Oral Session 4

OC-SMART for PACE: Retrieval of Coastal and Inland Water Properties based on Scientific Machine Learning Methods and Comprehensive Radiative Transfer Simulations

The OC-SMART platform has been extended for application to the PACE mission. OC-SMART is based on Scientific Machine Learning (SciML) methods to provide high quality estimation of water inherent optical properties (IOPs), for multi-spectral as well as hyperspectral ocean color sensors. OC-SMART for PACE makes use of some 200 OCI bands and employs independent SciML algorithms for (i) atmospheric correction (AC), (ii) cloud screening, and (iii) retrieval of water IOPs. The algorithms are trained with synthetic datasets generated by a robust coupled atmosphere-water forward radiative transfer model. The validation is based on Ocean Biology Processing Group (OBPG) datasets, AERONET-OC data, bio-optical measurements including hyperspectral remote sensing reflectances collected in coastal and Arctic environments. The SciML framework is particularly adequate for obtaining accurate retrieval of aquatic biological and ecological products from hyperspectral observations such as the PACE mission. The SciML approach used in OC-SMART for PACE is: (i) designed to work globally and to provide a smooth transition from the clear open ocean into turbid coastal waters, thus expanding the utility of climate-quality ocean color products into coastal areas and inland waters around the globe; (ii) extending the OC-SMART’s capabilities to provide reliable results at low solar elevations, required for accurate retrievals at high latitudes as well as for instruments deployed on geo-stationary platforms (GLIMR and GOCI). Finally, we emphasize that 1) SciML is particularly useful for turbid coastal waters where present algorithms for both AC and IOP retrievals often fail and 2) SciML provides uncertainties for the corresponding retrievals.

Knut Stamnes, Stevens Institute of Technology, USA, https://orcid.org/0000-0002-8880-6070

Wei Li, Stevens Institute of Technology, USA

Yongzhen Fan, University of Maryland, USA

Nan Chen, Stevens Institute of Technology, USA

Maria Tzortziou, City University of New York, USA

Atsushi Matsuoka, University of New Hampshire, USA

Code of Conduct
Privacy Policy