Statistical downscaling forecasting of precipitation and 2 m air temperature was obtained based on ECMWF model forecast data from March to November 2018. The interpolated precipitation was corrected using the frequency matching method firstly and then the threshold method, the interpolated temperature was corrected by using the Kalman filter-type decreasing average statistical downscaling technique, finally the hourly precipitation and temperature prediction were obtained. The results are as follows: (1) For the accuracy of rain probability forecast, it was obviously improved by using the frequency matching method and the threshold method for most forecasting time, and the maximum improvement range was more than 20% for the former. For the relative error, the threshold method had reduced the occurrence of false alarms considerably. For the short-term heavy rainfall with 1 h rainfall greater than or equal to 20 mm, the TS score was also improved significantly after using the frequency matching method. For the Typhoon “Amby” event in 2018, in addition to the above improvement effects, the frequency matching method improved the prediction capacity of the model about the amount and patterns of rainfalls, and the threshold method corrected false-alarm station completely. (2) For the test of temperature forecast of ECMWF model, the absolute error was the largest in March for almost forecast time. After using the Kalman filter-type decreasing average statistical downscaling technique, the absolute error of temperature in different months decreased to varying degrees. In general, the absolute error curve after correction still had the periodic fluctuation with the extension of forecast period, and the position of wave peak and trough was basically the same as those before correction, and the greater the absolute error, the greater the correction range was. For the temperature case, the accuracy of the spatial distribution of temperature prediction was retained, and the absolute error decreased significantly after correction.

 Based on conventional meteorological observation data in Tianjin and the ECWMF ERAInterim reanalysis data (0.125°×0.125°), the transformation processes of raintosleet and then sleettosnow were analyzed from October to next April during 2000-2015 in Tianjin. The humidity and geopotential thickness during the transformation process of precipitation phases were analyzed, and 9 factors, namely temperature on 850 hPa, 925 hPa,1000 hPa and surface, thickness from 1000 to 850 hPa and from 850 to 700 hPa, 0 ℃ layer height, -4 ℃ layer height and 925 hPa relative humidity, were identified, which had close relationship with transformation process of precipitation phases. Further, their corresponding threshold indexes for different precipitation phases were given. And through the analysis of 3 weather cases, the usability of the indicators had been further verified. The discriminant equation for precipitation phases was established by using the 9 factors, it was found that the backtesting discriminant accuracy for rain and snow was higher than 80%.