Wind Speed Retrieval Using Coastal Ocean-Scattered GNSS Signals

This paper aims to propose a new coastal observable based on the computation of the ratio between the coherent and incoherent averaging as an alternative to the coherent time and the effectiveness of incoherent averaging proposed by previous works. Experimental data have been processed to develop the relationship between wind speed and the proposed observable. The influence on fitting results of elevation angle, averaging samples, and the delay lag to align the local replica to the scattered signals are analyzed to obtain an optimal retrieval. When the number of averaged samples is less than 150, and the delay range relative to the specular reflection is from $-$0.4 to 0.4, fitting results with a root mean square error (RMSE) less than 2.0 m/s and a correction coefficient larger than 0.8 for signals from BeiDou GEO and GPS implemented compensation of an elevation angle are obtained. Finally, the feasibility of multisatellite observation using multiple regression and neural networks is demonstrated. Neural networks can get better fitting results than multiple regression for both BeiDou GEO and GPS. In addition, for GPS satellites, when the elevation angle of GPS is considered as an input of a neural network, the influence of the elevation angle is greatly weakened, so that the RMSE of 1.03 m/s and the correlation coefficient of 0.96 can be obtained.