Sea Surface Wind by Spaceborne SAR Using Machine Learning Methods

As a principal driver of global ocean circulation dynamics, sea surface winds (SSW) require accurate retrieval methodologies. This study develops and implements a backward propagation (BP) neural network framework for sea surface wind speed (SSWS) estimation using HH-polarized Gaofen-3 (GF-3) synthetic aperture radar (SAR) imagery. Comparative analyses reveal performance metrics of the BP model against reference datasets: buoy measurements yield wind speed retrieval bias of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.82 \mathrm{m} / \mathrm{s}$</tex>, root-mean-square error (RMSE) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.42 \mathrm{m} / \mathrm{s}$</tex>, and scatter index (SI) of 37.87 %, while ASCAT satellite data comparisons show improved metrics with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$-0.03 \mathrm{m} / \mathrm{s}$</tex> bias, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.36 \mathrm{m} / \mathrm{s}$</tex> RMSE, and 15.87 % SI for the testing ensemble. The experimental results demonstrate the BP neural network's promising capabilities in SSWS derivation from HH-polarized GF-3 acquisitions, suggesting its operational potential for marine meteorological monitoring applications.