Research On Retrieval Algorithm Of All Sky Atmospheric Temperature And Humidity Profiles From The FY-4A GIIRS

With the development of global climate and numerical weather forecast models,atmospheric temperature profiles,humidity profiles,trace gas content,and aerosol content are needed to be monitored accurately on a global basis.Atmospheric temperature and humidity profiles are the key parameters of the numerical weather forecast and atmospheric research,which are of great application value in weather analysis and numerical weather forecast models and the improvement in the accuracy of the weather forecast.Spaceborne infrared hyperspectral sounders can observe the atmospheric temperature and humidity profiles with high vertical resolution.This paper performs the cloud detection method based on the Level 1 observations from FY-4A/GIIRS(geostationary interferometric infrared sounder).On this basis,a physical retrieval approach based on the one-dimensional variational(1D-Var)algorithm and deep machine learning algorithm(one-dimensional convolutional neural network retrieval algorithm and three-dimensional convolutional neural network retrieval algorithm)are applied in this paper to simultaneously retrieve atmospheric temperature and humidity profiles under all weather conditions.The time-space matched reanalysis data of European Centre for Medium-Range Weather Forecasts(ECMWF),radiosonde observations from 81 upper-air stations in China,and level 2 operational products from the Chinese National Satellite Meteorological Center(NSMC)are used to validate the accuracies of the retrieved temperature and humidity profiles.The conclusions are as follows:The cloud detection method based on GIIRS Level 1 observations combines the traditional identification of clear FOV with the estimation of the number of cloud formations and test of the thermal contrast.This cloud detection method gets rid of the dependence on the imager.Compared with the real-time cloud detection product of the Advanced Geosynchronous Radiation Imager(AGRI)and the GIIRS observed brightness temperature at the window channel(900 cm^-1),the cloud detection results are in good agreement with the GIIRS observed brightness temperature and cloud detection product of AGRI.There is a slight difference between the edge of clear FOV and cloud FOV between the GIIRS cloud detection results and the cloud detection product of AGRI.Meanwhile,clear column radiances can be reconstructed under partial cloud cover and used in the subsequent one-dimensional variational algorithm.Using the test samples in the training sample set of 1D-CNN and 3D-CNN as the verification,the results show that the accuracy of the temperature retrieved by 1D-CNN is higher,the mean bias of all altitudes is near 0 K,and the root mean square error(RMSE)of temperature between 800-50 h Pa is less than 2 K.3D-CNN shows a better performance in humidity retrieval.The accuracy of the humidity retrieved by 3D-CNN increases with the increase of height,and the average RMSE of the whole layer is about 0.78 g/kg.The accuracy of retrieval algorithms is lower than that of test samples when observations are used.FY-4A/GIIRS observations during the periods from December 2019 to January 2020(winter)and from July 2020 to August 2020(summer)are used to retrieve temperature and humidity profiles.The results show that under clear sky conditions,among 1D-CNN,3D-CNN,GIIRS Level 2 operational products,and one-dimensional variational algorithm,1D-CNN has the largest retrieval error in terms of temperature and humidity.3D-CNN,one-dimensional variational algorithm and Level 2 operational products have the highest accuracy in temperature between 800-200 h Pa and the mean bias is mostly near 0 K both in winter and summer.Within the whole layer,the temperature RMSE retrieved by the one-dimensional variational algorithm is the smallest(800-200 h Pa RMSE<1 K),and the accuracy of GIIRS Level 2 operational products,3D-CNN and 1D-CNN is reduced in turn.When the GIIRS observations are affected by the clouds,the accuracy of GIIRS Level 2operational products with high precision under clear sky conditions is greatly reduced(the mean bias of some altitudes exceeds 2 K).The retrieval accuracy of the one-dimensional variational algorithm is the highest,and the mean bias of temperature at all altitudes(except near the ground)is near 0 K both in winter and summer.The mean bias of 3D-CNN at the height below200 h Pa is also small,which is equivalent to the one-dimensional variational algorithm.The temperature error retrieved by 1D-CNN is the largest.The temperature RMSE retrieved by the one-dimensional variational algorithm is less than that of the other three methods at all altitudes,especially the RMSE below 100 h Pa in summer is about 1 K;No matter in summer or winter,the retrieval accuracy of 3D-CNN is the second.In general,the retrieved temperature RMSE in summer of all algorithms(except 1D-CNN)is less than that in winter;the retrieved humidity RMSE in winter is less than that in summer.Similar to the clear sky conditions,the humidity retrieval accuracy of 3D-CNN and the one-dimensional variational algorithm is low near the ground.The humidity retrieved by both algorithms is generally lower than the radiosonde value.Except near the ground,the humidity retrieval accuracy of the one-dimensional variational algorithm is better than 3D-CNN.The one-dimensional variational algorithm and convolutional neural network retrieval algorithms developed in this paper can obtain the temperature and humidity profiles at the whole layer,which makes up for the defect that the GIIRS Level 2 operational products(which do not provide humidity profile product)released by NSMC only provide the temperature profile above the cloud top(when there are clouds in the FOV).The former algorithm shows a better performance in retrieval;however,no retrieval is performed under cloudy conditions.The latter algorithm can perform retrieval under all weather conditions and increases the retrieval speed.Taking advantage of the characteristics of high time resolution and regional mobile detection of FY-4A geostationary meteorological satellite,the temporal-spatial distribution and evolution characteristics of the warm-core structure and humidity field can be tracked when the typhoon is in different life-history stages such as development,maturity,and landing using GIIRS observations.