Linking satellites to genes with machine learning to estimate phytoplankton community structure from space

Abstract. Ocean color remote sensing has been used for more than 2 decades to estimate primary productivity. Approaches have also been developed to disentangle phytoplankton community structure based on spectral data from space, in particular when combined with in situ measurements of photosynthetic pigments. Here, we propose a new ocean color algorithm to derive the relative cell abundance of seven phytoplankton groups, as well as their contribution to total chlorophyll a (Chl a) at the global scale. Our algorithm is based on machine learning and has been trained using remotely sensed parameters (reflectance, backscattering, and attenuation coefficients at different wavelengths, plus temperature and Chl a) combined with an omics-based biomarker developed using Tara Oceans data representing a single-copy gene encoding a component of the photosynthetic machinery that is present across all phytoplankton, including both prokaryotes and eukaryotes. It differs from previous methods which rely on diagnostic pigments to derive phytoplankton groups. Our methodology provides robust estimates of the phytoplankton community structure in terms of relative cell abundance and contribution to total Chl a concentration. The newly generated datasets yield complementary information about different aspects of phytoplankton that are valuable for assessing the contributions of different phytoplankton groups to primary productivity and inferring community assembly processes. This makes remote sensing observations excellent tools to collect essential biodiversity variables (EBVs) and provide a foundation for developing marine biodiversity forecasts.