THERMOHALINE PROFILE PREDICTION IN THE GULF OF MEXICO USING DEEP LEARNING AND ARGO DATA
Accurate circulation modeling in the Gulf of Mexico (GoM) is limited by the scarcity of in-situ subsurface observations, which leads to errors in subsurface representations. These deficiencies reduce the reliability of ocean models and shorten the horizon of skillful forecasts. To address this challenge, we present the latest version of the Neural Synthetic Profiles from Remote Sensing and Observations (NeSPReSO), a data-driven framework designed to efficiently reconstruct subsurface temperature and salinity profiles from satellite-derived surface variables. This approach offers a robust alternative to conventional synthetic profile generation methods.
NeSPReSO employs Principal Component Analysis (PCA) to extract dominant modes of variability from Argo temperature and salinity profiles. A neural network is then trained to predict these modes using input features such as time, location, and satellite-based variables, including absolute dynamic topography, sea surface temperature, and sea surface salinity. Model performance was rigorously validated against independent Argo profiles and glider measurements in the GoM, and results show consistent improvements over traditional techniques such as Gravest Empirical Modes (GEM), Multiple Linear Regression (MLR), and the Improved Synthetic Ocean Profile (ISOP). NeSPReSO reduces both root mean square error and bias while effectively capturing key subsurface variability. The resulting synthetic profiles reproduce observed features such as thermoclines, haloclines, and the region’s characteristic temperature–salinity structure.
To support broad community use, we provide an accessible API that enables users to generate synthetic profiles for any location in the GoM across a range of spatial and temporal resolutions. This resource offers the scientific community a valuable tool for estimating quantities such as regional heat content, while enhancing research and forecasting capabilities in oceanography.