A land cover change framework analyzing wildfire-affected areas in bitemporal PRISMA hyperspectral images

Wildfires are becoming increasingly common events, and studying them, monitoring their effects, and assessing the damage they produce, is crucial for planning recovery efforts. The new generation of hyperspectral satellite sensors can provide highly detailed spectral information directly related to materials on the Earth’s surface, allowing the detection of potential changes in monitored areas. These instruments allow the detection of even small land changes, such as those in homogeneous areas of interest. Unlike binary change detection mechanisms that can only produce a map of changes in observed areas, our goal is to provide a mathematical framework to construct semantic maps of land change before and after an impactful event. This feature is particularly useful for monitoring land use and land cover (LULC), agriculture, and damage assessment in fire-affected areas. This paper presents a framework for remote sensing change analysis between bitemporal hyperspectral images, namely SemBLCC, whose core is a hierarchical clustering algorithm based on a rank-two nonnegative matrix factorization. SemBLCC is able to explicitly model the semantic “from-to” transitions between the two involved hyperspectral images, thanks to new spectral libraries specifically designed for the new data acquired by PRISMA (PRecursore IperSpettrale della Missione Applicativa) satellite. SemBLCC has been successfully used to produce LULC change maps of fire-affected areas, allowing accurate assessment of fire damage.