Therefore, this session invites contributions providing new insights into wildfire behavior through satellite data and artificial intelligence. It includes any extended application for air quality or climate extremes related to wildfires. This session also welcomes case studies of large fire events. The expected topics for this session are listed, but not limited to that.
- Wildfire monitoring and forecasting
- Smoke and air quality modeling
- Carbon emission estimation
- Wildfire risk assessment
- Ecosystem recovery and rehabilitation
- Wildfire behavior analysis (e.g. fire spread)
- Climate change and wildfire trends
Climate change is driving an alarming increase in wildfire frequency across arid ecosystems, highlighting the urgent need for more accurate susceptibility mapping to inform prevention efforts. This research evaluates fire risk in the oasis region of Morocco's Middle Ziz Valley through an integrated approach combining GIS technology, satellite remote sensing, and machine learning methods.
A total of 130 fire incidents recorded between 2010 and 2023 were analyzed using NASA's FIRMS database. Nine key factors were considered: topographic variables (slope and aspect), environmental conditions (NDVI, precipitation, temperature, and wind speed), and human influences (land use, road proximity, and distance to residential areas).Four machine learning algorithms were evaluated: Random Forest, Logistic Regression, Support Vector Machine, and XGBoost. Variable importance was determined using Information Gain, while model interpretability was enhanced through SHAP analysis. Ecological health and urban development were further assessed using the Remote Sensing Ecological Index and Night-Time Lights Index, respectively. Integrating these vulnerability measures with fire susceptibility data enabled comprehensive risk mapping across the region.
Random Forest achieved the highest predictive accuracy among the evaluated models. Temperature, wind speed emerged as the primary drivers of fire susceptibility. This adaptable methodological framework provides a robust approach for wildfire risk assessment applicable to other arid ecosystems globally.
Keywords: Wildfire susceptibility, Machine learning, Oasis ecosystems, Vulnerability assessment, Remote sensing, Morocco, Risk mapping.
How to cite: Bouargalne, Y., Tanarhte, M., Stour, L., and Lafif, M.: Wildfire Risk Assessment in Arid Oasis Ecosystems: An Integrated Machine Learning and Vulnerability Analysis Approach in Morocco., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1544, https://doi.org/10.5194/egusphere-egu26-1544, 2026.
Live Fuel Moisture Content (LFMC) is a key variable for understanding wildfire ignition and propagation, particularly in forest ecosystems. In this study, we develop a daily LFMC product designed to support operational fire danger management services in France. The product is built from in situ measurements provided by the French National Forest Office and estimated using a lightweight yet expressive neural network architecture specifically designed to generalize across space and time. The model can be directly coupled with land surface models, enabling near-real-time monitoring of vegetation hydric stress at national scale.
Our framework integrates outputs from a physically based land surface model with satellite-derived leaf area index observations to produce spatially consistent, high-resolution estimates of land surface variables. Model robustness was assessed through complementary cross-validation strategies to evaluate interannual stability, spatial transferability, and an operational “deployment-like” scenario. In addition, a sensitivity analysis quantified the variability in predictions associated with training randomness and data sampling.
Results show strong accuracy across most regions of France, while revealing specific areas where model uncertainty remains high. These spatially explicit insights highlight where additional in situ sampling or improved process representation could meaningfully reduce epistemic uncertainty. Overall, this work demonstrates the potential of combining AI, process-based modeling and satellite observations to deliver operational LFMC products, ultimately supporting more informed wildfire risk assessment and fire management strategies.
How to cite: Baehr, Y. and Calvet, J.-C.: Using artificial intelligence to monitor live fuel moisture content across France, based on a high resolution land surface analysis, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3300, https://doi.org/10.5194/egusphere-egu26-3300, 2026.
Global Navigation Satellite System Reflectometry (GNSS-R) is a remote sensing technique that uses reflected GNSS signals from the Earth's surface to monitor geophysical parameters. This research explores an innovative approach that leverages GNSS-R satellite data from the Cyclone GNSS (CYGNSS) together with machine learning techniques, to predict the Fire Weather Index (FWI). Derived from meteorological data to estimate fire danger, this index is widely adopted in climate research, yet its relationship with GNSS-R observations remains untapped.
For this experiment, we assembled a three-year dataset of CYGNSS parameters collected over a specific region. This dataset is used to build and challenge different machine learning models ranging from classic methods like regression/classification, Support Vector Machines (SVM) or ensemble techniques (like Decision Trees, Random Forest or XGBoost) to deep learning models such as Artificial Neural Network (ANN) using Multilayer Perceptron (MLP).
The results reveal that incorporating Delay Doppler Map (DDM) related parameters into the training dataset significantly enhances the predictive accuracy across most of the evaluated models. Moreover, we present a MLP implementation in which parameters such as DDM center and peak are identified as strong contributors, approaching the importance of their spatiotemporal counterparts.
The analysis demonstrates that machine learning techniques in general and deep learning models in particular can successfully be used to infer the FWI with an acceptable level of accuracy for wildfire risk assessment, offering very promising new research lines based on modern AI advanced techniques like attention mechanisms or transformer architectures.
How to cite: Rodriguez, H., Doctor, M., and Cardellach, E.: CYGNSS-Based Machine Learning Approaches for Predicting Wildfire Risk, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3784, https://doi.org/10.5194/egusphere-egu26-3784, 2026.
Wildfires represent a growing global challenge, with increasing impacts on ecosystems, air quality, infrastructure, and human safety. In Canada, wildfire activity has intensified in both frequency and severity over recent decades, underscoring the need for robust spatial analyses to better understand the conditions under which fires escape initial suppression. Numerous studies have leveraged advances in artificial intelligence and machine learning to model wildfire occurrence and behavior. Many of these approaches rely on event–non-event (case–control) study designs, where fire locations are contrasted with non-fire locations to identify controlling environmental and anthropogenic factors. While fire event locations are generally well defined in historical records, selecting non-event (non-fire) locations remains a critical and often under-addressed challenge. Existing studies have employed a range of strategies to define non-fire points, including random sampling, uniform grids, distance-based buffers, environmental stratification, and background sampling. Poorly defined non-event locations can introduce substantial spatial bias, distort background conditions, and ultimately undermine model inference and interpretation. In wildfire applications, non-fire locations must satisfy multiple constraints: they should be accessible to fire occurrence, respect land–water and administrative boundaries, and reproduce the spatial structure of observed fire patterns without clustering too close to fire events or dispersing into ecologically irrelevant regions. To address this issue, we propose a two-stage methodological framework specifically designed for wildfire case–control studies, demonstrated using escaped wildfires in Quebec, Canada.
