Annual area burned correlates with temperature and fuel aridity, yet extreme wildfire outcomes arise from a small fraction of fires and rapid-growth days. This asymmetry indicates that thermodynamic favorability sets background susceptibility but does not determine when extreme growth occurs. This Perspective proposes a cross-scale framework that distinguishes susceptibility from regime-conditioned event-scale realization. At seasonal and regional scales, temperature and humidity influence fuel dryness, ignition likelihood, and fire-season length, explaining substantial interannual variability in area burned. These variables vary smoothly in space and retain signal under aggregation. By contrast, extreme fire growth occurs during short-lived synoptic configurations that organize winds, pressure gradients, and stability into discrete opportunity windows that permit sustained spread. The strongest winds governing rapid spread are intermittent, terrain-structured, and often unresolved in coarse datasets or aggregated indices. Within these windows, terrain interactions, organized flow, and fire–atmosphere feedbacks can amplify growth until circulation patterns shift. Extreme wildfire behavior therefore operates as a gated joint-probability process requiring the coincidence of susceptibility (S), dynamical weather opportunity (W), and ignition (I). Separating susceptibility from realization reconciles strong climate–fire correlations with the dynamical control of event-scale extremes.

Wildfires have emerged as one of the most destructive natural disasters worldwide, causing catastrophic losses. These losses have underscored the urgent need to improve public knowledge and advance existing techniques in wildfire management. Recently, the use of Artificial Intelligence (AI) in wildfires, propelled by the integration of Unmanned Aerial Vehicles (UAVs) and deep learning models, has created an unprecedented momentum to implement and develop more effective wildfire management. Although existing survey papers have explored learning-based approaches in wildfire, drone use in disaster management, and wildfire risk assessment, a comprehensive review emphasizing the application of AI-enabled UAV systems and investigating the role of learning-based methods throughout the overall workflow of multi-stage wildfire management, including pre-fire (e.g., vision-based vegetation fuel measurement), active-fire (e.g., fire growth modeling), and post-fire tasks (e.g., evacuation planning) is notably lacking. This survey synthesizes and integrates state-of-the-science reviews and research at the nexus of wildfire observations and modeling, AI, and UAVs - topics at the forefront of advances in wildfire management, elucidating the role of AI in performing monitoring and actuation tasks from pre-fire, through the active-fire stage, to post-fire management. To this aim, we provide an extensive analysis of the existing remote sensing systems with a particular focus on the UAV advancements, device specifications, and sensor technologies relevant to wildfire management. We also examine the pre-fire and post-fire management approaches, including fuel monitoring, prevention strategies, as well as evacuation planning, damage assessment, and operation strategies. Additionally, we review and summarize a wide range of computer vision techniques in active-fire management, with an emphasis on Machine Learning (ML), Reinforcement Learning (RL), and Deep Learning (DL) algorithms for wildfire classification, segmentation, detection, and monitoring tasks. Ultimately, we underscore the substantial advancement in wildfire modeling through the integration of cutting-edge AI techniques and UAV-based data, providing novel insights and enhanced predictive capabilities to understand dynamic wildfire behavior.

Fire models predict fire behavior and effects. However, there is a need to know how confident users can be in forecasts. This work developed a probabilistic methodology based on ensemble simulations that incorporated uncertainty in weather, fuel loading, and model physics parameters. It provided information on the most likely forecast scenario, confidence levels, and potential outliers. It also introduced novel ways to communicate uncertainty in calculation and graphical representation and applied this to diverse wildfires using ensemble simulations of the CAWFE coupled weather–fire model ranging from 12 to 26 members. The ensembles captured many features but spread was narrower than expected, especially with varying weather and fuel inputs, suggesting errors may not be easily mitigated by improving input data. Varying physics parameters created a wider spread, including identifying an outlier, underscoring modeling knowledge gaps. Uncertainty was communicated using burn probability, spread rate, and heat flux, a fire intensity metric related to burn severity. Despite limited ensemble spread, maps of mean and standard deviation exposed event times and locations where fire behavior was more uncertain, requiring more management or observations. Interpretability was enhanced by replacing traditional hot–cold color palettes with ones that accommodate the vision-impaired and adhere to web accessibility standards.

