High Resolution Estimation of Ocean Dissolved Inorganic Carbon, Total Alkalinity and pH Based on Deep Learning

Abstract

This study combines measurements of dissolved inorganic carbon (DIC), total alkalinity (TA), pH, earth observation (EO), and ocean model products with deep learning to provide a good step forward in detecting changes in the ocean carbonate system parameters at a high spatial and temporal resolution in the North Atlantic region (Long. −61.00° to −50.04° W; Lat. 24.99° to 34.96° N). The in situ reference dataset that was used for this study provided discrete underway measurements of DIC, TA, and pH collected by M/V Equinox in the North Atlantic Ocean. A unique list of co-temporal and co-located global daily environmental drivers derived from independent sources (using satellite remote sensing, model reanalyses, empirical algorithms, and depth soundings) were collected for this study at the highest possible spatial resolution (0.04° × 0.04°). The resulting ANN-estimated DIC, TA, and pH obtained by deep learning shows a high correspondence when verified against observations. This study demonstrates how a select number of geophysical information derived from EO and model reanalysis data can be used to estimate and understand the spatiotemporal variability of the oceanic carbonate system at a high spatiotemporal resolution. Further methodological improvements are being suggested.