Key Words:IN-SITU; NEURAL-NETWORK; HEAT-CONTENT; TEMPERATURE; SALINITY; CIRCULATION; VARIABILITY; MECHANISMS; WIND; SST
Abstract:Accurate estimation of high-resolution thermocline depth is important for investigating ocean processes and climate variability on multiple scales. Due to the sparse coverage and high costs associated with in situ observations, reconstructing ocean interior structure from sea surface data serves as a valuable alternative. In this study, a new deep learning model named Enhanced Block Attention Module-Convolutional Neural Network (EBAM-CNN) was proposed to reconstruct thermocline depth in the tropical Indian Ocean (TIO) from 1993 to 2022. Absolute dynamic topography (ADT), sea surface temperature (SST), and sea surface wind (SSW), along with geographic information (latitude and longitude) and temporal data, were employed as input variables. In comparison with the traditional convolutional neural network (CNN) model, the proposed model demonstrates better performance, with an overall Root Mean Square Error (RMSE) of 5.29 m and a Pearson Correlation Coefficient (R) of 0.87. In addition, this study employs a downscaling approach to reconstruct higher-resolution thermocline depth data. An analysis of the downscaling results confirmed that the proposed framework effectively reconstructed mesoscale sea subsurface features from high-resolution surface observations, significantly enhancing thermocline depth estimates and providing robust data support for oceanic and climatic research.
Volume:196
Issue:
Translation or Not:no