In the first stage, six background-pool (BP) generation strategies were developed to create large sets of geographically plausible non-fire candidates. These strategies progressively incorporate wildfire-relevant constraints, including minimum distance buffers around escaped fires, land–lake masking, grid-based spatial stratification, density weighting, and explicit enforcement of the Quebec boundary. The final background-pool version integrates all constraints and introduces a hybrid distance-based acceptance scheme that combines a strict exclusion zone near fires with a smooth distance-decay function beyond this threshold.
In the second stage, five control-set (CS) selection methods were evaluated to construct 1:1-matched fire–non-fire datasets across multiple fire-size thresholds. The final method balances regional representation and spatial clustering by using an adaptive grid and a composite distance metric that accounts for both proximity to individual fires and distance to local fire centroids. This approach explicitly matches the spatial “clumpiness” of escaped wildfires rather than simply maximizing separation between events and controls.
Model performance was assessed using distance-based diagnostics, spatial variance metrics, and point-pattern validation based on Ripley’s K-function. The proposed framework consistently produced non-fire patterns that are statistically indistinguishable from observed escaped wildfire patterns. Overall, this study provides a transparent, wildfire-specific template for selecting non-event locations, thereby supporting more reliable spatial inference in wildfire risk assessment and fire behavior modeling.
How to cite: Bonakdari, H., Zaji, A. H., Farhadian, G., and Gumiere, S. J.: From Background to Benchmark: A Framework for Preserving Spatial Structure in Wildfire Occurrence Modeling, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6821, https://doi.org/10.5194/egusphere-egu26-6821, 2026.
During the summer of 2025, Europe experienced a devastating wildfire season, with over 2,000 wildfires and more than one million hectares of land scorched. The scale of the crisis severely impacted natural ecosystems. In response, Europe leveraged a robust, multi-agency approach, anchored by satellite data from the European Space Agency (ESA) and Copernicus Contributing Missions (CCM) and real-time impact assessments provided by the Copernicus Emergency Management Service (CEMS) On-Demand Mapping.
This response was made possible by seamless collaboration between two critical teams: the Copernicus Rapid Response Desk (CCM-RRD) and the CEMS On-Demand Mapping Service. The CCM-RRD, established by ESA in 2024 at the request of the European Commission, operates as a central space data hub and serves as the primary interface for sourcing very high-resolution optical and radar satellite data from Copernicus Contributing Mission operators. In parallel, the CEMS On-Demand Mapping Service, staffed by remote-sensing experts, rapidly analyzes these data and disseminates maps and products directly to emergency responders.
This partnership fosters a dynamic ecosystem integrating space-based capabilities and on-the-ground expertise. The system relies on the availability of satellite data, governed by factors such as satellite overpass schedules, tasking constraints, ground station access, delivery timeliness, and the quality of derived information, which depends on advanced processing, validation, and user feedback mechanisms.
This presentation showcases the end-to-end rapid response achieved in wildfire monitoring and impact assessment analysis during the intense 2025 season. It highlights the robust infrastructure and expertise deployed to deliver high-quality products to civil protections and fire fighters across Europe and beyond.
These achievements are made possible through a pioneering, sustained collaboration between ESA’s and CCMs’ emergency management teams, and remote sensing experts. This collaboration demonstrates the transformative potential of European integrated Earth observation systems in mitigating natural disasters.
How to cite: Losco, P., Di Ciollo, C., Amans, V., Korzeniowska, K., Fischer, P., Manzo, C., Dalmasso, S., and Ceccato, P.: Timely wildfires characterization through EO data in response to emergencies: a case study., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8081, https://doi.org/10.5194/egusphere-egu26-8081, 2026.
Fuel characterization plays a central role in every phase of wildfire risk management, from identifying priority areas for prevention to supporting wildfire danger evaluation and simulating fire spread. Despite their importance, producing fuel maps that are both up to date and spatially extensive remains a persistent difficulty in fire science. Highly detailed information on fuel structure and composition would improve fire behavior simulations and danger assessment, but collecting such data is often difficult or impractical at the required level of detail. As a result, wildfire management must often rely on simplified representations that trade detail for feasibility. This raises a critical operational question: can fuel classification systems be designed to remain effective and reliable while being simple enough for large-scale, operational applications?
In response to this need, the CIMA Foundation has developed an operational fuel classification methodology tailored to civil protection requirements. The approach integrates land cover data, vegetation typologies, and environmental variables with expert-driven rules and machine learning–based wildfire susceptibility analyses. Rather than aiming for exhaustive fuel descriptions, the method focuses on capturing the most relevant characteristics for operational decision-making that is, the susceptibility of the territory to wildfire spreading.
Originally conceived for static fire susceptibility mapping at multiple spatial scales - ranging from regional to pan-European - the methodology has since been expanded to account for drought conditions. This enhancement allows fuel susceptibility to vary over time, producing dynamic maps that better represent seasonal changes in vegetation flammability. Such temporal variability is especially important in the context of climate change, where prolonged droughts combined with extreme weather can amplify wildfire severity. Addressing these compound drivers is a key requirement for operational wildfire forecasting in civil protection systems.
The resulting fuel maps serve as a core input for the RISICO wildfire danger forecasting model, developed by CIMA Foundation and used by the Italian Civil Protection Department, regional authorities, and international partners. Dynamic fuel representations have been tested in pre-operational settings at the regional level in Italy, as well as in international applications, demonstrating their usefulness in supporting wildfire danger bulletins. In parallel, the static fuel map has been employed as an input for the PROPAGATOR fire spread model, extending its applicability across different components of the wildfire risk management cycle.
Although intentionally less detailed than some advanced fuel classification schemes, this approach has proven fit for purpose in operational contexts. It offers a pragmatic compromise between scientific rigor and usability, enabling the effective integration of scientific knowledge into decision-support tools for wildfire management.
