Sea Surface Wind Speed Retrieval Based on Machine Learning models with FY-3D MWRI
The L1-level bright temperature data of the microwave radiation imager(MWRI)in the Fengyun-3D(FY-3D)satellite can be used to retrieve the global sea surface wind speed.This paper discussed the use of multiple linear statistical regression model and machine learning models to retrieve the sea surface wind speed in clear sky and cloud areas.Four-day test sets are put into multiple linear statistical regression model,Random Forest(RF)model,Support Vector Regression(SVR)model,Convolutional Neural Network(CNN)model,and the Stacking Fusion(SF)model for the sea surface wind speed retrieval in the clear sky area,and the obtained optimal root mean square errors(RMSEs)are 1.56,1.31,1.24,1.29,and 1.27 m/s,respectively.Meanwhile,two-day test sets are put into multiple linear statistical regression model,RF model,SVR model,CNN model,and SF model for the sea surface wind speed retrieval in the cloud area and the obtained optimal RMSEs are 2.12,1.98,1.87,1.89 and 1.89 m/s,respectively.To further verify the reliability of the sea surface wind speed retrieval in the clear sky area,the buoy wind speed measured by the National Data Buoy Center(NDBC)in the United States is selected.The results show that the RMSE of the wind speed retrieved by the CNN model and the wind speed measured by the NDBC is 0.74 m/s,and the coefficient of determination(R-square,R2)is 0.80;the RMSE of the wind speed retrieved by the SF model and the wind speed measured by the NDBC is 0.85 m/s,and the R2 is 0.74.These results confirm that the machine learning models can effectively complete the brightness temperature retrieval tasks for global sea surface wind speed with the FY-3D MWRI.
国家科技期刊平台
NSTL
万方数据
