Generating Calibrated Ensembles of Physically Realistic, High-Resolution Precipitation Forecast Fields Based on GEFS Model Output
Generating Calibrated Ensembles of Physically Realistic, High-Resolution Precipitation Forecast Fields Based on GEFS Model Output
复制标题
基于 GEFS 模型输出生成物理真实的高分辨率降水预报场的校准集合
DOI：
--
复制
发表时间：
2018
影响因子：
3.8
通讯作者：
T. Hamill
中科院分区：
文献类型：
--
作者：
M. Scheuerer;T. Hamill
多变量后处理方法的增强，产生统计校准合奏的高分辨率降水预报字段与物理上现实的空间和时间结构的基础上，从全球环境预报系统（GEFS）的降水预报。校准的边缘分布得到异方差回归方法使用删失，移位伽玛分布。为了生成时空预测字段，最近提出的最小发散Schaake洗牌技术，它选择了一组历史日期，这样的方式，相关的分析字段的边缘分布，类似于校准的预测分布，提出了一个新的变种。该变体在预测网格尺度上执行单变量后处理，并通过导出乘法调整函数并使用它来修改历史分析字段以使其与校准的粗尺度降水预测相匹配来将这些粗尺度降水量分解到分析网格。此外，提出了系综copula耦合（ECC）技术的扩展。构造映射函数，其将每个原始集合预报场映射到高分辨率预报场，使得所得到的缩减集合具有规定的边缘分布。在一个地区，包括俄罗斯河流域在加州的案例研究，这表明，由这两种新技术产生的预测字段有一个物理上现实的空间结构。定量验证表明，它们也代表了亚网格尺度降水量的分布比标准的Schaake洗牌或ECC-Q重排方法产生的预报场更好。
Enhancements of multivariate postprocessing approaches are presented that generate statistically calibrated ensembles of high-resolution precipitation forecast fields with physically realistic spatial and temporal structures based on precipitation forecasts from the Global Ensemble Forecast System (GEFS). Calibrated marginal distributions are obtained with a heteroscedastic regression approach using censored, shifted gamma distributions. To generate spatiotemporal forecast fields, a new variant of the recently proposed minimum divergence Schaake shuffle technique, which selects a set of historic dates in such a way that the associated analysis fields have marginal distributions that resemble the calibrated forecast distributions, is proposed. This variant performs univariate postprocessing at the forecast grid scale and disaggregates these coarse-scale precipitation amounts to the analysis grid by deriving a multiplicative adjustment function and using it to modify the historic analysis fields such that they match the calibrated coarse-scale precipitation forecasts. In addition, an extension of the ensemble copula coupling (ECC) technique is proposed. A mapping function is constructed that maps each raw ensemble forecast field to a high-resolution forecast field such that the resulting downscaled ensemble has the prescribed marginal distributions. A case study over an area that covers the Russian River watershed in California is presented, which shows that the forecast fields generated by the two new techniques have a physically realistic spatial structure. Quantitative verification shows that they also represent the distribution of subgrid-scale precipitation amounts better than the forecast fields generated by the standard Schaake shuffle or the ECC-Q reordering approaches.
影响因子：
5.7
作者：
Roman Schefzik;T. Thorarinsdottir;T. Gneiting
通讯作者：
Roman Schefzik;T. Thorarinsdottir;T. Gneiting