How to cite: Perello, N., Trucchia, A., Meschi, G., Ghasemiazma, F., D'Andrea, M., Fiorucci, P., Gollini, A., and Negro, D.: An Operational Framework for Dynamic ML-informed Fuel Mapping for Wildfire Risk Management , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8114, https://doi.org/10.5194/egusphere-egu26-8114, 2026.
Wildfires are a crucial component of the global ecosystems, exerting momentous impacts on climate, ecosystems, biodiversity, carbon storage, and human health. Despite the consensus that human activities are the primary driver of the global decline in burned area, the trends and underlying mechanisms across elevation remain poorly understood. We leverage multi-source remote sensing data to reconstruct a high-resolution (500 m) global burned area dataset, revealing distinct burned area trends across elevation gradients for different fire types. Over the period from 2002 to 2020, the global annual average burned area derived from the 500 m resolution dataset was estimated at 768.5 ± 51.8 Mha (Mean ± standard deviation). The global burned area exhibited a pronounced decline at an average rate of -7.8 ± 1.2 Mha yr-2 (p < 0.05). The burned area declines in low-elevation regions (0–600 m) is strikingly rapid with a rate of -5.9 ± 0.9 Mha yr-2 (p < 0.05), contributed by savanna (-2.1 ± 0.5 Mha yr-2, p < 0.05), grassland (-2.7 ± 0.4 Mha yr-2, p < 0.05), and cropland burned areas (-1.6 ± 0.3 Mha yr-2, p < 0.05). Although climate drivers inherently expand global burned areas, anthropogenic activities have exerted an overriding offsetting effect to reduce burned areas in low-elevation regions, which is most pronounced for savanna, grassland, and cropland fires. Conversely, at high altitudes, the impact of human activities tends to be attenuated, with meteorological conditions and fuel availability becoming dominant factors, resulting in a slower rate of burned area decline (-1.9 ± 0.6Mha yr-2, p < 0.05). Forest fires show a persistent, albeit nonsignificant, upward trend in burned area across both low-elevation and high-elevation, underscoring their mounting susceptibility to wildfire, which is driven primarily by warming-induced fuel desiccation and higher ignition probability. This “human-driven vs. climate-driven” dichotomy pattern underscores the complex interaction between anthropogenic and environmental drivers in shaping fire dynamics across elevation gradients. These findings reveal an elevation-dependent divergence in wildfire regimes, trends, and drivers that are reshaping the Earth’s fire landscape substantially, with profound implications for global biodiversity conservation and carbon cycle dynamics in a warming future.
How to cite: Wang, N. and Zheng, B.: Human-induced reduction in low-elevation burned area shapes the global declining trends, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8494, https://doi.org/10.5194/egusphere-egu26-8494, 2026.
Building resilience to high-impact wildfire episodes, particularly in a warming climate, requires a deeper understanding of how fire regimes are changing across spatial and temporal scales. This need is especially critical in regions where wildfires have recently emerged as a significant environmental and societal hazard, despite historically being considered low-risk. One such region is South Asia, where preparedness and response mechanisms remain far less developed compared with regions that have a long-established history of wildfire threats.
This study analyses recent changes in the spatiotemporal characteristics of wildfires across South Asia using satellite-derived data from the Moderate Resolution Imaging Spectroradiometer (MODIS) for the period 2001–2023. The analysis also examines the relative susceptibility of homogeneous ecoregions within South Asia and assesses the extent to which these environments are experiencing holistic changes in fire characteristics. Statistically significant positive trends are identified in both fire frequency and intensity across much of the study region, including vegetated areas of central India and central Pakistan. These regions, along with Nepal, exhibit notable increases in fire intensity, as measured by Fire Radiative Power, whereas decreases in intensity are observed in the most fire-prone parts of Bangladesh, north-east India, and Sri Lanka. Furthermore, the analysis explores previously unexamined changes in the intra-annual timing of fire occurrence. Results indicate a shift towards an earlier annual peak in fire incidence across many parts of India and Pakistan, while other areas show evidence of later fire activity, underscoring an additional layer of vulnerability for these countries.
This work provides new insights into the regional and local nuances of wildfire dynamics across a complex and, in the global context of wildfire danger, understudied region. The presence of significant trends in fire characteristics within ecoregions associated with tropical forests is particularly concerning. Our findings highlight the need for further investigation into the implications of these shifts for fire management, risk reduction and evidence-based decision-making in South Asia.
How to cite: Zahoor, Z., Blackett, M., Chen, Y.-F., Yildiz, A., and Eden, J.: Observed changes in spatiotemporal characteristics of wildfires in South Asia , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10236, https://doi.org/10.5194/egusphere-egu26-10236, 2026.
Aerial application of fire retardant is an essential tool for controlling wildland fires. Retardant drops are carefully planned to optimize fireline effectiveness, enhance firefighter safety, protect valuable resources and assets, and reduce environmental impact. However, factors such as topography, wind, vegetation structure, and aircraft orientation can create differences between the planned drop points and the actual area covered by the retardant. Accurate information on the exact placement and extent of deposited retardant can assist wildland fire managers in (1) evaluating how well the retardant slows or stops fire spread, (2) adaptively managing resources during the event, and (3) documenting placement in relation to ecologically sensitive areas. Specifically, precise footprint mapping improves drop placement assessment and supports more effective wildfire suppression and asset protection. This study employs UAV multispectral imagery and UAV LiDAR to test an automated method for detecting and mapping retardant footprints at very high spatial resolution. Drone data are processed using Agisoft Metashape software to generate georeferenced orthomosaic models, which are then used to develop predictors for classification. We apply supervised machine learning trained on labeled reference polygons to distinguish retardant deposits from surrounding land cover conditions (e.g., vegetation, bare soil, and burned surfaces) in Single-class and Multi-class machine learning tests. The resulting maps outline the full extent of retardant coverage and provide a detailed footprint rather than simplified linear drop traces. This approach enables a standardized, reproducible workflow to evaluate retardant placement and enhances documentation of drop locations relative to sensitive environments, while allowing for a more objective assessment of whether the drop contributed to slowing fire spread and protecting valued resources.
