基于模式误差分布特征的中国夏季旱涝预测可信度研究

Predictive reliability of summer precipitation in China based on error distribution of numerical model

基于美国气候预报中心组合降雨分析资料对国家气候中心海气耦合模式多年夏季降水预报误差进行统计分析,以揭示模式预报误差的分布规律及特征.研究发现,模式预报误差分布特征满足正态分布,基于误差分布特征分析模式对中国夏季降水的预测能力,通过误差分布的高斯拟合曲线特征对比模式预报与动力统计方法的预测技巧,进而得到动力统计方法改进误差分布特征的两种类型:1)变幅型改进;2)均值型改进.在此基础上,根据模式误差的统计特征提出了一种新的旱涝预测可信度计算方法,用于定量化地评估动力统计方法在不同地区的预测可信度大小,并应用于实际预测.最后以2012和2013年夏季降水为例,给出了实况、预测以及可信度检验.结果显示该方法能较好地反映出预测的不确定性,较其他检验方法更具针对性,对异常降水的预测可信度反映更为准确,验证了方法的有效性,可进一步提高动力统计方法的预测服务水平.

In this paper, we made some statistical analyses on precipitation prediction errors of coupled global climate model of National Climate Center based on CPC merged analysis of precipitation. Through the statistics of the forecast results of summer precipitation in many years and the errors, we find that the distribution of the model prediction errors satisfies the Gaussian distribution. Based on the Gaussian distribution characteristics, the ability to predict the mode for summer precipitation in China can be analyzed and compared. The forecast performance of the dynamic-statistics scheme is significantlybetter than that of system error correction schemeof the model. The dynamic-statistics optimal combination of factor revise is corrected, relative to error distribution pattern of the system revised forecast, and there are obtained two kinds of error distribution characteristics： 1） improvement of the amplitude; 2） improvement of the displacement.Based on the statistical characteristics of the prediction error, measure of the credibility of model forecast results is proposed to quantitatively assess the credibility of the numerical model forecast results in different regions. Taking the summer precipitations in 2012 and 2013 for example, we analyze the credibility of actual observations, prediction and drought or flood level. Comparing actual observations and prediction, we find that there is a good consistency where the credibility of them is larger, while there are many bad predictions in the region where the credibility of them is smaller.What is more, compared with other reliability test method, the credibility of the abnormal precipitation forecast by this method is more accurate, showing the effectiveness of the method.