Research On The Sea Surface Wind Speed Inversion Model Of Spaceborne GNSS-R Based On Machine Learnin

Sea surface wind speed,as one of the important parameters reflecting the physical state of the ocean,plays an important role in global climate research.In order to obtain timely,accurate,comprehensive and high-quality data,it is necessary to carry out sea surface wind speed retrieval research.In view of the fact that the current sea surface wind speed inversion model is mostly established based on the reanalysis wind speed product,the error of the input independent variable will lead to the problem of poor accuracy of the built model.In this paper,based on the accuracy test of L2 wind speed products of Cyclone Global Navigation Satellite System(CYGNSS)in the United States,according to the nonlinear relationship between the observed characteristic values of CYGNSS satellite L1 data and wind speed,and taking the wind speed data measured by buoys as the target value,the research on the remote sensing model of sea surface wind speed is carried out,so as to realize the sea wind speed inversion,and then analyze the accuracy of wind speed inversion from two dimensions of season and sea area.In addition,this article also analyzes the temporal and spatial distribution characteristics of global sea surface wind speed to provide a reference for improving the accuracy of tropical cyclone forecasting and early warning.The specific research work and results are as follows:(1)The accuracy of CYGNSS L2 satellite wind speed products are evaluated by using the buoy measured wind speed of the National Buoy Data Center of the United States and the best path wind speed of the National Hurricane Center as reference data.The results show that the correlation coefficient(R)between the medium and low wind speed products estimated by the CYGNSS fully developed ocean(FDS)model and the wind speed measured by buoys is0.716 ～ 0.778,the root mean square error(RMSE)is 2.06 ～ 2.24 m/s,and the mean absolute error(MAE)is 1.54 ～ 1.65 m/s,and the mean bias(MB)is-0.35 ～ 0.16 m/s.There is a big difference between the high wind speed products estimated by the CYGNSS Incomplete Ocean(YSLF)model and the NHC best path wind speed.The correlation coefficient R between the two models is 0.548 ～ 0.950,the RMSE is up to 18.26 m/s,and the CYGNSS wind speed is lower than the NHC maximum wind speed.For the wind speed products estimated by the two models of CYGNSS,the normalized bistatic radar scatter The wind speed derived from cross-section(NBRCS)has better accuracy than that derived from leading slope(LES).(2)Using the L1 observation data of CYGNSS satellite and the measured wind speed data of buoys,three machine learning methods of support vector regression(SVR),random forest(RF)and BP neural network(BPNN)were selected to build a sea surface wind speed remote sensing model to realize the sea surface wind speed.inversion.The results show that the RF model has the highest accuracy of inversion wind speed,the correlation coefficient between the inversion wind speed and the measured wind speed is 0.84,the RMSE is 1.56m/s,the MAE is 1.15 m/s,and the MB is-0.07 m/s.The BPNN model Secondly,the correlation coefficient between the inverted wind speed and the measured wind speed is 0.79,the RMSE is 1.75 m/s,the MAE is 1.31 m/s,and the MB is 0 m/s.The inversion accuracy of the SVR model is the worst,and the inversion The correlation coefficient between the wind speed and the measured wind speed is 0.77,the RMSE is 1.81 m/s,the MAE is 1.18 m/s,and the MB is 0.03 m/s.(3)The inversion accuracy of the three sea surface wind speed remote sensing models was evaluated from the two dimensions of season and sea area.The results show that the inversion accuracy of the RF model is the best among the three models.In different seasons,the correlation coefficient between the wind speed retrieved by the RF model and the measured wind speed is 0.80 ～ 0.86,the RMSE is 1.37 ～ 1.70 m/s,the MAE is 1.05 ～ 1.23m/s,and the MB is between-0.06 ～ 0.08 m /s.In different sea areas,the correlation coefficients between the wind speed in coastal waters and the wind speed in open seas retrieved by the RF model and the measured wind speed are 0.80 and 0.83,the RMSE are1.75 m/s and 1.37 m/s,and the MAE is 1.27 m/s and 1.05 m/s,MB are 0.01 m/s and 0.16 m/s respectively.Finally,global sea surface wind speeds retrieved using the RF model with the best inversion accuracy are overall latitude dependent,mainly manifested in the fact that the wind speed in middle and high latitudes is greater than that in low latitudes,and the maximum wind speed area appears near 37° north and south latitude.In sea areas,the wind speed in the Atlantic Ocean is greater than that in the Pacific Ocean and the Indian Ocean,and the wind speed in the South Atlantic is stronger than that in the North Atlantic.In terms of seasons,the wind speed is higher in autumn and winter,and lower in spring and summer,and there are obvious seasonal differences in the distribution of the northern and southern hemispheres.The wind speed in the northern hemisphere is smaller than that in the southern hemisphere in summer,and the wind speed in the northern hemisphere is greater than that in the southern hemisphere in winter.