Range current retrieval fromsentinel-1 SAR ocean product based on deep learning

Abstract

ABSTRACT In this study, the feasibility of current retrieval from Sentinel-1 (S-1) synthetic aperture radar (SAR) in the radar look/range direction is investigated. S-1 Ocean (OCN) products acquired in interferometric wide (IW) mode in the regions with the western boundary current, i.e., the western Pacific and western Atlantic, are collected for the period from 2020 to 2022, which are collocated with the current field from HYbrid Coordinate Ocean Model (HYCOM). The OCN wind, HYCOM current, and Stokes drift estimated from the OCN wave parameters are geometric projected to be the range direction. In addition, the Doppler centroid anomaly (DCA) is estimated using the difference between the radar return Doppler frequency and the predicted Doppler shift, which are derived from the OCN products. The dependences of the upper ocean dynamics in the range direction on the DCA are studied, and it is found that the range Stokes drift, wind speed, and current speed are linearly related to the DCA. Based on deep learning, denoted as multi-layer perceptron, a range current retrieval algorithm from the OCN product is developed using two-thirds of the collocated dataset, and the root mean square error (RMSE) of the range current speed converges to 0.15 ms−1. In particular, the wave-induced surface Stokes drift is considered in the process. Comparison of one-third of the dataset yields an RMSE of 0.14 ms−1 for the range current speed, a correlation coefficient (r) of 0.85, and a bias of −0.001 ms−1. Validation against several moored buoys shows an RMSE of 0.12 ms−1 with a r of 0.74 and a bias of −0.01 ms−1. Under this circumstance, it is believed that the algorithm used in this study is applicable for range current retrieval from the S-1 OCN product.