Five of California's 10 largest wildfires occurred in 2020, with the largest complex exceeding the previous largest by more than 100%. The year follows a decade containing extraordinary fire activity. Previous trend investigations focused on changes in human activities and atmospheric thermodynamics, while the impacts of changing atmospheric dynamics are largely unknown. Here, we identify weather types (WTs) associated with historically large daily burned areas in eight Californian regions. These WTs characterize dominant fire weather regimes varying in fire behavior types (plume‐driven vs. wind‐driven fires) and seasonality. Most of the strongly large‐scale forced WTs such as Santa Ana and Diablo events increased in frequency during the 20th century particularly in the San Diego and Bay Area regions. These changes are likely not anthropogenically caused and the frequency of such events is projected to decrease under continuing climate change. However, significant future increases are found for WTs associated with thermal‐low‐pressure systems along the California coast and in the Sierra west region. These increases in southern California are mainly due to increasing greenhouse‐gas forcing and arise from the larger ocean‐land temperature gradient while aerosol forcing changes are driving most of the increased frequency in central and northern California due to a reduction of relative humidity over land and a strengthening of low‐pressure anomalies over the coast. These WT frequency changes could permit more weather favorable for large fire growth in summer and less in fall, further enhancing the risk of catastrophic fires due to hotter and drier summers in future climates.

Drone-based Unmanned Aerial Systems (UAS) provide an efficient means for early detection and monitoring of remote wildland fires due to their rapid deployment, low flight altitudes, high 3D maneuverability, and ever-expanding sensor capabilities. Recent sensor advancements have made side-by-side RGB/IR sensing feasible for UASs. The aggregation of optical and thermal images enables robust environmental observation, as the thermal feed provides information that would otherwise be obscured in a purely RGB setup, effectively "seeing through" thick smoke and tree occlusion. In this work, we present Fire detection and modeling: Aerial Multi-spectral image dataset (FLAME 2) [1], the first ever labeled collection of UAS-collected side-by-side RGB/IR aerial imagery of prescribed burns. Using FLAME 2, we then present two image-processing methodologies with Multi-modal Learning on our new dataset: (1) Deep Learning (DL)-based benchmarks for detecting fire and smoke frames with Transfer Learning and Feature Fusion. (2) an exemplary image-processing system cascaded in the DL-based classifier to perform fire localization. We show these two techniques achieve reasonable gains than either single-domain video inputs or training models from scratch in the fire detection task.

AbstractModeling smoke dispersion from wildland fires is a complex problem. Heat and emissions are released from a fire front as well as from post-frontal combustion, and both are continuously evolving in space and time, providing an emission source that is unlike the industrial sources for which most dispersion models were originally designed. Convective motions driven by the fire’s heat release strongly couple the fire to the atmosphere, influencing the development and dynamics of the smoke plume. This chapter examines how fire events are described in the smoke modeling process and explores new research tools that may offer potential improvements to these descriptions and can reduce uncertainty in smoke model inputs. Remote sensing will help transition these research tools to operations by providing a safe and reliable means of measuring the fire environment at the space and time scales relevant to fire behavior.

Abrupt changes in wind direction and speed can dramatically impact wildfire development and spread, endangering firefighters. A frequent cause of such wind shifts is outflow from thunderstorms and organised convective systems; thus, their identification and prediction present critical challenges for fire weather forecasters. Here, we develop a methodology and implement it in a software tool that can identify and depict convective outflow boundaries in high-resolution numerical weather prediction (NWP) models to provide guidance for fire weather forecasting. The tool can process model output, objectively identify gust fronts, and graphically display detected gust fronts and similar boundaries in NWP model forecasts. The tool is demonstrated with output from the Weather Research and Forecasting (WRF) model from the operational High-Resolution Rapid Refresh (HRRR) forecasting system and from a WRF ensemble run at the National Center for Atmospheric Research that can provide probabilistic information about model-predicted gust fronts. The tool can identify outflow boundaries in model forecasts of convective events occurring in both simple and complex terrain, both with and without concurrent wildfire activity. With accurate underlying model forecast output, the tool can reliably reveal areas of potential gust front activity and thus provide valuable guidance to incident meteorologists and command personnel.

The objectives of this study are to evaluate landscape-scale fuel and terrain controls on fire rate of spread (ROS) estimates derived from repetitive airborne thermal infrared (ATIR) imagery sequences collected during the 2017 Thomas and Detwiler extreme wildfire events in California. Environmental covariate data were derived from prefire National Agriculture Imagery Program (NAIP) orthoimagery and USGS digital elevation models (DEMs). Active fronts and spread vectors of the expanding fires were delineated from ATIR imagery. Then, statistical relationships between fire spread rates and landscape covariates were analyzed using bivariate and multivariate regression. Directional slope is found to be the most statistically significant covariate with ROS for the five fire imagery sequences that were analyzed and its relationship with ROS is best characterized as an exponential growth function (adj. R2 max = 0.548, min = 0.075). Imaged-derived fuel covariates alone are statistically weak predictors of ROS (adj. R2 max = 0.363, min = 0.002) but, when included in multivariate models, increased ROS predictability and variance explanation (+14%) compared to models with directional slope alone.

