FireCluster: A multi-source satellite active fire data clustering algorithm for comprehensive wildfire event identification

Spatiotemporal dynamic data characterizing the lifecycle of wildfires constitute a critical element for fire monitoring and early warning systems. However, comprehensive data documenting full fire evolution trajectories across large geographical areas remain scarce. To address the limitations of current multi-source satellite fusion approaches in handling spatiotemporal resolution disparities, this study proposes FireCluster, an innovative decision-level fusion algorithm. This two-stage hierarchical density clustering approach integrates multi-source active fire products (MODIS, VIIRS, Landsat-8, and Himawari-8) and introduces an adaptive clustering mechanism that incorporates fire data source and temporal features, building upon the conventional ST-DBSCAN to effectively mitigate fire omission issues caused by inappropriate neighborhood radius selection. Empirical analysis using 2020 wildfire data from mainland China demonstrates that the proposed algorithm significantly improves annual fire identification recall from 15.38% to 44.12%, with particularly notable gains during peak fire seasons (March–April), achieving detection rates of 54.27%–63.25%. Validation through the case study of the Sichuan Liangshan wildfire further highlights its effectiveness, as FireCluster successfully captured multiple spatiotemporally discrete fire clusters within the event that traditional ST-DBSCAN incorrectly merged into a single fire, thereby enabling precise reconstruction of wildfire evolution trends. The proposed technical framework offers an advanced multi-source fusion solution for satellite-based fire monitoring, providing high-precision fire event characterization that supports disaster assessment and emergency decision-making.