An intelligent algorithm for constructing quasi-real-time sea surface wind field
In this paper,the correction model of CMA-GFS numerical model wind field is constructed based on the deep learning U-Net network,and the construction of the quasi-real-time sea surface wind field is rapidly accom-plished by interpolation method using the corrected wind field with the correction model as the background field(CMA-GFS_Unet),and using the scatterometer sea surface wind data from the four satellites,namely,HY-2B/2C/2D and MetOp-B as the observation data.This intelligent algorithm can realize the generation of global sea surface fusion wind field(Fusion_QRT)with a spatial resolution of 0.25° and a temporal resolution of 6 hours in quasi-real time with a lag of 3 hours.The CMA-GFS,CMA-GFS_Unet and Fusion_QRT wind fields are evaluated using the CCMP fusion wind field data and the 10 m wind vector data from the Chinese offshore buoys,respect-ively.The results show that the quality of the CMA-GFS_Unet wind field has been significantly improved,and the quality of the wind speed of the Fusion_QRT wind field has been further improved but the quality of the wind direc-tion has been slightly reduced.The mean absolute errors(MAEs)of wind speed are 1.13 m/s,0.89 m/s and 0.84 m/s for the three wind fields by using CCMP data as reference,and the CMA-GFS_Unet and Fusion_QRT wind fields have improved by 21.3％and 25.7％compared to the CMA-GFS,respectively;while the MAEs of wind direction are 17.5°,15.5° and 16°,and have improved by 11.3％and 8.6％,respectively.The MAEs of wind speed are 1.50 m/s,1.36 m/s and 1.28 m/s for the three wind fields by using buoy data as reference,and have improved by 9.5％and 14.7％,respectively;while the MAEs of wind direction are 23.3°,22.7° and 24.0°,and have improved by 3.0％and-3.9％,respectively.
U-NetCCMPCMA-GFSHY-2B/2C/2DMetOp-Bquasi-real-timesea surface wind field
NETL
NSTL
万方数据
