Hyperspectral retrievals of phytoplankton abundance and absorption properties in optically complex waters

Following more than two decades of research and developments made possible through various proof-of-concept hyperspectral space-borne or airborne remote sensing missions, there is a consensus within the aquatic remote sensing community that hyperspectral imaging (HSI) ameliorates the quality of remotely sensed in-water products. However, thus far, most studies have been limited in geographic scope and lack global representativeness in coastal estuaries or freshwater ecosystems. Here, given a fairly large in situ database, we demonstrate the utility of hyperspectral radiometry for the retrievals of phytoplankton properties, i.e., its biomass and absorption characteristics, and compare those against products obtained from multispectral observations. Using a class of neural networks known as Mixture Density Networks (MDN) for retrievals, we show that hyperspectral data (409-719 nm @ ~ 5nm resolution) improve biomass retrievals by > 3X compared to that derived from band-ratio algorithms currently applied to multispectral data. Further, our inverted hyperspectral phytoplankton absorption spectra (a_ph) exhibit fine spectral structures that can significantly enhance our understanding of phytoplankton communities, their groups, pigment composition, and (when combined with environmental indicators) species. To further assess the value of HSI in optically complex waters across space and time, several images acquired by the Hyperspectral Imager for Coastal Ocean (HICO) are processed to biomass (i.e., chlorophyll-a) and a_ph followed by the validation against limited near-coincident in situ data. The image analyses performed in the lower Chesapeake Bay, Lake Erie, and Monterey Bay provide strong evidence on the utility of hypespectral data for identifying phytoplankton communities in optically complex waters.