Wildfire Prediction Research Articles

Enhancing Alaskan wildfire prediction and carbon flux estimation: a two-stage deep learning approach within a process-based model

Abstract Wildfires in boreal forests release substantial amounts of carbon into the atmosphere. However, current land-surface models are limited in their representation of fire processes, including their ignition and spread. This study thus developed FireDL, a novel data-driven machine-learning model for the prediction of natural wildfires, and combined it with a land-surface model to better understand the impact of fire on carbon fluxes. FireDL has a two-stage deep learning structure that sequentially combines a long short-term memory (LSTM) algorithm and an artificial neural network (ANN). Preliminary random forest analysis identified fire duration as an important factor in predicting the burned area. Thus, in FireDL, the LSTM algorithm was employed to predict fire occurrence and duration, utilizing lightning, vegetation, and climate datasets. Subsequently, the ANN predicted the total burned area using the LTSM-derived fire duration predictions and climate datasets as input. FireDL produced a robust performance in predicting large fires (>10 000 ha), achieving a correlation coefficient of 0.72. The daily-scaled burned area predictions derived from FireDL were integrated into the Community Land Model version 5—Biogeochemistry (CLM5-BGC) to produce CLM5-BGC-FireDL. This integration considerably improved carbon emission estimations. Notably, the total net ecosystem exchange (NEE) estimated using CLM5-BGC-FireDL in 2019, the year with the highest recorded burned area during our study, was twice that estimated using the standard CLM5-BGC. Discrepancies in the NEE can significantly influence atmospheric CO2 levels, highlighting the importance of our fire prediction model in forecasting the burned area and carbon emissions. The use of FireDL with future climate scenarios is thus anticipated to yield valuable insights into ecosystem management and climate change mitigation strategies.

Experimental and numerical study on data-driven prediction for wildfire spread incorporating adaptive observation error adjustment

In recent wildfire prediction research, data assimilation (DA) methods like Ensemble Kalman filtering have gained traction for integrating observation data to enhance prediction accuracy. Most previous studies trusted that the observation data were accurate, and set a small observation error, which causes unreliable predicted results for scenarios with large observation error. To tackle this, our study introduced a method that iteratively adjusted the potential range of observation errors by comparing observation and simulation data over time. We conducted a 30-m experiment and kilometer-scale numerical simulations. Unlike prior research, we adopted larger error ranges (the similarity index with true data ranges from 0.6 to 1) for both real and synthetic observation data. In the experiment, to increase the complexity of fire spread, a heterogeneous fuel arrangement was employed. Irregular flame fronts appeared due to incomplete combustion and were difficult to replicate in simulations. Better accuracy was achieved using real observation data to revise predictions. Furthermore, to improve the applicability of the algorithm, numerical simulations were designed to consider observation error changing over time or not. The Root Mean Square Errors for the fire front prediction using the proposed method remained lower than that of the traditional DA approach.

The distributed strategy for asynchronous observations in data-driven wildland fire spread prediction

Background Asynchronous observations refer to observations that are obtained at multiple moments. The observation moments of fire fronts may differ throughout an entire wildfire area. Asynchronous observations include historical data, which hinders the effectiveness of data assimilation due to the lack of timely updates on changing fire fronts. Aims This paper proposed a distributed strategy combined with the Ensemble Transform Kalman filter (ETKF-distributed) for asynchronous observations. It can assimilate fire fronts immediately at any location by using new matching schemes between prediction and observation. Methods The ETKF-distributed undergoes testing using a wildland fire generated based on real terrain, vegetation, and historical weather data from the local area. In addition, the ETKF and ETKF-centralised proposed in our previous work were employed as comparisons. Observing System Simulation Experiments were conducted to generate asynchronous observation fire fronts. Key results The benefit of immediate assimilation enables the new method to maintain high accuracy predictions. Conclusions The allocation of observation resources can be focused in regions with high rates of speed when employing ETKF-distributed. Implications The ETKF-distributed has high efficiency and adaptability, making it highly promising for implementation in wildfire prediction.

