Mapping the dynamics of global sea surface nitrate using ocean color data

Abstract

Sea Surface Nitrate (SSN) is crucial for assessing phytoplankton growth and the initiation of new production within the marine environment. Precise estimation of SSN concentrations plays a significant role in understanding marine ecosystem dynamics. In this study, the deep learning model TabularNet (TabNet) was assessed using quality-controlled in-situ measurements from the Global Ocean Data Analysis Project (GLODAP). These measurements included Sea Surface Temperature (SST), Sea Surface Salinity (SSS), Chlorophyll-a concentration (Chla), and nitrate, collected from various regions of the global ocean to achieve accurate SSN estimation. The TabNet model demonstrated superior performance and robustness, achieving accurate global SSN estimations using satellite data. The model yielded a root mean square error (RMSE) of 2.02 μmol/kg, a mean bias (MB) of −0.32 μmol/kg, a mean ratio (MR) of 0.78, and a coefficient of determination (R²) of 0.96. Furthermore, a comparative analysis of TabNet against Random Forest (RF) and Feed Forward Neural Network (FFNN) models was conducted. The results highlighted the robust performance of TabNet in accurately estimating SSN dynamics. TabNet effectively utilized in-situ and satellite data, providing accurate SSN dynamics. This technique offers valuable insights for monitoring global surface ocean nitrate dynamics, enhancing our ability to understand and manage marine ecosystems.