Optical and radar remote sensing for burn scar mapping in the grassland biome

Abstract

Wildfires remain a major ongoing threat to the integrity of the environment and therefore emphasis is placed on employing efficient assessment techniques, such as remote sensing. Grassland fires received lesser attention compared to forest fires, despite their significant contribution to global wildfire occurrences. This study, conducted in South Africa, utilized Sentinel-1 radar and Sentinel-2 optical data to map burn scars in grasslands, in a biome representative of grasslands found elsewhere. Employing the Random Forest (RF) and Support Vector Machine (SVM) algorithms within the Google Earth Engine (GEE) platform to classify the data, the study achieved high producer’s and user’s accuracies in identifying burn scars using optical data (>90 %). Comparison of variable importance showed the infrared as well as vegetation and fuel moisture indices being the most influential variables to the classification. However, radar data produced lower accuracies (<50 %) owing to significant confusion in distinguishing grass, bare land and water bodies since these features have structural compositions similar to burnt areas. Nonetheless, radar data proved effective in differentiating burn scars from shadows. Combining optical and radar data yielded comparable accuracies to the optical-alone data but improved the discrimination between burnt areas and shadows. This discrimination capability also agrees with the importance of radar data that ranked better than the visible bands of the optical data. The benefit of merging optical and radar data underscores the importance of radar data, which remains unaffected by atmospheric interference like smoke, haze and clouds, enabling continuous monitoring even during fire events.