Monitoring active fires in Borneo from Sentinel-2 MSI images

A detailed, spatially explicit fire inventory is essential for improving our understanding of biomass burning and for supporting the formulation of targeted fire mitigation policies. However, such fire inventories remain limited, especially in tropical regions. Existing active fire (AF) products derived from low-resolution sensors (e.g. MODIS and VIIRS) generally have high omission errors (OE), especially when detecting small and relatively colder temperatures fires. While moderate-resolution sensors offer unprecedented opportunities for detecting small and subtle fires, they face the dilemma of high commission errors (CE). To address this problem, we propose an object-oriented method to effectively detect AFs from Sentinel-2 MSI images, which focuses on suppressing the interference of various CEs through object-level inter-spectral criteria cloud filtering, seamline exclusion based on granule footprints, and false positive refinement based on random forest classification model. Using more than 55,000 Sentinel-2 MSI images acquired during 2016–2021, we have compiled a novel 20 m fire inventory covering forests and peatlands in Borneo. Initial assessment of the fire inventory suggests a CE of approximately 7.2% and an OE of 11.5%. Analysis of the Borneo fire inventory revealed the following: (<i>i</i>) A significant concentration of AFs was observed in Kalimantan, with Central Kalimantan accounting for approximately 55.9% of all detected peatland fires in Borneo, and West Kalimantan contributing 33.7% of forest fires. (<i>ii</i>) Peatland fires dominated widespread fires in Borneo in 2019, with 1.4 to 2.6 times the size and 3.1 to 16 times the number compared to other years in 2016–2021. (<i>iii</i>) The MSI AF detections show slight differences in spatiotemporal patterns compared to MODIS and VIIRS AF products, which is attributed to variations in sensitivity to small fires. Our study clarifies the spatial dynamic distribution of AFs in Borneo, providing fundamental support for local fire monitoring, fire regime, and carbon emission research.