AI-Aided Systematic Numerical Modeling of North Atlantic Gyre Surface Conditions with Multiple Deep-Learning Architectures

Understanding oceanic circulation is hugely important in helping to understand the health of the Earth system. Unfortunately, our understanding of the ocean's complex behavior is hugely limited by a lack of data, large uncertainties in our current models, and the possibility of the cycle's nature changing as a result of global heat and energy imbalances. In light of its potential influence on and by the Earth's climate, we seek an alternative approach in the realm of artificial intelligence (AI). We report here our continuing progress in utilizing deep-learning architectures to model ocean surface parameters and behaviors of the North Atlantic Gyre (NAG) using the University Corporation for Atmospheric Research's ICOADS dataset. We have investigated the applications of three different kinds of neural networks—a convolutional neural network (CNN), a convolutional long short-term memory network (ConvLSTM), and a spatiotemporal transformer (STT)—to this field in an attempt to acquire a comprehensive understanding of how well current AI architectures can simulate ocean surface conditions and new insights they can offer in this field, i.e. areas of the globe that are currently not observed or predictive capabilities in the context of deviating climate patterns. With a new learning regiment, we find our models can follow our observations to a close accuracy: in the case of mean North Atlantic sea surface temperature (SST), our convolutional long short-term model was able to achieve a root mean squared discrepancy of 0.927°C (loss = 0.0013) over the entire North Atlantic Ocean. We bring forth further findings and analyses of general model training and testing strategies, as well as provide a structural analysis of certain regions of the Atlantic Ocean to demonstrate the efficacy of our algorithms in modeling. We finally discuss artificial intelligence's manner of implementation in the field of ocean science and recommend additional strategies that would help supplement our understanding of geophysical phenomena.