How to cite: Zayat, Z., Ducros, L., Roig, B., and Legendre, D.: Influence Of Aerial Wildfire Long-Term Fire-Retardant Drops on Environmental Transfer , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18503, https://doi.org/10.5194/egusphere-egu26-18503, 2026.
In wildfire burn severity assessment, change detection approaches based on satellite-derived burn severity indices (BSIs) have emerged as effective alternatives to traditional field-based methods such as the Composite Burn Index (CBI). However, previous studies have primarily focused on improving model performance using carefully curated datasets, while paying relatively limited attention to a fundamental limitation—the phenological consistency between pre- and post-fire imagery. This study proposes an automated burn severity analysis framework that incorporates phenology detrending to address this limitation. The proposed framework integrates hotspot-based automatic extraction of regions of interest, acquisition of valid pre- and post-fire imagery free from cloud contamination, and a vegetation phenology adjustment procedure to generate analysis-ready BSI datasets. By introducing the adjusted differenced BSI (adBSI) as a core component, the framework substantially increases the number of usable image pairs and enhances the stability and reliability of burn severity estimates. Validation against CBI plots and burn area data from the Monitoring Trends in Burn Severity (MTBS) program across the contiguous United States demonstrates that adBSI consistently achieves performance comparable to or better than conventional differenced BSI (dBSI). The improvement is particularly pronounced under phenologically mismatched pre- and post-fire conditions, especially when phenology-sensitive indices such as the normalized difference vegetation index (NDVI) are applied to vegetation types with strong seasonal variability, including deciduous forests. Time-series analyses further confirm that adBSI effectively suppresses seasonal fluctuations, yielding more stable and robust results than conventional dBSI. The developed framework was successfully applied to the 2025 wildfire events in the Los Angeles region, demonstrating its practical applicability. Overall, this study presents a simple yet powerful solution to a long-standing challenge in change detection–based burn severity analysis. Future work will focus on incorporating additional environmental variables and nonlinear modeling approaches to further enhance performance and extend the applicability of the proposed framework beyond wildfire burn severity analysis to a broader range of change detection applications.
How to cite: Sung, T., Yang, S., Kim, W., Kang, Y., Son, B., Lee, J., and Im, J.: An Automated Burn Severity Analysis Framework with Vegetation Phenology Adjustment, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16665, https://doi.org/10.5194/egusphere-egu26-16665, 2026.
Extreme rainfall-induced debris flows in post-wildfire watersheds across the western United States pose critical threats to downstream communities and infrastructure. An accurate and a prompt prediction of potential risk is vital for effective mitigation and emergency response. To address this, we present a comprehensive machine learning (ML) framework to enhance prediction of debris flow probabilities. Our methodology integrates remotely sensed soil moisture data alongside rainfall intensity, determining how antecedent wetness influences the rainfall threshold to trigger debris flows.
The ML model is trained and tested on several historical California wildfire events, using approximately 50 geomorphological, hydrological, geological, and other fire-related parameters extracted and processed from high-resolution digital elevation models, satellite-derived burn severity products, and hydro-meteorological reanalysis datasets. This multi-parameter integration achieves superior prediction accuracy by capturing the complex interaction between meteorological triggers and surface conditions. To facilitate operational deployment, we are developing an interactive web-based dashboard that enables real-time debris flow probability assessment.
The dashboard acts as a cyber infrastructure where users simply input fire perimeter boundaries and select key parameters options, such as burn severity. The framework then automatically retrieves the necessary environmental data including near-real-time soil moisture and precipitation inputs to generate probabilistic hazard mapping. Using our tool, emergency managers and stakeholders benefit from enhanced decision-support for post-wildfire risk assessment.
How to cite: Poudel, S., Pradhan, N. R., and Talchabhadel, R.: Interactive Web Dashboard for Post-Wildfire Debris Flow Risk Assessment through a validated ML model , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-735, https://doi.org/10.5194/egusphere-egu26-735, 2026.
Recent studies have shown an increase in fire damage risks in northern latitudinal forests related to climate warming (Maes et al., 2020; Venäläinen et al., 2020). These forest systems are complex, with many feedback loops, such as between different types of damage and forest structure parameters. Due to this complexity, the resilience to damage and therefore ability of forests to mitigate climate on a regional scale are still poorly understood. Understanding this complexity requires model work and extensive literature research, as most studies only focus on a few aspects of the forest system (Lagergren & Jonsson, 2017; Konôpka et al., 2016).
Within Fennoscandia, fire risk warnings are mainly based on the output of the Canadian Fire Weather Index (FWI). This index is purely weather based and does not differentiate between vegetation types. Other fire risk prediction models, such as the US National Fire Danger Rating System (NFDRS), are more comprehensive and might therefore lead to more accurate results. The aim of this study was therefore to test the FWI and NFDRS models in their ability to predict forest fire size in boreal Fennoscandia. Expectation was that the more comprehensive NFDRS, which includes vegetation-specific information, would outperform the purely weather-based FWI in predicting forest fire risk in Fennoscandia.
Output from the 3km resolution HARMONIE Climate (HCLIM3) model was used as input in both the FWI and NFDRS to model forest fire risk in boreal Fennoscandia over the years 2001-2018. The output from these models was then compared to burned area data from a variety of data sources in Fennoscandia (MODIS and EFFIS burned area products) and Norway (DBS fire statistics, Skogbrand forest insurance, NIBIO National Forest Inventory).
Our results show that both the FWI and NFDRS fail to capture forest fire intensity in boreal Fennocandia. Neither model shows any pattern that relates historical forest fire size to predicted fire risk. Additionally, the different burned area datasets disagree with each other both in terms of number, location and date of historical fires, as well as fire size. In this poster presentation we will discuss these results, as well as the methodology we used to reach this conclusion.
How to cite: Ribbers, E., Lee, H., Mooney, P., Muri, H., Nieradzik, L., Kim, J.-S., and Cai, L.: When increased complexity does not help model accuracy: both FWI and NFDRS fail to accurately predict forest fire risk in Fennoscandia, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3604, https://doi.org/10.5194/egusphere-egu26-3604, 2026.
