Retrieval of Marine Parameters from Hyperspectral Satellite Data and Machine Learning Methods

The PRISMA hyperspectral mission of the Italian Space Agency, operational since 2019, is providing high spectral resolution data in the range 400–2500 nm, in support of multiple environmental applications, such as water quality and ecosystem monitoring. In this work we discuss how hyperspectral data can be used to simultaneously retrieve aerosol and marine properties, including sediment properties and chlorophyll, by using a coupled radiative transfer model (RTM). As physics-based methods are computationally expensive, we investigate the use of machine learning methods for emulation and hybrid retrievals, combining physics with machine learning. We find that assumptions on the covariance matrices strongly affect the retrieval convergence, which is poor in the coastal waters we considered. We also show that RTM emulation provides substantial speed-up and good results for AOD and sediment variables, however further parameter tuning seems necessary.