An operational Sentinel-2 methodology for burned-area mapping in heterogeneous landscapes

Reliable detection and mapping of burned areas are critical for managing the growing impact of fires on agricultural land and natural ecosystems. In southern and eastern Europe, fires cause economic losses and environmental damage through deforestation, habitat loss and carbon emissions, while disrupting ecosystem services, accelerating soil erosion and altering soil moisture and microclimate in ways that hinder recovery. Addressing these consequences requires effective, spatially explicit tools for monitoring, post-fire recovery, land-use planning, and climate-change adaptation. This study proposes a methodological approach for operational burned-area mapping based on Sentinel-2 data, designed for heterogeneous landscapes, where coarse-resolution products systematically under-report burns across mixed forest, shrub, grassland and agricultural cover. The core is a new composite multispectral index, NBR_BG, embedded in a three-component pixel rule combining pre-fire vegetation state, post-fire spectral signal and their signed difference. The rationale is that the classical Normalised Burn Ratio (NBR) and the Šebela SWIR-difference index have complementary limitations, and that adding pre-fire vegetation state as a third axis suppresses false positives typical of heterogeneous landscapes. The method was developed and validated on 96 candidate NBR_BG detections (94 confirmed as real fires) across two Sentinel-2 Military Grid Reference System (MGRS) tiles in Bulgaria, reaching an F1-score of 98.43% on the field-verified sample. Applied without recalibration to a nationwide Bulgarian assessment for August 2023, it mapped 639 fires, demonstrating scalability. The approach offers a practical, CPU-tractable tool aligned with Land Use, Land-Use Change and Forestry (LULUCF) reporting and land-use planning under Intergovernmental Panel on Climate Change (IPCC) guidance.