摘要
应用回归诊断方法分析发现 ,梅汛期大到暴雨回归预报模型的残差分布存在着不对称现象 ,这种不对称现象是由高杠杆点所引起。这些高杠杆试验点的残差存在着统计天气预报意义上的不合理性 ,导致了回归系数LS估计的误差 ,从而又引起暴雨预报的误差。针对这些问题提出了大到暴雨的回归诊断预报模型。实例计算说明 ,回归诊断预报模型要优于常规回归预报模型。进一步分析指出 ,梅汛期大到暴雨回归预报模型的不合理性并非个别例子的特殊性所造成 ,而是由模型的数学特点所决定 ,因此大到暴雨的回归诊断预报模型具有普遍意义。大量的试验和多年的业务应用表明 ,回归诊断对提高大到暴雨预报准确率具有明显的效果。
The regression analysis indicates that there are asymmetry characteristics of residual distribution in the Meiyu heavy/torrential rain forecasting model. These phenomena are caused by some Ligh Leverage Cases. There is irrationality in the residuals of these high leverage cases statistically,resulting in Least Square (LS) errors in estimating the regression coefficient,consequently errors of heavy/torrential rain forecasting.A regression diagnosis prediction model for heavy/torrential rain forecasting is proposed. The example analysis shows that this new model is superior to general regression prediction models. The further analysis indicates that the irrationality of the Meiyu heavy/torrential rain regression forecasting model is resulted from its mathematical characteristics,not by chance.
出处
《气象科技》
北大核心
2004年第2期81-85,共5页
Meteorological Science and Technology
基金
绍兴市科委课题 (2 0 0 0 1115 )资助
关键词
回归诊断
暴雨预报
模型
残差分布
高杠杆点
forecasting,heavy/torrential rain,regression diagnosis,residual distribution,high leverage case
