New Method of GNSS-R Wind Speed Retrieval Based on Empirical Orthogonal Function

Abstract

The geophysical model function (GMF) and the cumulative distribution function (CDF) are two main algorithms used for retrieving sea surface wind speed from GNSS-R observations. The difficulties in segment fitting and the complicity in parameter adjustment hinders the wide application of the GMF, while the wind speed retrieved by using the later renders a deviation of about –2 m/s when the wind speed is in the range of 0–3 m/s. This paper proposes a new algorithm based on the empirical orthogonal function (EOF) to retrieve wind speed from GNSS-R observations. Based on the EOF, two wind retrieval models are trained by using the delay doppler map average (DDMA) and the leading edge slope (LES) as the training set, respectively. In the training, the wind speed data (resolution: 30 km, 1 h) from European Centre for medium-range weather forecasts (ECMWF) reanalysis V5 (ERA5) are used as the ground truth data. The DDMA and LES are 80% of the resampled NASA’s cyclone global navigation satellite system (CYGNSS) and GNSS-R data of 2019 at an interval of 10 s. The final wind speeds are calculated from the two kinds of retrieval by a minimum variance (MV) criterion. At last, the test data set (20% of CYGNSS data) are used to evaluate the accuracy of the final wind speeds. The result shows that when the reference wind speeds are below 20 m/s, the mean bias and RMSE of the retrieved wind speeds are 0.026 m/s and 1.77 m/s when using the ERA5 wind speeds as the reference, which are 0.23 m/s and 1.67 m/s when using advanced scatterometer (ASCAT) wind speeds as the reference. This proves that the EOF algorithm has a good performance in retrieving sea surface wind speed.