Application of Level Set Methods for burned area mapping and evaluation against DLR's TET-1 hotspot data - a case study in Portugal

Abstract

Automated extraction of fire-affected areas from satellite images is a crucial task for large scale, near real-time damage assessment. The discrimination between burned and unburned pixels is usually done by differentiating pre-/post scenes and categorizing the change rate, or by using empirically derived thresholds on a single scene. These approaches, however, require the setting of a threshold value for the discrimination, which has to be derived empirically for constricted region of interest. It is therefore not well adoptable to different regions of interest, turning this approach inappropriate for automatic extraction of burned areas in large, heterogeneous study areas. The aim of this paper is to test the Active Contour Level Set method, which does not rely on any thresholds, regarding geometric accuracy of burned area extraction.