Combining optical and radar satellite image time series to map natural vegetation: savannas as an example

Up-to-date land cover maps are important for biodiversity monitoring as they are central to habitat and ecosystem distribution assessments. Satellite remote sensing is a key technology for generating these maps. Until recently, land cover mapping has been limited to static approaches, which have primarily led to the production of either global maps at coarse spatial resolutions or geographically restricted maps at high spatial resolutions. The recent availability of optical (Sentinel-2) and radar (Sentinel-1) satellite image time series (SITS) which provide access to high spatial and very high temporal resolutions, is a game changer, offering opportunities to map land cover using both temporal and spatial information. These data moreover open interesting perspectives for land cover mapping based on data combination approach. However, the usefulness of combining dense time series (more than 30 images per year) and data combination approaches to map natural vegetation has so far not been assessed. To address this gap, this contribution tests the idea that the combined consideration of optical and radar data combination and time series analyses can significantly improve natural vegetation mapping in the Pendjari National Park, a Sahelian savanna protected area in Benin. Results highlight that the combination of Sentinel-1 and Sentinel-2 SITS performs as well as Sentinel-2 SITS alone in terms of classification accuracy. Land cover maps are however qualitatively better when considering the data combination approach. Our results also clearly show that the use of dense/hypertemporal optical time series significantly improves classification outcomes compared to using multitemporal only a few images per year) or monotemporal data. Altogether, this work thus demonstrates the ability of dense SITS to improve discrimination of natural vegetation types using information on their phenology, leading to more detailed and more reliable maps for environmental management