POSTER Session 3

A FRAMEWORK TO ESTIMATE OPEN-OCEAN DIATOM CARBON BIOMASS FROM REMOTE SENSING OBSERVATIONS

Quantifying carbon biomass concentrations of diatoms, a major group of phytoplankton found throughout the world ocean and a key player in marine food webs and global carbon cycling, adds valuable knowledge to our understanding of the ocean ecosystem. Acquisition of large in situ datasets on diatom biomass from various locations across the globe using automated imaging-in-flow cytometry in recent years has enabled the development of a model to estimate diatom carbon biomass in the open ocean. Here we combine in situ observations collected during multiple expeditions during 2015-2022. We trained a Random Forest model to predict diatom carbon biomass from temperature, chlorophyll a, the magnitudes of Gaussian-function absorption representing chlorophylls b, c, and photoprotective carotenoid pigments, and particulate backscattering. We used group-aware cross-validation, a technique that accounts for the spatiotemporal dependence in surface oceanographic measurements (such as temperature and chlorophyll concentration) and thus prevents artificially increased accuracy metrics. Diatom carbon estimates from this model were compared to a prior model based only on chlorophyll a concentration. While model performance metrics (RMSE, MAE, Median AE) on average are similar in magnitude to baseline metrics calculated using the chlorophyll-based model, the Random Forest model reveals different spatial patterns in surface diatom carbon biomass distributions relative to chlorophyll-based estimates, as a result of the additional input information beyond chlorophyll a alone during model training. This demonstrates that the added information gleaned from hyperspectral ocean color measurements made by the PACE Ocean Color Instrument lends new insight into distributions of phytoplankton communities.

Alison Chase, University of Washington, USA, [email protected], https://orcid.org/0000-0002-1862-9822

Peter Gaube, University of Washington, Applied Physics Laboratory, USA, [email protected]

Valentina Staneva, University of Washington, eScience Institute, USA, [email protected]

Nils Haentjens, University of Maine, USA, [email protected]

Charles Stern, Columbia University, Lamont Doherty Earth Observatory, USA, [email protected]

Emmanuel Boss, University of Maine, USA, [email protected]

Lee Karp-Boss, University of Maine, USA, [email protected]

Guillaume Bourdin, University of Maine, USA, [email protected]

Code of Conduct
Privacy Policy