Abstract. Assessing the risk of a historical-level flood at a large scale is essential for regional flood protection and resilience establishment. Due to limitations on the spatiotemporal coverage of observations, the risk assessment relies on model simulations thus is subject to uncertainties from various physical processes in the chain of the flood frequency analysis (FFA). This study assessed the FFA performance as well as the uncertainties with different combinations of FFA variables (river water depth and water storage), fitting distributions and runoff inputs based on the flood characteristics estimated by a global hydrodynamic model CaMa-Flood. Results show that fitting performance is better if FFA is conducted on river water depth and if Wakeby function is selected as the fitting distribution. Deviations in the runoff inputs are the main source of the uncertainties in the estimated flooded water depth based on point analysis. This deviation is relevant to the model ability to reproduce the mean state of annual maximum flood extent and it is almost homogeneous for different flood return period. The uncertainty resulted from fitting distributions increases from the regular period to the rarer floods. The regional investigation of high-resolution inundation area over the lower Mekong River basin shows similar statistics as the point analysis, implying a large uncertainty with 20 % deviation in the total inundation area between different runoff inputs. Regional validation of the CaMa-Flood with two other flood hazard maps proves the reliability of the inundation in space and values. Global analysis on the floodplain water depth implies an increasing contribution of uncertainties in fitting distribution to the total uncertainties for rarer floods in almost all land grids. While the changes in contribution of uncertainties in runoff inputs differentiates in regions. The much higher contribution of runoff uncertainty for rarer floods in wet/flat regions necessitates special attention on rainfall-runoff model calibration (or runoff bias correction) if gauge discharge observations are available. Different adaptions to the large floods are needed for regions with different flood water depth and with different inundation agreements among simulations.

中文翻译：

---

水动力模型模拟洪水频率分析的不确定性

摘要。大规模评估历史洪水的风险对于区域洪水保护和恢复力建设至关重要。由于时空覆盖范围的限制，风险评估依赖于模型模拟，因此受洪水频率分析（FFA）链中各种物理过程的不确定性的影响。这项研究根据全球水动力模型CaMa-Flood估算的洪水特征，评估了FFA的性能以及FFA变量（河流水深和蓄水量），拟合分布和径流输入的不同组合的不确定性。结果表明，如果在河水深处进行FFA且选择Wakeby函数作为拟合分布，则拟合性能会更好。径流输入的偏差是基于点分析估算的淹没水深的不确定性的主要来源。该偏差与模型再现年最大洪水范围的平均状态的能力有关，并且对于不同的洪水恢复期几乎是同质的。拟合分布导致的不确定性从正常时期到罕见的洪水增加。湄公河下游流域高分辨率淹没区域的区域调查显示，统计数据与点分析相似，这意味着不确定性很大，不同径流输入之间的总淹没区域有20％的偏差。CaMa-Flood的区域验证与其他两个洪水灾害图一起证明了淹没在空间和价值上的可靠性。对洪泛区水深的全球分析表明，拟合分布中的不确定性对几乎所有陆地网格中罕见洪水的总不确定性的贡献越来越大。径流输入不确定性贡献的变化在不同地区之间存在差异。如果可以使用量表流量观测，则对于湿润/平坦地区的罕见洪水，径流不确定性的贡献要大得多，因此必须特别注意降雨径流模型的校准（或径流偏差校正）。对于模拟中不同洪水深度和不同淹没协议的地区，需要对大洪水进行不同的适应。径流输入不确定性贡献的变化在不同地区之间存在差异。如果可以使用量表流量观测，则对于湿润/平坦地区的罕见洪水，径流不确定性的贡献要大得多，因此必须特别注意降雨径流模型的校准（或径流偏差校正）。对于模拟中不同洪水深度和不同淹没协议的地区，需要对大洪水进行不同的适应。径流输入不确定性贡献的变化在不同地区之间存在差异。如果可以使用量表流量观测，则对于湿润/平坦地区的罕见洪水，径流不确定性的贡献要大得多，因此必须特别注意降雨径流模型的校准（或径流偏差校正）。对于模拟中不同洪水深度和不同淹没协议的地区，需要对大洪水进行不同的适应。