Towards consistent inland water body mapping across space and time from optical Earth observation systems

Inland water bodies, while covering less than 4% of the Earth surface, are essential to many global dynamic processes such as biogeochemical cycles, biodiversity, climate change and ecosystem services. Mapping the temporal distribution of terrestrial water is thus crucial for scientific research as well as for sustainable ecosystem management. The current deployment of an unprecedented Earth observation satellite constellation provides a unique opportunity to monitor quantitatively our changing environment. The tsunami of heterogeneous spatial datasets recently available required appropriate methods to extract relevant information. In this context, we aimed at mapping consistently inland water bodies with optical remote sensing by developing methods taking into account spatial and temporal resolutions independently of the environmental context and the observation sources. First, the thesis proposes a framework improving delineation of water bodies by handling sub-metric multi-source data acquired in heterogeneous observation conditions. Secondly, we assess the minimum size of water body mappable at sub-pixel level from 10-m sensors, with a specific interest for Sentinel-2 instrument potential. Thirdly, mapping water bodies using twice-daily 250-m MODIS observation was successfully demonstrated to produce maps and indicators describing the location, the intra-annual and the inter-annual behavior of all African inland water bodies. Finally, we address the challenge of water body map validation by proposing and applying an original validation strategy specific for a land cover class underrepresented at a global scale. Altogether, the frameworks and methods developed during this thesis contribute to consistently map inland water bodies.