People

Fresh and coastal waters provide critical ecosystem services. Specifically, these waters serve as sources for food and drinking water as well as spaces for recreation. We can monitor the health of these systems via the ocean’s color, which in these optically complex waters is primarily derived from interactions between light and a combination of phytoplankton, suspended sediment, and colored dissolved organic matter (CDOM). The Ocean Color Instrument (OCI) onboard the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) provides hyperspectral data with improved radiometric capabilities critical for developing advanced products to monitor these optically complex waters on global scales. Prior to mission launch, we developed tools that leverage a machine learning architecture, termed mixture density networks (MDNs), to simultaneously estimate biogeochemical parameters (BPs) and inherent optical properties (IOPs), and the associated estimation uncertainty in each product. Specifically, the BPs include chlorophyll a, phycocyanin, and total suspended solids, which serve as proxies for phytoplankton biomass, freshwater cyanobacteria biomass, and water clarity. The IOPs include absorption due to phytoplankton pigments, non-algal particles, and CDOM, which can provide information on phytoplankton community composition (PCC), non-algal particle mineral versus organic composition, and the source and composition of dissolved organic matter.

To advance these products for global fresh and coastal water monitoring, we will first attempt to reduce uncertainties in the ocean color retrievals (remote sensing reflectance, Rrs) from these optically complex systems, by developing an atmospheric correction model specific to inland and coastal waters. Second, we will improve the generalization capabilities of the simultaneous BP and IOP MDN via inclusion of a bio-optical model during model training. Third, we will combine the uncertainty estimates for our Rrs retrievals with the downstream product uncertainty estimates to capture an overall pipeline uncertainty. Fourth, we will generate PCC product maps with both supervised and unsupervised approaches. Each of these products will be validated against co-located in situ measurements. The refined and developed Rrs, BP, IOP, and PCC products, and their associated uncertainties, will expand the utility of PACE for global fresh and coastal water monitoring.