Operational evaluation of data-driven forest fire forecasting models

Machine learning is being increasingly used to predict where active wildfires will burn the following day, helping inform evacuation boundaries and containment lines. Most models are evaluated using Average Precision (AP), which summarizes performance across all decision thresholds, although acting on a forecast requires choosing one. We benchmarked five discriminative architectures and one generative model on WildfireSpreadTS using a shared evaluation pipeline and two input configurations. We found that model rankings varied depending on whether performance was measured by AP or by threshold-dependent metrics like F1 and IoU. At the default 0.5 threshold, the highest-AP model flagged 4 to 5 times the observed active-fire area and ranked fifth of six on F1 and IoU, and the most recall-heavy model flagged 16 to 23 times. The three models whose flagged area stayed close to the observed area had AP scores 24 to 37% lower. Across models, we identified three distinct prediction profiles: over-predicting, balanced, and under-predicting, which AP alone could not distinguish. Expanding the input from 7 to 23 channels changed AP by 0.03 on average, against a 0.21 to 0.24 spread across models. Ultimately, these results show that AP alone can favor models whose predictions are poorly suited for operational wildfire forecasting.

Wildfire prediction is important for early warning and resource allocation, yet existing Earth foundation models (Earth FMs) are pretrained for general atmospheric and geophysical objectives rather than wildfire forecasting. To address this gap, we introduce WILDFIRE-FM, the first foundation model pretrained specifically for wildfire prediction using weather, active-fire observations, topography, vegetation, and static environmental data. However, introducing a domain-specific backbone alone does not solve the evaluation problem: wildfire events are sparse in space and time, making transfer conclusions highly sensitive to matching rules and evaluation settings. To address this problem, we introduce a fixed-contract evaluation framework with two controlled checks: a fixed-output check for matching-rule effects and a fixed-feature check for head-selection effects. Under matched contracts, we compare WILDFIRE-FM with ten Earth-FM baselines across occupancy, spread, retrieval, and regression tasks. Our results show that wildfire transfer conclusions depend strongly on evaluation design and task formulation. We hope this framework and WILDFIRE-FM provide a foundation for future wildfire-specific Earth-FM research and benchmarking. Our code is available at https://anonymous.4open.science/r/Wildfire-fm-evaluation-contracts-5AE9/.

Accurate next-day active fire forecasts can support early warning, disaster response, forest risk assessment, and downstream estimation of fire-related carbon emissions. Existing machine learning approaches to wildfire forecasting typically predict wildfire danger or fire probability on kilometre-scale daily grids, which is useful for regional warning but does not directly represent localized fire events. We propose Wildfire Ignition Set Predictor (WISP), a query-based model that reformulates next-day active fire forecasting as point-set prediction. From 48 hours of covariates including meteorology, satellite vegetation products, static land, and fire history, WISP predicts a fixed-size ranked set of future active fire cluster centres on a 375 m grid across globally distributed regions. The model is trained end-to-end with Hungarian matching; to address the conflicting roles of the classification score in assignment, ranking, and query activation, we use asymmetric classification-localization weighting in matching and loss. We further construct a globally distributed, hourly, multi-source benchmark for this task. On a held-out test set spanning fire regions worldwide, the best WISP variant achieves 38.2% average precision (AP) for ranked fire-centre detections, covers 53.4% of fire cluster mass weighted by fire radiative power (FRP), and localizes 54.1% of observed clusters within 5 km. These results establish sparse set prediction as a viable formulation for high-resolution wildfire forecasting and provide a benchmark for future work in this regime.