Wildfires play a central role in the global carbon cycle, but the forest fire is intensifying globally. Controlling wildfires is crucial for managing carbon emissions, but the extinction processes remain poorly quantified. We analyzed global forest fire extinction drivers from 2001 to 2020 using satellite-derived firelines and machine-learning attribution at 500 m resolution. Totally, we identified 98,090 individual fire events worldwide and classified fireline pixels into fire and extinction states, then Random Forest models were used to model extinction procedure. The model showed a robust performance across regions (accuracy 0.72–0.85). Fire extinction mechanisms differ across biomes: in high-latitude forests, extinction is mainly controlled by climatic and fuel conditions, whereas in tropical regions fires more often terminate when constrained by terrain features such as rivers, roads, and topographic breaks. Temporal trends form 2001-2020 present a significant decreased trend of natural climate-driven extinction capacity, with reduced effectiveness of VPD, and a relative strengthening of terrain-related constraints in North America and central Asia (slope = -0.629–-0.318). While the effectiveness of fuel and terrain conditions intensified by time in North America and Asia, with a slope of 0.006–0.032. Especially for extreme fires, the extinction relied more on terrain barriers as climatic suppression fails. Our results imply that with the climate warming, high‑latitude forests require enhanced fire monitoring, while tropical and other fire‑prone regions must strengthen infrastructure and leverage terrain barriers, especially against extreme fires where natural climate‑driven suppression is weakening.
How to cite: Wang, N. and Li, W.: Drivers of fire extinction in global forests over 2001-2020, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4338, https://doi.org/10.5194/egusphere-egu26-4338, 2026.
The present work aims at determining the factors affecting forest regeneration after wildfire events and quantifying their impact using data from 45 major wildfires in Greece during 2017-2023. Wildfires in Greece have increased markedly during the last decade, in parallel with persistent drought conditions. We used the Normalized Difference Vegetation Index (NDVI) as an indicator of post-fire vegetation recovery and modelled NDVI using two approaches: a Generalized Linear Model (GLM) and an Artificial Neural Network (ANN). Predictors used in both models were soil moisture (SM) in four soil depths (0-7 cm, 7-28 cm, 28-100 cm, 100-289 cm) from ERA5-Land, Burn Severity estimated by the differenced Normalized Burn Ratio (dNBR) from MODIS Terra Surface Reflectance, Slope and Aspect of the topography, Land Cover type, and Time elapsed since the wildfire event occurred. Significant predictors in the GLM were top layer SM (SM1) and SM in the deepest soil layer (SM4), Slope, Aspect, Land Cover type, and Time, with SM4 showing the highest regression coefficient. The GLM achieved a mean squared error (MSE) of 0.007. For the ANN, we evaluated 63 candidate architectures using repeated 60/20/20 train/validation/test splits (10 repeats) and selected hyperparameters based on validation performance (10 random initializations per architecture). The best-performing ANN used 11 input neurons (after dummy encoding of categorical predictors) and two hidden layers with 12 and 6 neurons (12-6), achieving mean validation MSE of 0.00306 ± 0.00029 and mean test MSE of 0.00324 ± 0.00042 across repeats. Permutation feature importance (reference split, R=50) highlighted Slope, Aspect, Land Cover type and SM4 as the most influential predictors, confirming the key role of soil moisture—especially at deeper horizons—in the regeneration process of burned land. Our research reveals areas where natural regeneration is effective and policies can, therefore, prioritize passive regeneration while mandating for more intensive methods is areas affected by adverse forest regeneration conditions.
How to cite: Gemitzi, A. and Chaleplis, K.: Investigating the factors that affect forest regeneration after major wildfire events, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5163, https://doi.org/10.5194/egusphere-egu26-5163, 2026.
Biomass burning constitutes a significant source of global atmospheric pollution, profoundly impacting regional air quality and the global carbon cycle. However, current global characterizations of fire emissions rely predominantly on polar-orbiting satellite data, whose limited temporal resolution hinders the capture of rapid evolution and diurnal variations in fire emissions. To address this, we identifies global hourly fire incidents from 2015 to 2025 using active fire products from multiple geostationary satellites. By integrating ERA5 vegetation and meteorological data, we construct an Estimated Biomass Burned Index (EBBI), which enables a unified physical quantification of available fuel load across global vegetation zones. Subsequently, we evaluate the dynamic increments of multiple pollutants during fire events, quantify regional disparities in post-fire emissions, and decouple the nonlinear mechanisms by which meteorological dispersion conditions and fuel attributes drive surface pollutant concentrations. Our study effectively bridges the gap in global high-frequency fire emission monitoring, providing a critical scientific basis for understanding short-term pollutant transport mechanisms and improving emission inventories.
How to cite: Zhang, L., Yue, L., and Yuan, Q.: Global Spatiotemporal Patterns and Drivers of Fire-Driven Pollutant Emissions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15568, https://doi.org/10.5194/egusphere-egu26-15568, 2026.
When most people think of countries affected by wildfires, Scotland is usually at the bottom of the list. Despite these preconceptions however, wildfire activity in Scotland has increased in recent decades, driven by shifting climate patterns, evolving land‑use practices, and the growing frequency of extreme weather events. This relative lack of wildfire occurrences compared to other warmer European regions is one of the reasons why wildfire research in Scotland has historically been a low-priority area.
In 2025, I undertook a secondment with the National Centre for Atmospheric Science at the University of St Andrews, with my goal being to use machine learning techniques to gain a deeper understanding of wildfire activity in Scotland and provide a method for modelling and predicting wildfire occurrences. With only 3-months to tackle this project, I quickly realised that building an AI model from scratch wouldn’t be feasible.
Through collaboration with researchers who had designed a wildfire prediction model for Southern European regions, my project quickly shifted to adapting and tailoring this model for Scotland. This involved finding & integrating Scottish environmental data into the model, running it, then evaluating the results to assess its applicability. The results revealed that whilst atmospheric weather variables are usually the most important factor in wildfire occurrences, Scotland’s more temperate climate means that the weather holds much less significance compared to other countries. Instead, physical features like landcover type become a lot more impactful in the model, reflecting both the unique vegetation present in Scotland and the common land management practice - muirburning, which can intensify and spin out of control.
