Abstract
This paper presents a method to invert the sea surface wind field from Sentinel‐1 synthetic aperture radar (SAR) interferometric wide swath (IW) band mode ground range detected (GRD) vertical polarization (VV) data based on the deep residual network (ResNet). Compared with the traditional algorithm (i.e., using geophysical model functions (GMFs)), the wind field inversion based on the ResNet deep learning method solves the problem of separate wind direction and wind speed inversion. The ResNet model is developed based on the results of 212 VV polarized IW images and cross calibrated multiplatform (CCMP) data collected in 2021 in the offshore waters of China. The root mean square error, standard deviation, and bias of the inverse wind direction of the ResNet model are 12.5°, 9.8°, and 7.4°, respectively, and the root mean square error, standard deviation, and bias of the inverse wind speed are 2.33, 1.495, and −1.375. The ResNet is better than the conventional local gradient and C‐band model models, further confirming the advantages shown by the deep learning algorithm in nonlinear fitting and its potential application in sea surface wind field inversion.
Reference24 articles.
1. Spatial Scale Effect on Wind Speed Retrieval Accuracy Using Sentinel-1 Copolarization SAR
2. Remote sensing of surface wind and wave from spaceborne synthetic aperture radar;Xiaofeng Li;Journal of Radars,2020
3. A review for retrieving wind fields by spaceborne synthetic aperture radar;Baowei Z.;Journal of Sensors,2022
4. HH polarization SAR wind speed inversion evaluation;Baowei Z.;Remote Sensing Information,2022
5. The comparative analysis of C band geophysical model functions;Yi Z.;Remote Sensing Information,2012