The understanding and prediction of large wildland fire events around the world is a growing interdisciplinary research area advanced rapidly by development and use of computational models. Recent models bidirectionally couple computational fluid dynamics models including weather prediction models with modules containing algorithms representing fire spread and heat release, simulating fire-atmosphere interactions across scales spanning three orders of magnitude. Integrated with weather data and airborne and satellite remote sensing data on wildland fuels and active fire detection, modern coupled weather-fire modeling systems are being used to solve current science problems. Compared to legacy tools, these dynamic computational modeling systems increase cost and complexity but have produced breakthrough insights notably into the mechanisms underlying extreme wildfire events such as fine-scale extreme winds associated with interruptions of the electricity grid and have been configured to forecast a fire's growth, expanding our ability to anticipate how they will unfold. We synthesize case studies of recent extreme events, expanding applications, and the challenges and limitations in our remote sensing systems, fire prediction tools, and meteorological models that add to wildfires' mystery and apparent unpredictability.

Fire is a powerful ecological and evolutionary force that regulates organismal traits, population sizes, species interactions, community composition, carbon and nutrient cycling and ecosystem function. It also presents a rapidly growing societal challenge, due to both increasingly destructive wildfires and fire exclusion in fire-dependent ecosystems. As an ecological process, fire integrates complex feedbacks among biological, social and geophysical processes, requiring coordination across several fields and scales of study. Here, we describe the diversity of ways in which fire operates as a fundamental ecological and evolutionary process on Earth. We explore research priorities in six categories of fire ecology: (a) characteristics of fire regimes, (b) changing fire regimes, (c) fire effects on above-ground ecology, (d) fire effects on below-ground ecology, (e) fire behaviour and (f) fire ecology modelling. We identify three emergent themes: the need to study fire across temporal scales, to assess the mechanisms underlying a variety of ecological feedbacks involving fire and to improve representation of fire in a range of modelling contexts. Synthesis: As fire regimes and our relationships with fire continue to change, prioritizing these research areas will facilitate understanding of the ecological causes and consequences of future fires and rethinking fire management alternatives.

An important property of wildfire behaviour is rate of spread (ROS). The objectives of this study are to evaluate the uncertainty of landscape-scale ROS estimates derived from repetitive airborne thermal infrared (ATIR) georeferenced imagery and the utility of such estimates for understanding fire behaviour and controls on spread rates. Time-sequential ATIR image data were collected for the Cedar, Detwiler, and Rey Fires, which burned in California during summers of 2016 and 2017. We analyse error, uncertainty, and precision of ROS estimates associated with co-location accuracy, delineation of active fire front positions, and generation of fire spread vectors. The major sources of uncertainty influencing accuracy of ROS estimates are co-registration accuracy of sequential image pairs and procedures for delineating active fire front locations and spread vectors between them; none of these were found to be substantial. Median ROS estimates are 11 m min(-1) for the Cedar Fire and 8 m min(-1) for the Detwiler Fire, both of which burned through mixed shrub and tree areas of the Sierra Nevada foothills and were estimated for downslope spread events. Of the three study fires, the fastest spread rates (average spread of 25 m min(-1) with maximum of 39 m min(-1)) are estimated for the Rey Fire, which burned on variable directional slopes through chaparral shrubland vegetation.

The level-set method is typically used to track and propagate the fire perimeter in wildland fire models. Herein, a high-order level-set method using fifth-order WENO scheme for the discretization of spatial derivatives and third-order explicit Runge-Kutta temporal integration is implemented within the Weather Research and Forecasting model wildland fire physics package, WRF-Fire. The algorithm includes solution of an additional partial differential equation for level-set reinitialization. The accuracy of the fire-front shape and rate of spread in uncoupled simulations is systematically analyzed. It is demonstrated that the common implementation used by level-set-based wildfire models yields to rate-of-spread errors in the range 10-35% for typical grid sizes (Delta =12.5-100 m) and considerably underestimates fire area. Moreover, the amplitude of fire-front gradients in the presence of explicitly resolved turbulence features is systematically underestimated. In contrast, the new WRF-Fire algorithm results in rate-of-spread errors that are lower than 1% and that become nearly grid independent. Also, the underestimation of fire area at the sharp transition between the fire front and the lateral flanks is found to be reduced by a factor of approximate to 7. A hybrid-order level-set method with locally reduced artificial viscosity is proposed, which substantially alleviates the computational cost associated with high-order discretizations while preserving accuracy. Simulations of the Last Chance wildfire demonstrate additional benefits of high-order accurate level-set algorithms when dealing with complex fuel heterogeneities, enabling propagation across narrow fuel gaps and more accurate fire backing over the lee side of no fuel clusters.

