基于重组降水集合预报的洪水概率预报

The Probabilistic Flood Prediction Based on Implementation of the Schaake Shuffle Method over the Huaihe Basin

采用条件亚正态模型方法,生成了具有包含不同可能性的降水集合预报。为了保持各子流域降水集合预报变量之间的空间相关性,采用集合预报重组方法对降水集合预报进行重新排列。使用重组后的降水集合预报驱动水文模型,实现了淮河上游大坡岭一息县、淮河上游息县一王家坝和汝河一洪河上游3个子流域的12次洪水过程的洪水概率预报,并对1988年9月7日和1991年7月31日两次洪水概率预报进行个例分析。结果表明:相对于单一确定性预报,通过条件亚正态分布模型生成降水集合预报后,再经过Schaake洗牌法空间相关性重新组合的降水集合预报,捕捉洪峰出现时间和流量的能力更强。对洪水概率预报来说,降水概率预报更能达到对未来的水文事件进行最大可能估计的目的,并尽可能综合了降水预报不确定性因素,同时也说明维持变量原有的空间相关特征对于降水概率预报具有重要意义。

Daily precipitation records of 19 rain gauges over the Huaihe Wangjiaba-Dapoling catchment and single-value forecasts of 24-hour cumulative precipitation of the Global Forecast System （GFS） with lead time up to 14 days from 1 January 1981 to 31 December 2003 are employed to construct a probability forecast model which can generate ensemble forecast based on conditional meta-Gaussian distribution. Several single-value forecasts could be computed by this model using forecasts of the GFS for daily mean areal precipitation （MAP） and cumulative MAP for each lead time （1--14 days） over 3 sub-catchments in the Huaihe Basin. Then a method is implemented to reorder the ensemble output to recover the space-time variability in precipitation, namely Schaake shuffle method. Ensembles are then reordered to match the original order of the selection of historical data. Using this approach, the observed inter sub-catchments correlations, intervariable correlations, and the observed temporal persistence are almost entirely recovered. This reordering methodology is applied in recovering the space-time variability in modeled streamflow for twelve flood processes over the Huaihe Basin. Results demonstrate that the observation of discharge is included in the interval between the 5th percentage and the 95th percentage forecasts of discharge that is generated by MAP ensemble forecasts which is calculated from the conditional meta-Gaussian distribution model and Schaake shuffle. Several members can capture the flood peak flow and the corresponding peak time. Using approach of Schaake Shuffle, sub-catchment correlations of each ensemble member forecasting could be recovered, which are closer to the observation. A test of flood forecasting result from precipitation probability forecasts of conditional meta-Gaussian distribution model and Schaake shuffle for the stream between Dapoling to Wangjiaba Hydrologic Station is carried out. It shows that MAP ensemble forecasts can provide the maximum estimation of possibility of the future hydrologic events for flood forecasting comparing to the single-value MAP forecast of GFS model. And a comprehensive interval which includes the factor that can lead to hydrologic uncertainty is also given.