Susanne Elizabeth Craig, Erdem M. Karaköylü

In this study, we present Bayesian machine learning approaches to predict the spectral phytoplankton absorption coefficient - a proxy of phytoplankton biomass - from top of atmosphere measurements of ocean color. This presents a significant advance in ocean color research as it permits the bypassing of conventional atmospheric correction, which is notoriously challenging in optically complex waters such as costal seas and inland waters - regions particularly vulnerable to anthropogenic forcing and climate change. Model prediction accuracy is significantly greater than conventional algorithms, and most, importantly, offers a method to derive biogeochemically relevant information from ocean color in scenarios where it may otherwise not be possible.