I tailored a machine‑learning framework for Scotland using atmospheric, land‑cover, topographical, and human‑activity datasets spanning a 15 year period to create an AI-ready dataset that provides a great launchpad for analysis with machine learning algorithms such as support vector machines and random forests. Developing these methods not only provides new insights into Scottish wildfires, but it also lays out a roadmap that someone looking to analyse wildfires in their local region could follow in the future.
Considering the challenge of interpretability and trustworthiness in ML and AI, I used SHAP values to quantify the contribution of each predictor to model outputs, which provides a unique insight into the AI ‘black box’. These values quantify the impact different features have on wildfire prediction and are also a mechanism for explainable AI, showcasing the reasoning and weights the model uses when identifying the strongest drivers of wildfire likelihood.
Using the trained model, I created a national‑scale wildfire risk map which displayed spatial patterns of wildfire susceptibility and demonstrated how integrated modelling outputs can support risk‑informed decision‑making for land managers, emergency response planners, and climate‑risk practitioners. The groundwork also provides the ability to predict short-term wildfire likelihood across Scotland in the short-term by inputting forecasted weather variables or outlining future trends and patterns by utilising longer-term climate projections. Highlighting my full process and the model used was vital to ensure transparency, reproducibility, and community reuse.
How to cite: Alexander, J., Colfescu, I., Georgina Marie Meuriot, O., and Soto Martin, J.: My secondment adapting a Southern European ML wildfire prediction model for Scotland, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22119, https://doi.org/10.5194/egusphere-egu26-22119, 2026.
Wildfires are posing ecological and social challenges in the United Kingdom [1]. Therefore, it is important to better understand this phenomenon and locate the wildfires to identify their contributing factors. However, the analysis of UK wildfires is mainly based on the European Forest Fire Information System (EFFIS) [2], which uses MODIS and VIIRS imagery [1]. Due to the moderate pixel resolution of these satellites, only fires greater than 30 hectares are reliably recorded. This is a limitation for the study of UK wildfires as 99% are smaller than 30 hectares [1]. Thanks to higher resolution sensors such as Sentinel-2 MSI and Landsat 8 OLI, it has now become possible to map smaller wildfires [3]. However, we have found no evidence that these sensors were used to locate smalla wildfires in the UK.
Recently, geospatial foundation models have made significant improvements in the processing of satellite imagery. More specifically, Prithvi-EO-2.0 [4] and TerraMind [5] have outperformed typical machine learning models in many applications.
With this work, we studied how the Prithvi-EO-2.0 and TerraMind geospatial foundation models generalized and performed for the detection of wildfires in the UK. First, we created a dataset made of Harmonized Landsat and Sentinel images matched to UK EFFIS wildfire polygons (1409 large wildfire polygons covering the UK), as well as wildfire polygons from the UK Dorset region (typically smaller wildfire polygons obtained from 1147 wildfire intervention records of the Dorset Fire Intervention service). Then, we compared the performance of Prithvi-EO-2.0 and TerraMind over this dataset, using different fine-tuning configurations to analyze their performance and generalization capabilities. These models were also compared with typical ML and rule-based wildfire detection methods in order to confirm the relevance of our models.
We demonstrated that the use of geospatial foundation models, once fine-tuned over UK wildfire data, allowed us to increase the detection of the wildfire from 0.58 MIoU (rule-based baseline models) and 0.73 MIoU (ML based baseline models) to 0.78 (Prithvi-EO-2.0) and 0.81 (TerraMind) MIoU. We have found that this increase in performance is especially important for the detection of smaller wildfires relevant to our study.
This work thus provides a novel approach to detect smaller wildfires in the UK and the rest of the world using geospatial foundation models, but also highlights the necessity to train the geospatial foundation models with diverse data to improve its generalizability.
[1] Belcher, C. M et al.: UK wildfires and their climate challenges. Expert Led report prepared for the third climate change risk assessment (2021).
[2] San-Miguel-Ayanz, J. et al.: Towards a coherent forest fire information system in Europe: the European Forest Fire Information System (EFFIS) (2002).
[3] Filipponi, F.: Exploitation of sentinel-2 time series to map burned areas at the national level: A case study on the 2017 italy wildfires. Remote Sensing, 11(6), 622 (2019).
[4] Jakubik, J. et al.: Foundation Models for Generalist Geospatial Artificial Intelligence. Preprint Available on arxiv:2310.18660 (2023).
[5] Jakubik, J. et al.: TerraMind: Large-Scale Generative Multimodality for Earth Observation. IEEE/CVF International Conference on Computer Vision (ICCV) (2025).
How to cite: Vandaele, R., Belcher, C., Williams, H., Pope, E., and Luo, C.: Detection of wildfire burn scars in the UK using geospatial foundation models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14183, https://doi.org/10.5194/egusphere-egu26-14183, 2026.
Biomass burning plays a fundamental role in shaping landscapes and global ecosystem dynamics, with far-reaching impacts on the carbon balance, biodiversity, atmospheric composition, climate, air quality, and human health. In South America, which accounts for approximately 15% of global biomass burning emissions, accurate and accessible emission estimates are essential for long-term monitoring and for delineating policies to support neutral carbon development. We introduce the South American Biomass Burning Inventory (SAMBBI), the first open-access, continuous biomass burning emission inventory for South America, based on the regional model Brazilian Biomass Burning Emissions Model with Fire Radiative Power (BEM_FRP). SAMBBI represents a major advancement in understanding biomass burning emission dynamics and patterns by providing continuous, regularly updated emission estimates from 2003 onwards. This inventory will include emission estimates for the following species released during biomass burning: carbon monoxide (CO), carbon dioxide (CO₂), methane (CH₄), and fine and coarse particulate matter (PM₂.₅ and PM₁₀). SAMBBI aims to achieve five key goals: (1) automating routines and processes to ensure continuous and standardised emission estimates; (2) ensuring the continuity and consistency of emission estimates in the post-MODIS era; (3) facilitating access to emission estimates for researchers, policymakers, and society; (4) predicting biomass burning emissions using artificial intelligence; and (5) quantifying the extent to which fire suppression in the Amazon improves air quality in the largest cities of Brazil, including the São Paulo Metropolitan Area with a population of over 20 million inhabitants. To achieve these goals, SAMBBI will (1) develop a pioneering approach to integrate data from multiple sensors, ensuring continuity in emission time series; (2) create a web platform for dashboard visualisation and seamless access to emission estimates across multiple spatial and temporal resolutions by scientists and stakeholders; (3) develop and train artificial intelligence models using environmental, climatic, and land-use predictors to forecast biomass burning emissions; and (4) conduct air quality simulations with and without Amazonian fire emissions using SAMBBI-driven inputs to quantify the urban pollution burden attributable to Amazonian fires and the potential gains from fire suppression. With these advancements, SAMBBI will constitute an innovative and accessible inventory for enhanced regional and global air pollution assessments, serving as a reference for environmental research and evidence-based policymaking.