Research Trends in Wildland Fire Prediction Amidst Climate Change: A Comprehensive Bibliometric Analysis

Wildfire prediction plays a vital role in the management and conservation of forest ecosystems. By providing detailed risk assessments, it contributes to the reduction of fire frequency and severity, safeguards forest resources, supports ecological stability, and ensures human safety. This study systematically reviews wildfire prediction literature from 2003 to 2023, emphasizing research trends and collaborative trends. Our findings reveal a significant increase in research activity between 2019 and 2023, primarily driven by the United States Forest Service and the Chinese Academy of Sciences. The majority of this research was published in prominent journals such as the International Journal of Wildland Fire, Forest Ecology and Management, Remote Sensing, and Forests. These publications predominantly originate from Europe, the United States, and China. Since 2020, there has been substantial growth in the application of machine learning techniques in predicting forest fires, particularly in estimating fire occurrence probabilities, simulating fire spread, and projecting post-fire environmental impacts. Advanced algorithms, including deep learning and ensemble learning, have shown superior accuracy, suggesting promising directions for future research. Additionally, the integration of machine learning with cellular automata has markedly improved the simulation of fire behavior, enhancing both efficiency and precision. The profound impact of climate change on wildfire prediction also necessitates the inclusion of extensive climate data in predictive models. Beyond conventional studies focusing on fire behavior and occurrence probabilities, forecasting the environmental and ecological consequences of fires has become integral to forest fire management and vital for formulating more effective wildfire strategies. The study concludes that significant regional disparities in knowledge exist, underscoring the need for improved research capabilities in underrepresented areas. Moreover, there is an urgent requirement to enhance the application of artificial intelligence algorithms, such as machine learning, deep learning, and ensemble learning, and to intensify efforts in identifying and leveraging various wildfire drivers to refine prediction accuracy. The insights generated from this field will profoundly augment our understanding of wildfire prediction, assisting policymakers and practitioners in managing forest resources more sustainably and averting future wildfire calamities.

Physical, social, and biological attributes for improved understanding and prediction of wildfires: FPA FOD-Attributes dataset

Abstract. Wildfires are increasingly impacting social and environmental systems in the United States (US). The ability to mitigate the adverse effects of wildfires increases with understanding of the social, physical, and biological conditions that co-occurred with or caused the wildfire ignitions and contributed to the wildfire impacts. To this end, we developed the FPA FOD-Attributes dataset, which augments the sixth version of the Fire Program Analysis Fire-Occurrence Database (FPA FOD v6) with nearly 270 attributes that coincide with the date and location of each wildfire ignition in the US. FPA FOD v6 contains information on location, jurisdiction, discovery time, cause, and final size of >2.3×106 wildfires in the US between 1992 and 2020 . For each wildfire, we added physical (e.g., weather, climate, topography, and infrastructure), biological (e.g., land cover and normalized difference vegetation index), social (e.g., population density and social vulnerability index), and administrative (e.g., national and regional preparedness level and jurisdiction) attributes. This publicly available dataset can be used to answer numerous questions about the covariates associated with human- and lightning-caused wildfires. Furthermore, the FPA FOD-Attributes dataset can support descriptive, diagnostic, predictive, and prescriptive wildfire analytics, including the development of machine learning models. The FPA FOD-Attributes dataset is available at https://doi.org/10.5281/zenodo.8381129 (Pourmohamad et al., 2023).

Smart City Wildfire Risk Analysis with Fuzzy Multi-Criteria Decision-Making

Uncontrolled wildfires pose a significant threat, potentially causing extensive damage to biodiversity, soil quality and human resources. It’s crucial to swiftly detect and predict these wildfires to minimize their catastrophic consequences. To address this, our research introduces a wildfire prediction model that ranks cities based on risk leveraging multi-criteria decision-making (MCDM) to systematically assess conflicting factors in decision-making. This model integrates wildfire risks into a city’s resilience strategy, utilizing fuzzy set theory to manage imprecise data and uncertainties. As part of this approach, we compile a new dataset encompassing weather patterns, vegetation types, terrain features and population density across various Californian cities. Ultimately, the model assesses and ranks the wildfire risk for each city in California.

