Estimation of Nitrate Concentration and Its Distribution in the Northwestern Pacific Ocean by a Deep Neural Network Model

As a fundamental nutrient for marine biogeochemical processes, the magnitude and spatial distribution of nitrate concentrations are insufficiently measured in the interior ocean. In the present study, a deep neural network (DNN) model was developed to estimate nitrate concentrations in the upper northwestern Pacific Ocean (NPO). This model takes the temperature and salinity profiles as the primary input variables. Since the subtropical and tropical regions are featured by different spatial patterns of nitrate concentrations, we separately trained the model to improve the prediction skill. The predictive results indicate that the DNN model performs well in depicting both spatial and seasonal variability of nitrate concentrations. The sensitivity experiments show that temperature is the dominant factor for the nitrate estimation, while salinity has a relatively small effect, but it cannot be ignored in improving the prediction accuracy. Furthermore, using the temperature and salinity data from World Ocean atlas (2018), we found our DNN model has a good generalization ability on nitrate estimation in NPO. This model can be applied to further studies on nitrate’s spatiotemporal variability and mechanism around the global ocean.

Deep Sea Research Part I: Oceanographic Research Papers