Evaluation of Indices for Automatic Active Fire and Burn Scar Detection in Forest Zones Using Satellite Imagery in Chiang Mai, Thailand

Abstract

Wildfire detection and burn scar mapping in the mountainous areas of Northern Thailand remain challenging due to cloud cover, smoke, revisit intervals, and complex terrain, making index and sensor selection critical for reliable fire monitoring. This study aimed to evaluate and compare the performance of three remote sensing indices for classifying Active Fire and Burn Scar areas in Chiang Mai Province, Thailand, and to identify the most appropriate index and satellite platform for each burned area category. The tested indices included the Burned Area Index (BAI) derived from Landsat 8–9 and Sentinel-2 imagery, the Burned Area Index for Sentinel-2 (BAIS2), and the Normalized Burn Ratio Thermal (NBRT) derived from Landsat 8–9 imagery. Reference areas were prepared by visual interpretation of false-color composite images, supported by hotspot data and wildfire suppression coordinates. Optimal index ranges were determined using a statistical thresholding approach, and classification accuracy was assessed using error matrices. NBRT from Landsat 8–9 performed best for active fire detection, with an optimal range of -0.289–0.260, a thematic coverage of 70.54%, and an Overall Accuracy of 88.16%. NBRT also classified burn scars effectively, with an optimal range of 0.326–0.543, a thematic coverage of 77.14%, and an Overall Accuracy of 91.84%. The Sentinel-2 reflectance-based indices did not provide reliable active fire classification under the applied thresholding approach, possibly because they lack the thermal infrared data needed to capture fire-emitted heat. For burn scar mapping, BAIS2 showed the highest performance, with an optimal range of 0.860–1.031, a thematic coverage of 77.14%, and an Overall Accuracy of 96.58%. These findings demonstrate that an index effective for burn scar mapping is not necessarily suitable for active fire detection, underscoring the need to match index and sensor selection to the intended monitoring objective.