Wildfire Prediction in the United States Using Time Series Forecasting Models

Wildfires are a widespread phenomenon that affects every corner of the world with the warming climate. Wildfires burn tens of thousands of square kilometres of forests and vegetation every year in the United States alone with the past decade witnessing a dramatic increase in the number of wildfire incidents. This research aims to understand the regions of forests and vegetation across the US that are susceptible to wildfires using spatiotemporal kernel heat maps and, forecast these wildfires across the United States at country-wide and state levels on a weekly and monthly basis in an attempt to reduce the reaction time of the suppression operations and effectively design resource maps to mitigate wildfires. We employed the state-of-the-art Neural Basis Expansion Analysis for Time Series (N-BEATS) model to predict the total area burned by wildfires by several weeks and months into the future. The model was evaluated based on forecasting metrics including mean-squared error (MSE)., and mean average error (MAE). The N-BEATS model demonstrates improved performance compared to other state-of-the-art (SOTA) models, obtaining MSE values of 116.3, 38.2, and 19.0 for yearly, monthly, and weekly forecasting, respectively.

How well do multi-fire danger rating indices represent China forest fire variations across multi-time scales?

To better support wildfire predictions and risk assessment, multiple fire danger rating indices (FDRIs) have been developed but their credibility in China remains obscure. Compared with the satellite fire observations, 13 FDRIs are evaluated for the historical (2003–2021) forest fire frequency in China from four different time scales: active seasons, trends, interannual variations (IAVs) and discrimination of fire/non-fire days (DFDs). Most FDRIs effectively capture the double active seasons over Southwest China and the dominant active season over Northeast and South China but fail over the other regions. FDRIs with cloud cover perform better in capturing climatological fire seasonality. All FDRIs fail to reproduce the significant decreasing trend of forest fires speculatively due to local fire management and discordant changes in meteorological elements. Most FDRIs have the advantages of the IAVs and DFDs over Southwest China but exhibit deficiencies over the other regions. FDRIs incorporating wind speed perform best in representing both IAVs and DFDs, indicating the indispensable effect of surface wind on the interannual/daily variation of fire danger. This study provides a credible reference for utilizing FDRIs in China, as well as offers insights for developing better regional FDRIs to represent different time-scale variations.

An escape route planning model based on wildfire prediction information and travel rate of firefighters

Background When firefighters evacuate from wildfires, escape routes are crucial safety measures, providing pre-defined pathways to a safety zone. Their key evaluation criterion is the time it takes for firefighters to travel along the planned escape routes. Aims While shorter travel times can help firefighters reach safety zones faster, this may expose them to the threat of wildfires. Therefore, the safety of the routes must be considered. Methods We introduced a new evaluation indicator called the safety index by predicting the growth trend of wildfires. We then proposed a comprehensive evaluation cost function as an escape route planning model, which includes two factors: (1) travel time; and (2) safety of the escape route. The relationship between the two factors is dynamically adjusted through real time factor. The safety window within real time factor provides ideal safety margins between firefighters and wildfires, ensuring the overall safety of escape routes. Key results Compared with other models, the escape routes planned by the final improved model not only effectively avoid wildfires, but also provide relatively short travel time and reliable safety. Conclusions This study ensures sufficient safety margins for firefighters escaping in wildfire environments. Implications The escape route model described in this study offers a broader perspective on the study of escape route planning.

