针对动力气候模式对区域或更小空间尺度内的日降水预测技巧偏低的问题,应用最优子集回归(OSR)方法对国家气候中心业务化的月动力气候模式(DERF)输出的高度场、风场和海平面气压场进行降尺度处理用于降水预测,旨在提高预测准确率.1982-2006年交叉检验结果表明:OSR方法能显著提高降水预测技巧,其中11～40 d改善效果最为显著.在此基础上,应用一步法和两步法两种统计降尺度方法预测极端降水日数,交叉检验结果表明:两种方法均优于随机预测,冬季两步法预测技巧略高于一步法,夏季一步法略优于两步法.综合认为OSR,OSR结合随机天气发生器(WG)两种统计降尺度方法对月尺度降水或极端降水日数的预测均具有较高的技巧,可作为短期气候预测的重要参考信息.%The predicion of precipitation especially extreme precipitation is important but difficult. Dynamical climate models play important roles in the climate prediction and show good skills in large-scale circulation prediction. However, its prediction skill of daily precipitation is limited on regional or smaller spatial scale. So dynamical or statistical downscaling is developed to provide prediction with high resolution. Statistical downscaling can make full use of the large -scale circulation information with high skill of global climate model, and simulate everyday climate variables on the regional or point scale. It has become a popular method in climate prediction and climate change research.Dynamical Extension Regional Forecast Model (DERF) by National Climate Center, CMA has been used in the climate prediction for nearly ten years. Like other global climate models, it has good skills in predicting circulation fields such as height, wind, and sea level pressure. Optimum subsets regression (OSR) is used to predict precipitation anomaly at 133 stations in China for 6 periods (1 -10 days, 11-20 days, 21 -30 days, 31 - 40 days, 1-30 days, 11-40 days) using geopotential height, zonal wind, merid ional wind and sea level pressure as predictors by DERF. The OSR models are verified with cross valida tion method using data from 1982 to 2006. Five operational sores (Ratc, CLTc, P, ACC and TS) are compared with the results directly forecasted by DERF. The results show that OSR can improve prediction skill to different extents, especially for 11- 40 days. Then two statistical downscaling methods are used to predict number of extreme precipitation days. One is predicting directiy as predictant with OSR method using large circulations from DERF as predictors (named as 1-step method), which is similar to precipitation anomaly prediction. The other one is to compute the day number using simulation results of weather generator (WG) under the condition of precipitation anomaly predicted by OSR downscaling (named as 2 step method). Random prediction is compared with the two methods. Crossing verification from 1982 to 2006 show that the predict skill of the two statistical methods is better than that of random prediction. The skill of 2-step method is better than 1-step method to predict number of extreme precipitation days in win ter, but worse in summer. It can be concluded that the methods of OSR and combingtion of OSR and WG have high skill to predict precipitation and number of extreme precipitation days. The prediction informa tion can provide important information for short-range climatic prediction.
展开▼
