 July 2015. The wildfire ignition occurred in the isolated area, which prolonged the response time from emergency agencies. Random Forest (RF) with input Landsat 8 spectral bands and indices resulted in the highest classification accuracy in the 1st classification level with an overall agreement of 83.6%. The vegetation subclasses from the 2nd classification level were matched to the 13 standard fuel models for the input in FARSITE software. The predicted wildfire evaluation showed the highest mean accuracy of 0.906 for the first two days, which decreased to 0.722 in the latter stages of the active wildfire caused by overprediction. The proposed two-step fuel model classification presented a cost-efficient solution to the fuel map creation in any part of the world, with a disadvantage of no in-situ ground truth identification and accuracy assessment for 2nd classification level. The evaluation of wildfire growth prediction with 250 m images enabled high spatial and temporal resolution of the assessment, while its limitations of wildfire overprediction and the negative effects of wildfire smoke in MODIS images should be addressed in future research.

Important meteorological predictors for long-range wildfires in China

Wildfire predictions provide useful information for fire management planning and implementation. Temperature and precipitation have been used as the primary meteorological predictors for wildfires in China. This study is to improve the prediction skills of long-range (monthly, seasonal, and annual) wildfires in China by identifying other important meteorological predictors. Provincial data during 1999–2020 were used to calculate the correlations between fire properties (fire count and burned area) and meteorological variables of maximum temperature, precipitation, relative humidity, wind speed, and vapor pressure deficit (VPD) and drought indices of Keetch-Bryam Drought Index (KBDI), Palmer Drought Severity Index (PDSI), and Standardized Precipitation Index (SPI). The fitting rates of the linear regression fire prediction models were compared among these meteorological variables and drought indices. The results indicate that the number of provinces with significant correlations and / or high fitting rates is the largest with VPD for monthly fires, KBDI for seasonal fires, and KBDI, PDSI, and SPI for annual fires. The number is larger in Northeast, Central, and South China than in other China regions. The number is comparable between spring and other seasons for KBDI but often smaller in spring for meteorological variables. The number is generally smaller for burned areas than fire count. It is concluded that the skills of long-range fire predictions are expected to be improved in many provinces of China by using VPD and KBDI as well as some other drought indices.

MODIS-FIRMS and ground-truthing-based wildfire likelihood mapping of Sikkim Himalaya using machine learning algorithms

Wildfires in limited extent and intensity can be a boon for the forest ecosystem. However, recent episodes of wildfires of 2019 in Australia and Brazil are sad reminders of their heavy ecological and economical costs. Understanding the role of environmental factors in the likelihood of wildfires in a spatial context would be instrumental in mitigating it. In this study, 15 environmental features encompassing meteorological, topographical, ecological, in situ and anthropogenic factors have been considered for preparing the wildfire likelihood map of Sikkim Himalaya. A comparative study on the efficiency of machine learning methods like Generalized Linear Model, Support Vector Machine, Random Forest (RF) and Gradient Boosting Model (GBM) has been performed to identify the best performing algorithm in wildfire prediction. The study indicates that all the machine learning methods are good at predicting wildfires. However, RF has outperformed, followed by GBM in the prediction. Also, environmental features like average temperature, average wind speed, proximity to roadways and tree cover percentage are the most important determinants of wildfires in Sikkim Himalaya. This study can be considered as a decision support tool for preparedness, efficient resource allocation and sensitization of people towards mitigation of wildfires in Sikkim.

Boots on the Ground and Eyes in the Sky: A Perspective on Estimating Fire Danger from Soil Moisture Content

With increasing forest and grassland wildfire trends strongly correlated to anthropogenic climate change, assessing wildfire danger is vital to reduce catastrophic human, economic, and environmental loss. From this viewpoint, the authors discuss various approaches deployed to evaluate wildfire danger, from in-situ observations to satellite-based fire prediction systems. Lately, the merit of soil moisture in predicting fuel moisture content and the likelihood of wildfire occurrence has been widely realized. Harmonized soil moisture measurement initiatives via state-of-the-art soil moisture networks have facilitated the use of soil moisture information in developing innovative applications for wildfire prediction and risk management applications. Additionally, the increasing availability of remote-sensing data has enabled the monitoring and modeling of wildfires across various terrestrial ecosystems. When coupled with remotely sensed data, field-based soil moisture measurements have been more valuable predictors of assessing wildfire than alone. However, sensors capable of acquiring higher spectral information and radiometry across large spatiotemporal domains are still lacking. The automation aspect of such extensive data from remote-sensing and field data is needed to rapidly assess wildfire and mitigation of wildfire-related damage at operational scales.

Interannual variability and climatic sensitivity of global wildfire activity

Understanding historical wildfire variations and their environmental driving mechanisms is key to predicting and mitigating wildfires. However, current knowledge of climatic responses and regional contributions to the interannual variability (IAV) of global burned area remains limited. Using recent satellite-derived wildfire products and simulations from version v1.0 of the land component of the U.S. Department of Energy's Energy Exascale Earth System Model (E3SM land model [ELM] v1) driven by three different climate forcings, we investigated the burned area IAV and its climatic sensitivity globally and across nine biomes from 1997 to 2018. We found that 1) the ELM simulations generally agreed with the satellite observations in terms of the burned area IAV magnitudes, regional contributions, and covariations with climate factors, confirming the robustness of the ELM to the usage of different climate forcing sources; 2) tropical savannas, tropical forests, and semi-arid grasslands near deserts were primary contributors to the global burned area IAV, collectively accounting for 71.7%–99.7% of the global wildfire IAV estimated by both the satellite observations and ELM simulations; 3) precipitation was a major fire suppressing factor and dominated the global and regional burned area IAVs, and temperature and shortwave solar radiation were mostly positively related with burned area IAVs; and 4) noticeable local discrepancies between the ELM and remote-sensing results occurred in semi-arid grasslands, croplands, boreal forests, and wetlands, likely caused by uncertainties in the current ELM fire scheme and the imperfectly derived satellite observations. Our findings revealed the spatiotemporal diversity of wildfire variations, regional contributions and climatic responses, and provided new metrics for wildfire modeling, facilitating the wildfire prediction and management.

Identifying Key Drivers of Wildfires in the Contiguous US Using Machine Learning and Game Theory Interpretation.

Understanding the complex interrelationships between wildfire and its environmental and anthropogenic controls is crucial for wildfire modeling and management. Although machine learning (ML) models have yielded significant improvements in wildfire predictions, their limited interpretability has been an obstacle for their use in advancing understanding of wildfires. This study builds an ML model incorporating predictors of local meteorology, land‐surface characteristics, and socioeconomic variables to predict monthly burned area at grid cells of 0.25° × 0.25° resolution over the contiguous United States. Besides these predictors, we construct and include predictors representing the large‐scale circulation patterns conducive to wildfires, which largely improves the temporal correlations in several regions by 14%–44%. The Shapley additive explanation is introduced to quantify the contributions of the predictors to burned area. Results show a key role of longitude and latitude in delineating fire regimes with different temporal patterns of burned area. The model captures the physical relationship between burned area and vapor pressure deficit, relative humidity (RH), and energy release component (ERC), in agreement with the prior findings. Aggregating the contribution of predictor variables of all the grids by region, analyses show that ERC is the major contributor accounting for 14%–27% to large burned areas in the western US. In contrast, there is no leading factor contributing to large burned areas in the eastern US, although large‐scale circulation patterns featuring less active upper‐level ridge‐trough and low RH two months earlier in winter contribute relatively more to large burned areas in spring in the southeastern US.