Evaluation of Flood Prediction Uncertainties and its Application to Ungauged Basins

为了获得设计洪水，首先确定降雨时空分布随预定回归期大小的变化规律，然后利用径流模型将降雨分布转换为河流流量。在这里，给定的降雨分布有不同的模式，径流模型有局限性。因此，估计的河流流量包括来自各种来源的不确定性。因此，评估设计洪水的预测不确定性，明确不确定性的来源，对确定设计洪水具有重要意义。洪水预测的不确定性主要来自于输入的不确定性（时空降水观测不足）、模型结构的不确定性（水文过程认识不足）和模型参数的不确定性（模型参数辨识不足）。这些不确定性主要源于水文观测不足。为了解决这个问题，加强观测系统是必要的，但由于预测的目标是百年一遇的洪水，这是不现实的。因此，需要开发一种充分利用已测量流域信息的未测量流域水文预测方法。本研究针对未测量流域的设计洪水进行了以下研究：1)建立基于降雨极值尺度特征的强度-持续时间-频率关系；2)建立反映降雨极值特征的降雨时间序列生成方法；3)建立基于预测不确定性的降雨-径流模型性能评估方法。

To obtain design flood, at first some patterns of spatiotemporal rainfall distributions with the magnitude of pre-determined return period is determined, then the rainfall distributions are converted to river discharges using a runoff model. Here, the given rainfall distributions have various patterns and the runoff model has limitations. Thus, the estimated river discharges include uncertainties which come from various sources. Therefore, it is important to evaluate the predictive uncertainty and clarify the sources of the uncertainty to determine design flood.The predictive uncertainty in flood prediction mainly comes from 1) input uncertainty (insufficiency of time and space rainfall observation), 2) model structure uncertainty (insufficiency of the understandings of hydrologic processes), and 3) parameter uncertainty (insufficiency of model parameter identification). These uncertainties essentially stem from insufficient hydrologic observations. To solve the problem, to enhance the observation system is essential, but not practical because the target of prediction is flood which happens e.g. once a hundred years. Therefore, it is requested to develop a hydrologic prediction method for ungauged basins which fully utilize the information at gauged catchments. In this research, followings are conducted to obtain design flood at ungauged basins :1) Development of a intensity-duration-frequency relationships based on scaling characteristics of rainfall extremes2) Development of a rainfall time series generation method which reflects characteristics of rainfall extremes3) Development of a method to assess the performance of rainfall-runoff models in terms of predictive uncertainty.