Predictability of extreme floods

第二阶段研究的总体目标仍然是在不同的空间和时间尺度下，研究极端洪水的可预测性，作为其集水区和气候驱动因素的功能。洪水预测在这里被定义为对未观测到的地点和/或未来时间段的洪水概率分布的估计。可预测性则是指洪水预测的不确定性的量化和归因。可预测性被扩展到对多变量和非平稳洪水概率分布估计的不确定性的量化和归因。将提出三个工作包。首先，考虑数据的非平稳性，处理多变量洪水特性的可预测性。基于单变量数据的区域洪水频率分析将扩展到多变量分析，即考虑洪峰流量之外的洪水量和持续时间。二是利用考虑非平稳性的雷达数据探索随机天气生成。在第一阶段研制的天气发生器将进一步利用气象雷达资料，用以缩小未来气候的尺度。它将被应用于生成降雨和气候情景，用于第三个工作包中基于模型的洪水模拟。这是研究考虑模型非平稳性的多变量洪水特性的可预测性。从模型预测洪水也将扩展到分析多变量和非平稳洪水频率分析。三个工作包的结果可能最终产生一个新的广义统计模型，用于直接根据集水区和气候条件下的降雨频率预测洪水频率。

The overall goal of the proposed research in the second phase will remain the investigation of the predictability of extreme floods as function of their catchment and climate drivers considering different space and time scales. Prediction of floods will be defined here as the estimation of flood probability distributions for unobserved locations and/ or future time - periods. Predictability refers then to the quantification and attribution of the uncertainty of this flood prediction. Predictability is extended to the quantification and attribution of the uncertainty of the estimation of multivariate and non-stationary flood probability distributions. Three work packages will be proposed. The first is dealing with the predictability of multivariate flood properties considering non-stationarity from data. The data based univariate regional flood frequency analyses will be extended to multivariate ones, i.e. taking into account flood volume and duration beside the peak flows. The second one is exploring stochastic weather generation using radar data considering non-stationarity. The weather generator which has been developed in phase one will be further advanced by utilizing weather radar data and employed to downscale future climate. It will be applied to generate rainfall and climate scenarios for the model based flood simulations in the third work package. This is investigating the predictability of multivariate flood properties considering non-stationarity from modelling. Flood prediction from modelling will also be extended to the analyses of the multivariate and non-stationarity flood frequency analyses. The outcomes of the three work packages may eventually result in a new generalized statistical model for prediction of flood frequencies directly from rainfall frequencies conditioned on catchment and climate.