How to cite: Mataveli, G., Sanchez, A., Pereira, G., Longo, K., R. Freitas, S., Amaral, C., de Oliveira, G., Anderson, L., Maure, L., Pinho, I., W. Jones, M., Artaxo, P., Sitch, S., and E. O. C. Aragão, L.: SAMBBI: A New Open-access Biomass Burning Inventory for South America, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10746, https://doi.org/10.5194/egusphere-egu26-10746, 2026.
Wildfires are among Europe’s most damaging natural hazards, with significant impacts on ecosystems, economies, and society. Assessing how wildfire risk may evolve under climate change remains challenging, as fire occurrence depends not only on meteorological conditions but also on topography, land use, and human activity. However, most future-oriented studies rely on traditional weather-based indices, such as the Fire Weather Index, which do not explicitly account for these additional drivers. Machine-learning (ML) approaches offer a powerful alternative by integrating multiple sources of information, yet their application to future wildfire risk under combined climate and land-use change scenarios remains limited.
In this study, we develop a data driven ML wildfire risk model for Southern Europe trained on historical data. Several ML algorithms are evaluated, with XGBoost (XGB) model having the best performance (AUC = 0.93; F1 = 0.83). Explainable AI techniques are used to interpret model behavior and identify the most influential predictors of wildfire risk. The trained model is then applied to future climate projections using a regional multi-member ensemble of the Canadian Regional Climate Model version 5 (CRCM5) covering the European CORDEX domain at a high spatial resolution (0.11°, 12 km). Wildfire risk is investigated under the Shared Socioeconomic Pathways SSP1-2.6 and SSP3-7.0. Simulations driven exclusively by greenhouse gas (GHG) forcing are compared with simulations that also incorporate land-use change (LUC). Future projections indicate an increase in wildfire risk by the end of the century (2081–2100), under the SSP3-7.0 scenario, with a stronger rise when including both LUC and GHG changes compared to the one including GHG alone. These findings show the important role of land-use change in shaping future wildfire risk and highlight the need of integrating socio-environmental drivers along with climate change in wildfire risk assessments.
How to cite: Soto Martin, J., Meuriot, O., and Drews, M.: Future wildfire risk in Southern Europe under changing land use and climate scenarios, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1985, https://doi.org/10.5194/egusphere-egu26-1985, 2026.
For a long time, wildfires in Belgium were not considered a major risk. However, climate change is causing more frequent and longer periods of droughts, and when combined with high population density, a considerable wildland urban interface, and limited expertise and awareness, Belgium is facing an emerging wildfire risk. Belgium has already experienced wildfires that were difficult to control, such as those in Baelen (2011) and Achouffe (2025). Simultaneous wildfire events have also occurred, such as in April 2020, when three wildfires in the provinces Antwerp and Limburg stretched the capacity of the emergency services.
Belgium currently lacks a standardized method for wildfire data collection. The only available database, compiled by our research group at Ghent University (Prof. J. Baetens), is based on digitized newspapers dating back to 1830 and intervention reports. However, this database is incomplete: for many events only the date and municipality are known, with no additional information on burnt area, fire perimeter or flame height, nor any environmental data such as landcover type or meteorological conditions.
To improve wildfire data collection in Belgium, we are developing a semi-automatic method to register wildfires using satellite imagery. A major challenge is the small size of most wildfires in Belgium, often limited to a few hectares, which makes existing satellite-based systems such as the EFFIS Current Situation Viewer unsuitable. Our approach starts from emergency phone calls, where wildfire related calls are identified using a specific incident code, providing a date and approximate location. A spatial buffer is applied to account for the fact that callers are not located directly at the fire site. This results in a list of potential wildfire events.
For each potential event, time series of Sentinel-1 and Sentinel-2 images are collected. Pre- and post-fire images are processed using a customized wildfire detection algorithm designed specifically for the Belgian landscape. Based on spectral indices (e.g., NDVI or NBR), backscatter differences and thermal anomalies, the algorithm distinguishes true wildfire events from false positives by analysing conditions before and after the reported incident.
The detection results are validated using field-based wildfire perimeter measurements, which we collected for the wildfire season of 2025, covering approximately 100 events identified from newspaper reports. Combined with the historical database from 1830, these data enable us to understand the wildfire dynamics in Belgium. Finally, based on the historic dataset, we developed the Belgian Wildfire Viewer, an interactive dashboard that allows users to explore wildfires events and increases public awareness of wildfire risk. This viewer not only shows information about the number of wildfires we had in Belgium but also provides derived information such as the landcover type and meteorological conditions.
How to cite: Coppens, F., Baetens, J., and Vancoillie, F.: Registering small-scale wildfires in Belgium using satellite data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5551, https://doi.org/10.5194/egusphere-egu26-5551, 2026.
Climate change is intensifying wildfire risks globally, yet the most devastating impacts are concentrated in underserved regions. While global wildfire forecasting systems are established, there is significant potential to enhance their effectiveness for these vulnerable areas. Currently, many vulnerable regions lack the precise, localized information necessary for effective fire preparedness.
In this study, we employ a novel AI-based model that predicts fire weather index with a lead time of up to 31 days. Our research aims to better understand the intersection of wildfire risks and social vulnerability. We found that our AI-driven approach significantly reduces prediction bias compared to traditional methods derived from the ECMWF. This improvement is most pronounced in the Global South, where the convergence of high poverty and intense wildfire activity makes accurate forecasting essential.
By providing more reliable and actionable data to these underserved regions, our research demonstrates that AI can be a powerful tool for information equity. This study represents a critical step toward ensuring that all nations have access to high-quality tools to manage the escalating risks of climate change.
