An Improved Bias Removed Method for Precipitation Prediction and Its Application

On the basis of traditional bias removed (BR) method, grading bias removed (GBR) method is designed by adding the step of correcting according to three precipitation orders, which are more than 0.1 mm, 25 mm and 50 mm, respectively. Then, using observations of precipitation and numerical precipitation prediction of ECMWF from April to August in 2011 and 2012, the real-time precipitation forecast of 1-5 days at summer (June-August) over China in 2012 is corrected by GBR method using two different training periods, i.e., the mixed training phase and 60-day running training phase, and the results of them are called GBR_h and GBR_60, respectively. In order to contain information of heavy precipitation in forecast phase as much as possible, the mixed training period is composed of a 30-day period before the forecast phase and two 15-day periods before and after the same phase one year ago, according to characteristics of summer monsoon rainfall of China.Equitable-threat scores (ETS) of forecast over China at many thresholds of precipitation are examined, in order to compare results of the mixed training and the 60-day running training period using GBR. It reveals that both of two corrected results have higher skill than precipitation prediction of ECMWF, at the threshold of beneath 25 mm, the improving amplitude of them are very close (the improvement of GBR_h and GBR_60 are 19.5% and 19.1%, respectively). However, for those above 25 mm, GBR_h apparently has bigger amplitude which is up to 73.5%, and GBR_60 is only 55.9%. Especially in the situation of correcting the local heavy precipitation prediction, the correcting effect of GBR_h is much better. Furthermore, the correlation coefficient is also calculated, and the result shows that the pattern of precipitation prediction is also modified by GBR_h and GBR_60, and the former also has better performance.By analyzing errors of three orders calculated through two different training periods, it is clear that the key point of successfully improving the initial ECMWF forecasts is to add the step of grading bias removed, and a larger improvement of ETS can be expected if more appropriate mixed training period is chosen. It is assumed that according to the obvious effect of this experiment which are easy to apply in operation, this grading bias-removing method of mixed training period will make a very useful product for real time events and have favorable application prospects.