基于Tweedie分布的日降水量统计降尺度模型

基于Tweedie分布的广义线性模(generalized linear model,简称GLM),并结合Kriging模型,发展了日降水量统计降尺度的GLM-Kriging模型.首先用GLM拟合研究区域内日降水量与数值模式输出的影响局地降水的物理量之间的关系,日降水量的空间相关性反映在模型的残差中;然后用Kriging模型来拟合GLM的随机化百分位残差(randomized quantile residuals,简称RQ残差).结合NCEP再分析资料应用于2007年7月沂沭泗流域的42站日降水观测,结果表明GLM-Kriging降尺度模型较好地还原了主要降水过程,整体上取得了较高的准确度,可用于气候变化影响评估或数值天气预报产品的释用,还可进一步扩展为日降水量的时空统计模型.

STATISTICAL DOWNSCALE MODEL FOR DALLY PRECIPITATION BASED ON TWEEDIE DISTRIBUTION

Based on generalized linear model (GLM) of Tweedie distribution and Kriging model, a combined GLM-Kriging model for statistical downscaling of daily precipitation has been developed. GLM is first fitted to represent relations between daily precipitations and corresponding model output variables influencing rainfall in studied area; spatial dependency of daily precipitation is inherited by model residuals. A Kriging model is then fitted to randomized quantile residuals of fitted GLM. The model has been applied to 42 daily precipitation observations for the Yishusi Basin in July 2007 together with NCEP Reanalysis II data. Data show that the GLM-Kriging downscaling model successfully retrieved main rainfall processes during that month and obtained high prediction accuracy. The GLM-Kriging downscaling model can be used for climate change impact assessment and model output statistics of numerical weather product, and can be further expanded into a spatial-temporal model for daily precipitation.