ABSTRACTS

Oral Session 2 | Monday, October 3, 14:40–15:00 | Abstract 588

A Hyperspectral Inversion Framework for Estimating Inherent Optical Properties and Biogeochemical Parameters in Inland and Coastal Waters

The complex task of simultaneously retrieving multiple inherent optical properties (IOPs) and biogeochemical variables across a wide range of optically distinct waterbodies is complicated by uncertainties in hyperspectral satellite imagery. Our machine-learning model, Mixture Density Networks (MDNs), simultaneously estimates both IOPs and biogeochemical variables by leveraging the covariance between these products to achieve increased performance relative to operational algorithms. Our MDNs are trained on a large (N>7,000), globally distributed, set of co-aligned hyperspectral remote sensing reflectance (Rrs) measurements and associated products. Specifically, these products consist of absorption due to phytoplankton, non-algal particles, and colored dissolved organic matter as well as concentrations of chlorophyll-a, total suspended sediment, and phycocyanin. The accuracy of the model on the in situ dataset, when trained on half the dataset and validated on the other half, has increased performance relative to operational algorithms. Sensitivity of our model to uncertainties in the Rrs typical of satellite imagery, which result from instrument noise and uncertainties in the atmospheric correction process, is assessed using in situ matchups coaligned with hyperspectral imagery from HICO and PRISMA. The model is demonstrated on HICO and PRISMA imagery in preparation for future hyperspectral missions, such as the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE); it will serve as a key component for phytoplankton pigment composition quantification and phytoplankton functional type identification from remotely sensed hyperspectral satellite imagery.

Ryan O’Shea, Science Systems and Applications, Inc., and NASA Goddard Space Flight Center

Nima Pahlevan, Science Systems and Applications, Inc., and NASA Goddard Space Flight Center

Brandon Smith, Science Systems and Applications, Inc., and NASA Goddard Space Flight Center

Emmanuel Boss, School of Marine Sciences, University of Maine

Raphael Kudela, University of California Santa Cruz