Precipitation Detection Algorithm Over Land For FY3C MWHS-2 And Its Application In LAPS-WRF Assimilation System

Studies have shown that the error of the initial value of water vapor in the troposphere is an important source of uncertainty in numerical forecasts,especially in short-term(0–12 h)forecasts.Compared with traditional conventional observation data,satellite-borne microwave humidity data can make up for the insufficiency of conventional observations,due to its wide coverage,high vertical resolution,and the ability to penetrate non-precipitation clouds to detect atmospheric humidity information in cloud.Therefore,the assimilation of microwave humidity data is of great significance for improving the initial field of water vapor in the troposphere and increasing the accuracy of numerical prediction.However,due to the factors such as latitude,scan angle,surface emissivity,cloud and precipitation,etc.,the microwave humidity data has a system deviation before entering the assimilation system,which will affect the assimilation effect of the satellite data in the numerical model.In order to make full use of the satellite-borne microwave humidity data to improve the ability of the regional numerical model to predict short-term precipitation,this paper focuses on the precipitation detection algorithm over land for Microwave Humidity Sounder-2(MWHS-2)onboard the Fengyun-3 C satellite(FY3C)and quantitatively evaluates the the assimilation of MWHS-2 in LAPS(Local Analysis and Prediction System)– WRF(Weather Research and Forecasting Model)model.Firstly,the variational characteristics of FY3 C MWHS-2's O-B(Observation minus Background)bias with the radar reflectivity factor are analyzed.It is found that there is a certain positive correlation between the O-B bias for FY3 C MWHS-2 and the radar reflectivity factor.The larger the radar reflectivity factor is,the greater the O-B bias exists.And for the channels2–6 and 11–14,the range of O-B bias in the area where the radar reflectivity factor is less than10 d BZ can be kept between ± 2 K.On the basis of this theory,a PDL(Precipitation Detection over Land)algorithm based on the machine learning method GBDT(Gradient Boosting Decision Tree)is established by using the characteristics of brightness temperature and its difference,latitude,zenith angle and so on.This precipitation detection algorithm can effectively eliminate most of the data affected by precipitation,and the accuracy can reach more than 80 %.After the precipitation detection,the observed brightness temperature of FY3 C MWHS-2 can be more linearly correlated with the simulated brightness temperature.This method can complete precipitation detection effectively by FY3 C MWHS-2 itself,and has a good application prospect.After improving the quality control method for FY3 C MWHS-2,the FY3 C MWHS-2observations are assimilated in the LAPS regional model.The results show that the assimilation of FY3 C MWHS-2 in LAPS-WRF has a certain positive effect on the analysis and prediction of humidity field and wind field.The distribution and falling area of the precipitation are also improved.Additionally,the assimilation of FY3 C MWHS-2 data improved the Ts score by about 23.9 %,especially for the rainfall greater than 50 mm.