Fine Surface Temperature Forecasting Method Research Based On Statistical Downscaling Techniques In China

The statistical downscaling technique based on large scale numerical forecasting productions is one of the effective methods for fine forecast.Researchers interiorly use interpolation methods such as bilinear interpolation and inverse distance interpolation to make a downscaling forecasting. In recent years, the Kalman filter-typed self-adapting decaying average downscaling technique is designed overseas for downscaling forecasting which is better than MOS method. First, this paper will compare the effect of the decaying average downscaling technique of National Centers for Environmental Prediction (NCEP) with interpolation method and then make an improvement in skill for NCEP method. Last, the decaying average downscaling technique will be applied on numerical forecasting productions.1)Low resolution "forecast field" and fine "analysis field" are made with daily average temperature data of weather stations without the effect of forecasting error in numerical forecasting productions, in order to make a downscaling ideal test by using decaying average technique and analyze the feasibility of the method. The result shows that the scheme is feasible in china. The averaging RMSE of l-3d forecasts is1.5℃which is50%of the result by interpolation method. In addition, the scheme improves the prediction result in West China.2)Change the critical self-adapting parameter W of NCEP method into a self-adapting function. The result shows that the method using the self-adapting function is better than the one using the self-adapting parameter W as the error decrease by4%.3)Based on the T213ensemble forecast outcomes,the ERA-Interim reanalysis data and the hourly surface temperature data of weather stations, a forecasting test is designed and the forecasting results of decaying average technique and running-mean window skill are analyzed. Firstly, use the decaying average algorithm and the running-mean window technique to remove the deviation of T213ensemble forecast original outcomes, which turns the systematic bias to0℃and reduces by30%in RMSE. Secondly, contrast various interpolation methods and choose inverse distance squared interpolation method to calculate fine estimated value. Last, apply the NCEP constant decaying average algorithm, the improved decaying average algorithm and the running-mean window technique to make a downscaling forecast. All of the three schemes perform well and reduce the systematic bias made by interpolation effectively, especially reducing the big forecast gap between East China and West China. The best downscaling scheme is the improved decaying average method whose absolute error is1.7-2.7℃in1-7d forecast and less30%than the result of the original outcomes interpolation scheme.4)Analyze the high temperature prediction above30℃in different schemes, which found that all of the schemes make a lower forecast and have a position drifting phenomenon compared with observation result. But comparison with the original outcomes interpolation scheme, the improved decaying average method is ideal for intensity and good for position. With the second correction, the RMSE is close to spread, Talagrand distribution approachs the ideal value and forecast probability approximates the observed frequency for the high temperature events, that is the reliability of the forecasting system and recognition ability for the summer high temperature events are improved.