Assessment of space-based tropical smouldering peatlands: mixed pixel analysis

Abstract

Detection of tropical smouldering peatlands from space remains a challenge in a mixed pixel issue. This study aims to discriminate the mixed pixel containing smouldering fire resulting from Tropical Peatland Combustion Algorithm (ToPeCAl) applied to Landsat-8. The assessment of the smouldering pixel was conducted by applying a modification of the contextual test to select the smouldering pixel and then proceed it to linear spectral mixture analysis. The smouldering and burnt areas used for endmembers were determined from field observation. Our study areas span from South to West Kalimantan’s peatlands covering six path-rows of Landsat-8 in the fire seasons of 2018-2019. There were 53,224 candidate smouldering pixels from ToPeCAl which 80% of them passed the contextual test. Based on mixed pixel analysis, most of the selected smouldering pixels consisted of more than 0.79 of smouldering fraction. This was confirmed by the smoke features on the ground from the high spatial PlanetScope images acquired on the same day with Landsat-8. The actual false alarms pixel may consist of 0.74 of smouldering fraction and more than 0.20 of the burnt area or vegetated area fraction with no smoke features on the ground. Further research on smoke mapping or aerosol properties over burning peatland could enhance the performance of smouldering pixel detection.