The recent development of CubeSat constellations (fleets consisting of numerous small satellites) provides an unprecedented opportunity for monitoring spatiotemporal dynamics of inland waters. The meter-scale spatial resolution and daily revisit frequency of the recent SuperDove constellation can potentially take the remote sensing of inland waters to the next level. However, there is a need to assess the radiometric quality of this new source of data for aquatic applications. This study examines the potential of SuperDove imagery in retrieving bathymetric and water constituent information in inland waters. We adopt both physics-based and machine-learning models to derive the biophysical parameters. The radiative transfer inversion model implemented in water color simulator (WASI) processor is parametrized to retrieve the concentration of chlorophyll-a and total suspended matter in Lake Trasimeno (Italy) using level-2 bottom-of-atmosphere SuperDove imagery. The flexibility of WASI in parametrization allowed us to effectively mitigate the atmospheric and sun-glint artifacts that are a severe problem. Furthermore, we leverage a neural network-based (NN) regression model to infer bathymetry of the Colorado River (US) with depths up to 10 m. The NN-based model automatically handles the feature extraction applied to level-1 top-of-atmosphere SuperDove imagery. The results indicate the promising radiometric quality of the SuperDove data that provided robust retrievals of bathymetric and constituent parameters in inland waters based on physical and machine learning models.

Primary Presenter: Milad Niroumand-Jadidi, Fondazione Bruno Kessler (mniroumand@fbk.eu)

Authors:

Milad Niroumand-Jadidi, Fondazione Bruno Kessler  (mniroumand@fbk.eu)

Francesca Bovolo, Fondazione Bruno Kessler  (bovolo@fbk.eu)