Mapping coastal ecosystems at very high temporal and high spectral resolutions using the maximum entropy classifier on 12-band Venμs and 8-band SuperDove imagery

Abstract. Coastal landscapes are composed of littoral ecosystems in interaction with human societies. The ecosystem services delivered range from oxygen and soil production, food provisioning, ocean-climate regulation, to recreo-tourism activities. However, the sustainability of those air-land-sea interfaces are increasingly jeopardized by global changes and local stresses due to the anthropogenic drivers. The natural protection offered by those ecosystems to cope with the near future’s sea-level rise and wind-wave intensification requires to be finely and precisely monitored. 
High to very high spatial, spectral and temporal spaceborne observation constitutes a relevant solution to address this issue. However the trade-off in remote sensing constrains to emphasize one resolution at the detriment of the other ones. This research innovatively proposes to compare the classification (maximum entropy) performance, over five representative coastal landscapes, of two state-of-the-science satellite sensors provided with very high spectral and temporal resolutions along with a high spatial resolution, namely the 8-band SuperDove at 3 m, and the 12-band Venμs at 4 m. 
Firstly, the addition of the intermediate visible and near-infrared bands to the standard blue-green-red dataset strongly improved the classification score of both SuperDove and Venμs orthorectified surface reflectance imageries: +8 and +7% for beach-dune, +18 and +4% for salt marsh, +2 and +1% for mangrove, +3 and +12% for seagrass, and +9 and +8% for coral, respectively. Secondly, the full optical SuperDove dataset outperformed the full optical Venμs dataset for salt marsh (3%), mangrove (3%) and coral reef (10%); but equalled it for the beach-dune (0%); and was outperformed for seagrass meadow (−4%). Both spatial and spectral resolutions (10 versus 12 bits for SuperDove and Venμs, respectively) were discussed to explain those findings.