Deep learning modeling of human activity affected wildfire risk by incorporating structural features: A case study in eastern China

Wildfire risk prediction is a critical component of disaster prevention and mitigation, often closely associated with local human activities in most regions. Recent studies demonstrate that employing joint modeling techniques using diverse datasets alongside Convolutional Neural Networks-Long Short-Term Memory Networks (ConvLSTM) produces favorable predictive results. However, previous research inadequately explored the different impact of factors across different categories and spatial orientations, and neglected the impact of fuels and human activities inside the samples. This study focuses on the six eastern provinces of China, utilizing a multi-source dataset comprising satellite-monitored wildfire products from 2012 to 2022, along with various factors indicating terrestrial and human activities, simulated meteorological elements and high-resolution vegetation imagery. By introducing channel and spatial attention mechanisms and visual transformer mode, this research optimizes the ConvLSTM wildfire prediction model. Results indicate a noteworthy enhancement, elevating accuracy, Kappa coefficient, and AUC of ROC curves from 91.15%, 80.87%, and 97.01% to 92.79%, 84.48%, and 97.90%, respectively. Consequently, it reinforces the accuracy by increase of the structural features within samples and quantifying the differences in the importance of different factors, which is also validated by prediction application of the samples in the entire year of 2023. Sensitivity analysis reveals that the current model is still highly dependent on the meteorological factors. Notably, the impact of structural features significantly surpasses the influence of terrain and terrestrial ecology elements, which should be considered in further models. Thus, this study has developed a methodology integrating multiple attention mechanisms and sample structural features, which could furnish high-precision daily kilometer-level wildfire risk prediction products. This method could improve the efficiency of prevention and control by improving the accuracy and narrowing the high-risk areas.

Global Wildfire Danger Predictions Based on Deep Learning Taking into Account Static and Dynamic Variables

Climate change will intensify the danger of wildfires, significantly impacting human life. Deep Learning (DL) has been extensively applied in wildfire prediction research. In the realm of wildfire prediction, previous deep learning methods have overlooked the inherent differences between static positional information and dynamic variables. Additionally, most existing deep learning models have not integrated the global system characteristics of the Earth’s features and teleconnection during the learning phase. Here, we propose a static location-aware ConvLSTM (SLA-ConvLSTM) model that is aware of static positional elements and interconnected with global information and teleconnection. The model we propose can discern the influence of dynamic variables across various geographical locations on predictive outcomes. Compared with other deep learning models, our SLA-ConvLSTM model has achieved commendable performance. The outcomes indicate that the collaborative interplay of spatiotemporal features and the extraction of static positional information present a promising technique for wildfire prediction. Moreover, the incorporation of climate indices and global feature variables enhances the predictive capability of the model in wildfire prediction.

A global fuel characteristic model and dataset for wildfire prediction

Abstract. Effective wildfire management and prevention strategies depend on accurate forecasts of fire occurrence and propagation. Fuel load and fuel moisture content are essential variables for forecasting fire occurrence, and whilst existing operational systems incorporate dead fuel moisture content, both live fuel moisture content and fuel load are either approximated or neglected. We propose a mid-complexity model combining data driven and analytical methods to predict fuel characteristics. The model can be integrated into earth system models to provide real-time forecasts and climate records taking advantage of meteorological variables, land surface modelling, and satellite observations. Fuel load and moisture is partitioned into live and dead fuels, including both wood and foliage components. As an example, we have generated a 10-year dataset which is well correlated with independent data and largely explains observed fire activity globally. While dead fuel moisture correlates highest with fire activity, live fuel moisture and load are shown to potentially enhance prediction skill. The use of observation data to inform a dynamical model is a crucial first step toward disentangling the contributing factors of fuel and weather to understand fire evolution globally. This dataset, with high spatiotemporal resolution (∼9 km, daily), is the first of its kind and will be regularly updated.

Loss functions for spatial wildfire applications

Spatial predictions of wildfire spread are used operationally and in risk estimation. It is important that their outputs are validated to quantify predictive performance and uncertainty. There are numerous loss functions for this simulation validation. Our paper synthesises ten common and five novel loss functions for evaluating the performance of wildfire predictions against observed or simulated wildfires. We describe each loss function; test their sensitivity to scale, rotation and translation; and demonstrate their application with three case study wildfires in south-eastern Australia. Based on their purpose, there are three categories of loss functions: general overlap, partial overlap and distance. Within these categories, there is functional redundancy as many loss functions are appropriate and perform similarly. Using loss functions from across the three different categories is the most comprehensive for evaluating wildfire models, but the choice depends on the simulation purpose and interpretation.