To protect ecosystem services and the increasing wildland urban interface in a world with fire, comprehensive maps of wildland fuels are needed to predict fire behavior and effects. Traditionally, fuels have been categorized into a classification scheme whereby a single metric represents vegetation composition and structure, which can then be parameterized based on variable vegetation amount and condition. Remote sensing has been used to extrapolate between known field plots across the landscape, however until recently, those technologies have had limited ability to characterize fuels (e.g., composition, horizontal and vertical connectivity). Using new technologies (imaging spectroscopy and LIDAR), the objectives of this study are to assess: 1) how fuel characteristics observed from remote sensing affect categorical fuel classifications, and 2) how fuel characteristics affect landscape-scale fire behavior (spread rate, areal extent and perimeter). The analysis was conducted over the 2014 California King Fire that burned ~40,000 ha over lands with varying use and history and has unique remote sensing observations from before and after the fire. This analysis compares fuel classifications from a synergistic field, model, and Landsat approach (LANDFIRE) and products derived from the Airborne Visible/Infrared Imaging Spectrometer and LIDAR (MapFUELS). Each classification focuses on different fuel characteristics, which were then used to compare differences in a fire simulation model (CAWFE) and actual fire behavior. The results show that fuel characteristic inputs such as horizontal connectivity or fuel type and vertical structure affect fire spread rate and final fire extent (respectively). These results present the opportunity for future integration of fuel characteristics observed at coarser resolutions (900 m2) into predictions of fire behavior a similar spatial resolutions (as opposed to the current standard based on empirical relationships between fuel and fire behavior at ~12 m2 resolution).

On 8–9 October 2017, fourteen wildfires developed rapidly during a strong Diablo wind event in northern California including the Tubbs Fire, which travelled over 19 km in 3.25 h. Here, we applied the CAWFE® coupled numerical weather prediction-fire modeling system to investigate the airflow regime and extreme wind peaks underlying the extreme fire behavior using simulations that refine from a 10 km to a 185 m horizontal grid spacing. We found that as Diablo winds travelled south down the Sacramento Valley and fanned out southwestward over the Wine Country, their strength waxed and waned and their direction wavered, creating varying locations near fire origins where wind overrunning topography reached 30–40 m/s, along with streaks and bursts of strong winds in the lee of some topographic features and stagnation downstream of others. Despite a statically stable layer in the lowest 1.5 km, the high Froude number flow sometimes resembled a hydraulic jump. Elsewhere, the flow behaved similarly to neutrally-stratified flow over small hills, creating wind extrema that exceeded 40 m/s at the crest of some lesser hills including near the Tubbs fire ignition, but which shed bursts of high speed winds that travel downstream at approximately 5–7-min intervals. Nonetheless, simulated fire growth lagged satellite detection of fire arrival in Santa Rosa by up to 1 h, although whether the data detect fire line or spotting is ambiguous. A forecast simulation with a 370 m horizontal grid spacing produced an on-time fire line arrival in Santa Rosa, with calculations executed 4 times faster than real time on a single computer processor.

A newer generation of models that interactively couple the atmosphere with fire behavior have shown an increased potential to understand and predict complex, rapidly changing fire behavior. This is possible if they capture intricate, time-varying microscale airflows in mountainous terrain and fire-atmosphere feedbacks. However, this benefit is counterbalanced by additional limitations and requirements, many arising from the atmospheric model upon which they are built. The degree to which their potential is realized depends on how coupled models are built, configured, and applied. Because these are freely available to users with widely ranging backgrounds, I present some limitations and requirements that must be understood and addressed to achieve meaningful fire behavior simulation results. These include how numerical weather prediction models are formulated for specific scales, their solution methods and numerical approximations, optimal model configurations for common scenarios, and how these factors impact reproduction of fire events and phenomena. I discuss methods used to adjust inadequate outcomes and advise on critical interpretation of fire modeling results, such as where errors from model limitations may be misinterpreted as natural unpredictability. I discuss impacts on other weather model-based applications that affect understanding of fire behavior and effects.