Machine-learning estimates of global marine gross primary production

Yibin Huang, Duke University, Durham, United States, David Nicholson, Woods Hole Oceanographic Institution, Department of Marine Chemistry and Geochemistry, Woods Hole, MA, United States, Bangqin Huang, Xiamen University, State Key Laboratory of Marine Environmental Science, Xiamen, China and Nicolas Cassar, Duke University, Earth and Ocean Sciences, Nicholas School of the Environment, Durham, NC, United States
Abstract:
Approximately half of global primary production occurs in the oceans. While the large-scale variability in net primary production (NPP) has been extensively studied, ocean gross primary production (GPP) has thus far received less attention. In this study, we derived two satellite-based GPP models by training two machine-learning algorithms with light-dark bottle incubations (GPPLD-O2) and the triple isotopes of dissolved oxygen (GPP17Δ-O2). The Random Forest and Support Vector Regression algorithms predicted global GPP distributions in agreement with our understanding of the mechanisms regulating primary production, with generally higher values at high latitudes of the northern hemisphere, and in equatorial and coastal regions. Global GPP17Δ-O2 was higher than GPPLD-O2 by an average factor of 1.7 which varied meridionally. The discrepancy between GPP17Δ-O2 and GPPLD-O2 simulations can be partly explained by the known biases of each methodology. After accounting for some of these biases, the GPP17Δ-O2 and GPPLD-O2 approaches converged to 10.3-12.2*1015 mol O2 yr-1, or 107-126 Gt C yr-1 for the global ocean. Our result suggests similar NPP/GPP ratios for land and ocean which contrast with the common axiom of a higher fraction of GPP consumed by autotrophic respiration on land. This surprising result may be due to biases in the measurements, warranting further study.
Back to: Gases As Tracers of Ocean Biogeochemical and Physical Processes Posters