LAUSR

Abstract Habitat mapping is an essential descriptor to monitor and manage natural or semi-natural ecosystems. Habitats integrate both the environmental conditions and the related biodiversity. However, it remains challenging to map certain habitats such as inland wetlands due to spectral, spatial and temporal variability in the vegetation cover. Currently, no satellite constellations optimize the spectral, spatial and temporal resolutions required to map wetlands according to the habitats discriminated from in situ surveys. Our approach aims to combine satellite and unmanned aerial vehicle (UAV) data to exceed their respective limitations. Both data sources were combined through a spectral unmixing algorithm with the hypothesis that endmembers from UAV data are pure enough to enhance plant community abundances estimated from satellite data. The experiment was conducted on the regional preserve of the Sougeal marsh, a wet grassland of 174 ha located upstream of the Mont-Saint-Michel Bay. Two satellite data sources - Sentinel-2 and Pleiades - and three acquisition periods - November 2017, April 2018 and May 2018 - were considered. A reference map of plant community distribution was produced from UAV multitemporal data and floristic surveys to validate the unmixing of satellite data. This study shows innovative results and perspectives: while UAV can improve habitat discrimination, results vary among acquisition periods and habitats. Results illustrate well the great potential of combined UAV and satellite data but also demonstrate the influence of endmembers on the unmixing process and technical limitations (e.g. spectral mismatches between sensors), which can be overcome using domain adaptation.