Enhancing Wildfire Forecasting Through Multisource Spatio-Temporal Data, Deep Learning, Ensemble Models and Transfer Learning

This paper presents a novel approach in wildfire prediction through the integration of multisource spatiotemporal data, including satellite data, and the application of deep learning techniques. Specifically, we utilize an ensemble model built on transfer learning algorithms to forecast wildfires. The key focus is on understanding the significance of weather sequences, human activities, and specific weather parameters in wildfire prediction. The study encounters challenges in acquiring real-time data for training the network, especially in Moroccan wildlands. The future work intends to develop a global model capable of processing multichannel, multidimensional, and unformatted data sources to enhance our understanding of the future entropy of surface tiles.

Deep Learning Surrogate Models of JULES-INFERNO for Wildfire Prediction on a Global Scale

Deep Learning Surrogate Models of JULES-INFERNO for Wildfire Prediction on a Global Scale

Analysis of Wildfire Danger Level Using Logistic Regression Model in Sichuan Province, China

Sichuan Province preserves numerous rare and ancient species of plants and animals, making it an important bio-genetic repository in China and even the world. However, this region is also vulnerable to fire disturbance due to the rich forest resources, complex topography, and dry climate, and thus has become one of main regions in China needing wildfire prevention. Analyzing the main driving factors influencing wildfire incidence can provide data and policy guidance for wildfire management in Sichuan Province. Here we analyzed the spatial and temporal distribution characteristics of wildfires in Sichuan Province based on the wildfire spot data during 2010–2019. Based on 14 input variables, including climate, vegetation, human factors, and topography, we applied the Pearson correlation analysis and Random Forest methods to investigate the most important factors in driving wildfire occurrence. Then, the Logistic model was further applied to predict wildfire occurrences. The results showed that: (1) The southwestern Sichuan Province is a high-incidence area for wildfires, and most fires occurred from January to June. (2) The most important factor affecting wildfire occurrence is monthly average temperature, followed by elevation, monthly precipitation, population density, Normalized Difference Vegetation Index (NDVI), NDVI in the previous month, and Road kernel density. (3) The Logistic wildfire prediction model yielded good performance, with the area under curve (AUC) values higher than 0.94, overall accuracy (OA) higher than 86%, true positive rate (TPR) values higher than 0.82, and threat score (TS) values higher than 0.71. The final selected prediction model has an AUC of 0.944, an OA of 87.28%, a TPR of 0.829, and a TS of 0.723. (4) The results of the prediction indicate that extremely high danger of wildfires (probability of fire occurrence higher than 0.8) is concentrated in the southwest, which accounted for about 1% of the area of the study region, specifically in Panzhihua and Liangshan. These findings demonstrated the effectiveness of the Logistic model in predicting forest fires in Sichuan Province, providing valuable insights regarding forest fire management and prevention efforts in this region.

Antecedent Hydrometeorological Conditions of Wildfire Occurrence in the Western U.S. in a Changing Climate

AbstractWildfires have significant hydrological and ecological impacts in the western U.S. Using a high‐resolution regional climate simulation and wildfire observations for 1984–2018, this study investigates the antecedent hydrometeorological conditions (AHCs) of wildfires in the western U.S. During the warm season (April‐September), the wildfire AHCs feature diverse surface pressure (PS), soil moisture, and longwave/shortwave radiation (LW/SW) conditions. K‐means clustering classifies wildfires into four types with distinct AHCs: low‐PS‐type and high‐PS‐type with lower and higher PS anomalies, respectively, LW‐type featuring intense LW but weak SW anomalies, and wet‐soil‐type with wet soil anomalies. Each fire cluster represents 22%–27% of all the wildfires, featuring different combinations of climate and vegetation conditions and their diverse relations to regional hydrometeorological conditions, with wet‐soil‐type fires often exhibiting opposite correlations with AHCs compared to those of the other three types. In five major Köppen climate zones over the western U.S., clustering‐based predictions improve the seasonal wildfire prediction accuracy (R2) by 10% compared to prediction without classification. Such improvement comes from separating the opposite relationships between wet‐soil‐type fires and their seasonal AHCs from the other three types, along with separating LW‐type fires, which include most of the lightning‐ignited fires that occur more randomly. Increases in wildfire occurrence during 1984–2018 are dominated by the increases in the LW‐type fires, while the wet‐soil‐type fires have decreased, consistent with the long‐term drying in the western U.S.

