Research On Quantitative Inversion Method And Application Of Three-dimensional Wind Field Based On Machine Learnin

The Geosynchronous Interferometric Infrared Sounder(GIIRS)onboard the Fengyun-4A(FY-4A)is the first geostationary hyperspectral infrared sounder.It can provide continuous three-dimensional(3D)information of atmospheric temperature and water vapor,and the atmospheric horizontal wind field at different heights through tracking the movement of water vapor.In this study,GIIRS data from FY-4A observations with 30-minute intervals during Typhoon Maria(2018)were used to test 3D horizontal wind field algorithm in both clear and partially cloudy skies,with the focus on how to improve the 3D wind field retrieval from partially cloud filled GIIRS footprints by including the collocated high spatial resolution cloud products from Advanced Geostationary Radiation Imager(AGRI)onboard the same platform.Independent ERA5,CRA40 analysis and radiosonde data were used as reference data to verify the 3D horizontal wind field retrievals in both clear and partially cloudy skies.Results indicated that:(1)the tropospheric wind field could be obtained from GIIRS brightness temperature(BT)measurements from two consecutive times.In the clear skies,RMSE was less than 1.5m/s,and the absolute direction difference between retrievals and reference data was about 15° in this particular case;under some clear skies,RMSE was 1.5 ～1.7m/s,and the absolute direction difference was about 20°.Compared with the traditional optical flow(OF)method,GIIRS BT based direct retrieval shows better accuracy with RMSE and direction absolute difference smaller than those from the OF method.(2)When classified by cloud coverage(CC)and cloud-top pressure(CTP),it is shown that the more cloud cover or the higher cloud top,results in the greater RMSE.When CC and CTP are included as additional input parameters,the GIIRS BT based 3D wind retrievals reduced RMSE decreases,indicating that high spatial resolution AGRI cloud products can improve wind field retrieval accuracy in some cloud covered areas.(3)The wind profile retrievals from GIIRS BT based method consistent with the wind profiles of CRA40 analysis and dropsonde data,indicating the feasibility of deriving 3D horizontal wind fields from geostationary hyperspectral infrared BT measurements.In order to verify the applicability of DNN wind field inversion algorithm,three other independent typhoon cases(not involved in the training and testing of DNN model)were also used to test the DNN model.They are Typhoon Ampil in July 2018,Typhoon Mangkhut in September 2018,and Typhoon Lekima in August 2019.At the same time,in order to verify the applicability of the model to the conventional observation data of GIIRS in China,the wind field retrievals model was also applied to the observation data of GIIRS in China.The time resolution of the data was 2 hours.This study provides a practical scientific method(from 1D wind field to 3D wind field)for obtaining 3D wind products with high time resolution from stationary orbit hyperspectral infrared detector data,and proves that combining the imager with the same platform can improve the wind field retrieval accuracy in some cloud field areas.It is also proved that both the height and the amount of clouds in the infrared hyperspectral field of view will affect the results of wind field retrieval products,which indicates that the cloud characteristic information in the infrared hyperspectral field of view can be used as one of the quality control basis of wind field retrieval products.If the application of wind farm products is involved,the wind farm products in deep cloud areas should be carefully used.In addition,the cloud characteristic information in the infrared hyperspectral field of view can also be used as the quality marking code of wind field products directly or indirectly.This method theoretically proves the feasibility and reliability of three-dimensional wind field retrieval with stationary orbit hyperspectral detection data with intensive time observation period,which will provide added value for a variety of quantitative applications such as data assimilation,approaching weather monitoring and situation awareness.