14:30–14:45 | ABSTRACT 1357
DECODING PHYTOPLANKTON COMMUNITIES FROM HYPERSPECTRAL OCEAN COLOR
Phytoplankton community composition (PCC) plays a key role in ocean biogeochemical cycling, climate regulation, and marine ecosystem dynamics and the ability to characterize it from ocean color measurements is one of the key objectives of the NASA PACE mission. Historically, ocean PCC algorithms have been trained using HPLC-determined taxonomic marker pigments. However, marker pigments are an imperfect proxy of PCC – they can co-occur across very different taxonomic groupings and size classes, and their absolute and relative abundances can be strongly affected by environmental and physiological factors. In this work, we demonstrate that quantitative PCC algorithms can be derived from only ocean color and easily available environmental data. We used a coupled biogeochemical-radiative transfer model to simulate hyperspectral remote sensing reflectance (Rrs). The model was constrained by the inherent optical properties of six phytoplankton groups – diatoms, chlorophytes, cyanobacteria, coccolithophores, dinoflagellates, and Phaeocystis. We then developed an extreme gradient boosting (XGBoost) regression model to predict the abundance of the six phytoplankton groups. Validation was performed on an independent test set, and the model achieved high accuracy for five groups (R² > 0.95), but lower skill for dinoflagellates (R² = 0.53). Feature attribution with SHAP showed that, alongside specific Rrs spectral regions, sea surface temperature strongly influenced predictions, suggesting that integrating non-PACE temperature data could improve performance. Finally, we introduce strategies for further model development using the rapidly growing PACE validation datasets – specifically from instruments that provide taxonomic identification and enumeration using imaging flow through cytometry.
Susanne Craig, NASA Goddard Space Flight Center/UMBC GESTAR II, [email protected], https://orcid.org/0000-0002-8963-0951
Erdem Karaköylü, Independent consultant, [email protected]
Ian Carroll, NASA Goddard Space Flight Center/UMBC GESTAR II, [email protected]