Performance of Fire Danger Indices and Their Utility in Predicting Future Wildfire Danger Over the Conterminous United States

AbstractPredicting current and future wildfire frequency and size is central to wildfire control and management. Multiple fire danger indices (FDIs) that incorporate weather and fuel conditions have been developed and utilized to support wildfire predictions and risk assessment. However, the scale‐dependent performance of individual FDIs remains poorly understood, which leads to large uncertainty in the estimated fire sizes under climate change. Here, we calculate four commonly used FDIs over the conterminous United States using high‐resolution (4 km) climate and fuel data sets for the 1984–2019 period. The relationships of these four FDIs to the observed wildfire sizes show that higher values of FDIs correlate to larger total fire sizes; this correlation is more robust at larger spatial scales. Sensitivity analysis indicates that the daily minimum relative humidity and precipitation are the most important drivers of the annual mean fire danger. In the instances of extreme fire danger, wind speed becomes a critical factor and should be considered in the calculation of the FDI. To assess the impact of climate change on future fire size, we calculate the present‐day and end‐of‐century FDIs using the 12 km regional climate model simulations. The four FDIs generally predict consistent changes in future fire potential, suggesting an overall higher fire potential in conjunction with a prolonged wildfire season in future climate. Regionally, the four FDIs also reveal similar seasonal patterns as the enhancement arises mostly in spring and summer over the southwest US while in summer and fall over the northern and eastern US.

Incorporating Stochastic Wind Vectors in Wildfire Spread Prediction

The stochastic nature of environmental factors that govern the behavior of fire, such as wind and fuel, exposes wildfire modeling to a degree of uncertainty. In order to produce more realistic wildfire predictions, it is, therefore, necessary to incorporate these uncertainties within wildfire models in a way that reflects the influence of environmental stochasticity on wildfire propagation. Otherwise, the risks of the potential danger of a given wildfire may be under-represented. Specifically, environmental stochasticity in the form of wind variability results in considerable uncertainty in the output of fire spread models. Here, we consider two stochastic wind models and their implementation in the spark fire simulator framework to capture the environmental uncertainty related to wind variability. The results are compared with the output from purely deterministic wildfire spread models and are discussed in the context of the potential ramifications for wildfire risk management.

A fire danger index assessment method for short-term pre-warning of wildfires: A case study of Xiangxi, China

Wildfire is a significant threat that can result in severe injuries and economic damage. Conducting wildfire danger assessments in specific areas is necessary to mitigate the negative consequences associated to them. A useful tool to achieve them is satellite remote sensing technology, which provides global observations of the earth surface and enables wildfire danger assessment, hotspot detection, and flame monitoring. Based on this, a fire danger index assessment method for short-term wildfires based on the Moderate-Resolution Imaging Spectroradiometer (MODIS) was proposed. The fire danger composite index (FDCI) could be obtained by using the Normalized Differential Vegetation Index (NDVI), Land Surface Temperature (LST), Temperature Vegetation Dryness Index (TVDI), and forest vegetation types (VT). A fire danger rating and a fire danger map of wildfires can be developed, which could identify high and low danger areas. Additionally, it can be found that NDVI in Xiangxi was at a relatively high level, reflecting good vegetation conditions during 2014–2018. The study also showed that NDVI, TVDI, LST, and VT were key parameters affecting wildfire danger. Low FDCI values indicated a low wildfire danger and vice versa. Furthermore, a case study in Xiangxi was conducted in September 2018. The application shows that there is consistency between the actual high danger areas and the calculated high danger areas by this method. The case study reflects that the method benefits the danger classification, prediction, and rescue of wildfires in specific areas to some extent.