How to cite: Kang, Y., Lee, S., Cho, D., and Im, J.: Towards Equitable Wildfire Forecasting for Vulnerable Communities, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6259, https://doi.org/10.5194/egusphere-egu26-6259, 2026.
Underground coal fires (UCFs) represent a persistent global hazard, causing resource loss, land subsidence, toxic emissions, ecological degradation, and severe threats to mining safety. Unlike surface wildfires, the concealed and protracted nature of UCFs makes accurate risk assessment and dynamic monitoring exceptionally challenging. This study presented a robust remote sensing framework to characterise the spatiotemporal evolution of UCFs and assesed the effectiveness of suppression efforts in the Midong coalfield, Xinjiang, China, utilising time-series Landsat-8 Thermal Infrared (TIRS) imagery from 2013 to 2020.
For the coal fires identification and delineation, land surface temperature (LST) was retrieved from TIRS data using the Radiative Transfer Equation model. The retrieved LST effectively distinguishes fire areas from their surroundings, with significantly higher temperatures observed—up to 7.4°C higher in summer and 5.8°C in cold seasons. A comparative analysis of four thermal thresholding algorithms (Mean+2SD, Hotspot analysis, EDA, and Fractal model) was conducted. Due to the strong spatial dependence of UCF distribution, the Hotspot Analysis (HSA) model was identified as optimal for delineating fire boundaries, achieving a 65% area overlap accuracy and 70% location precision for fire spot identification. To further mitigate false alarms caused by solar radiation and surface heterogeneity, a Hotspot Sequential Frequency Extraction (HSFE) method was developed. This technique filters transient noise by identifying pixels with a high recurrence frequency (>75%) as high-probability fire risks.
Regarding the spatiotemporal analysis of coal fire evolution, the thermal severity and distribution assessed by the Coal-fire Thermal-island Intensity Ratio (CTIR) remain consistent with UCF development. The analysis captures the initially rapid fire growth, marked by a CTIR increase of 0.024 a-1 and a total areal expansion rate of 1.29×105 m2·a-1. However, the application of this risk evaluation successfully quantified the effectiveness of fire interventions: following suppression measures, the CTIR shifted to a decrease of 0.005–0.006 a-1. Similarly, Sequence Overlap Dynamic Analysis (SODA) reveals significant reductions of up to 74% in specific sections. Furthermore, the Thermal Anomaly Density Centre (TADC) concept was introduced to track migration, revealing that fire centroid movement is not simply unidirectional expansion but exhibits multi-directional, bilateral, and round-trip propagation.
This research demonstrates that integrating advanced spatiotemporal algorithms with satellite thermal data can effectively reconstruct the coal fire life cycle. The study also elucidates the complex coupling mechanism of anthropogenic and natural factors on UCF evolution, specifically characterizing their joint impacts on heat release, spatial distribution, and migration trajectories. The firedynamic behaviours reveal a strong "zoning effect", where thermal anomalies cluster along geological stratigraphic strikes and fracture zones. While geological fractures and faults fundamentally dictate fire initiation and propagation, frequent mining activities act as primary catalysts accelerating spread. Conversely, the implementation of targeted fire control and mining restrictions leads to the rapid disintegration and decline of large-scale fire zones. Ultimately, this framework not only offers critical data support for mining fire detection and spontaneous combustion safety management, but also demonstrates scalable, broad applicability for monitoring peat fires and other smoldering wildfires, providing generalized solution for integrated environmental management and dynamic fire risk mitigation.
How to cite: Deng, J., Xue, Y., Shi, B., Torero Cullen, J. L., and Han, L.: Identification, Spatiotemporal Evolution, and Risk Assessment of Underground Coal Fires Based on Time-Series Satellite Thermal Anomalies: A Case Study in Midong, China, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3169, https://doi.org/10.5194/egusphere-egu26-3169, 2026.
Climate change is amplifying wildfire risk in many regions worldwide due to a variety of factors, such as the increasing frequency and intensity of heatwaves and droughts. To support mitigation strategies, accurate and timely prediction of wildfire susceptibility is essential.
We present a wildfire susceptibility prediction model based on the Extreme Gradient Boosting (XGBoost) algorithm, designed to generate daily regional-scale susceptibility maps throughout the wildfire season. The model is implemented over Catalonia and trained using a diverse set of inputs, including population density, distance to the electric network, terrain elevation, the Normalized Difference Vegetation Index (NDVI), land cover classifications, and historical Fire Weather Index (FWI) and burned-area records. The training dataset covers the period 2007–2022 and includes 65 documented wildfire events, three of which correspond to large-scale fires affecting extensive areas, while the remaining events were of smaller magnitude. Model training focuses on the fire season (April–September), and performance is evaluated through external validation using data from 2023–2024. To ensure robust and generalizable predictions, we applied an extensive hyperparameter optimization procedure combined with a 5‑fold cross‑validation strategy, enabling the development of an optimized model and the creation of a consistent historical fire susceptibility dataset.
Evaluation of the model predictions for the 2020–2024 period using the quadratic weighted Kappa metric shows moderate to strong agreement with the official fire danger maps produced by the regional forest fire prevention service across most of Catalonia. Reduced skill is observed in southern Lleida and in high‑elevation sectors of the northern Pyrenees, where additional analysis will be required to better understand the sources of these regional discrepancies and guide future model improvements. Importantly, the developed model consistently outperforms fire‑danger assessments based solely on the Fire Weather Index. For a comparable recall level (0.6), it achieves twice the precision, demonstrating substantially higher predictive skill in identifying areas at risk of ignition.
This model is currently under development within the MedEWSa project, funded by the EU Horizon Europe Programme (grant agreement No 101121192) and represents a step toward operational tools for wildfire risk management and climate adaptation in Mediterranean environments.
Keywords: wildfire susceptibility, machine learning, XGBoost, fire danger prediction
How to cite: Martin-Gomez, V., Von Ruette, J., Chiva, B., Senande-Rivera, M., Pinilla, M., Burgues, J., and Baladima, F.: Machine Learning-Based Wildfire Susceptibility Modeling for Catalonia , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18293, https://doi.org/10.5194/egusphere-egu26-18293, 2026.
