BAM: A physics-informed self-supervised framework for near-real-time wildfire burned area mapping from multi-source earth observation

Abstract

The operational monitoring of global wildfires is frequently constrained by the delayed availability of reliable reference labels, the latency of standard surface reflectance products, and spectral confusion between charred biomass and naturally dark surfaces. These limitations are especially problematic for near-real-time wildfire monitoring, where burned-area maps must be produced before manual training data or fully processed reference products are available. To address these limitations, this study presents the Burn Area Mapper (BAM), a fully automated, physics-informed machine learning framework implemented in Google Earth Engine (GEE). The framework utilizes a Physics-Informed Self-Supervision paradigm that eliminates the need for manual annotation by leveraging the stoichiometry of combustion to generate scene-specific training labels. Central to this approach is the novel Automated Temporal Burn Index (ATBI), which utilizes multiplicative scaling of NIR and SWIR bands to enforce a strict bimodal distribution, enabling an adaptive Otsu algorithm to automatically isolate high confidence burn signals. These physics-derived labels are refined by a Gradient Tree Boost (GTB) classifier incorporating textural and topographic features. We validated the framework using internal spatially blocked assessment across 15 wildfire events spanning 6 continents, achieving a global mean F1-score of 0.994 without regional retraining. Comparative benchmarking against the MODIS MCD64A1 coarse-resolution product revealed that BAM captures up to 93% more burned area in fragmented landscapes. Furthermore, for one test event, validation against 3 m PlanetScope imagery at 30 m aggregated scale confirms (R2 = 0.80) that BAM’s probabilistic output effectively proxies the sub-pixel burn fraction, resolving fine-scale fire fronts and unburned refugia that are invisible to coarse-resolution systems. By combining rapidly available Landsat 8/9 TOA observations, SREM atmospheric correction, automated pseudo-label generation, and GTB refinement within GEE, BAM provides a scalable open-source framework for near-real-time burned-area mapping and post-fire assessment.