Integration of satellite and in situ observations into machine learning for coastal water quality

Nearshore processes that impact water quality for boating, swimming and aquaculture happen at scales where current satellite data lack spatial or spectral resolution. Upcoming government and commercial satellite missions aim to fill this gap. To prepare to maximize the use of these Earth observations and address this challenge, we have been working closely with stakeholders around the Chesapeake Bay to explore satellite-derived indicators that could assist practitioners, i.e. clarity, harmful algal blooms, bacterial indices. Several key assets have recently been deployed to improve the potential for this effort: namely, a new NASA Aerosol Robotic Network for Ocean Color (AERONET-OC) site in the Chesapeake Bay to support ocean color atmospheric correction and validation, hyperspectral satellite data from DESIS and PRISMA, and in situ sampling for satellite calibration and validation. In coordination with these activities, we developed an artificial intelligence (AI) framework for feature discrimination using a single satellite sensor, starting with Sentinel 3a&b OLCI. We are currently extending our initial methodology to integrate multiple satellite data sets of differing spatial, spectral, and temporal resolution, namely MODIS-Aqua with its long record, and DESIS and PRISMA for their hyperspectral and higher spatial resolution. With this Deep learning for Environmental and Ecological Prediction-eValuation and Insight with Ensembles of Water quality (DEEP-VIEW) framework we hope to improve predictions of estuarine impacts of runoff and pollution from land, changes in water clarity, and other metrics that are needed by resource managers and other stakeholders to safeguard health and safety around the Chesapeake Bay.