中小河流洪水预报的不确定性控制与动态评估研究

Flash flood occurs frequently in the medium and small river basins in China and usually leads to severe damages. In these regions, the existing network of hydrological stations is sparse and the historical observations are extremely short for a large number of catchments. The accuracy of flood prediction is too low to fulfill the requirement of flood prevention. The uncertainties of prediction, model parameters and the parameter estimation in data-limited regions are the main reasons for the poor performance of flood forecasting. Based on the rainfall-runoff data series and the simulation of distributed hydrological models, this study aims to analyze the dominating factors that accounts for the uncertainty of flood prediction，investigate the error sources and uncertainty reducing methods of quantitative precipitation forecast and parameter estimation, quantitatively evaluate the dynamic uncertainty of flood forecasting in medium and small river basins. Firstly, the accuracy of typical quantitative precipitation forecast products will be evaluated according to historical rainstorms and the suitable quantitative precipitation forecast models for every study region will be identified. The second focus of this study is to investigate the uncertainty of model parameters, as well as the transferability of model parameters in time and space. Common parameters will be calibrated based on the simultaneous calibration for a number of similar catchments. The idea of reducing uncertainty in parameter estimation for data-limited basins will be proposed. Finally, based on the joint distribution function of uncertainties in precipitation and model parameters, the dynamic quantitative analysis method will be explored to evaluate the uncertainty in flood forecasting. In the application of flood forecasting in the medium and small river basins, the time-varying multiple uncertainty could be reduced and comprehensively evaluated. This project shows great scientific significance and application value as it offers approaches for promoting flood prediction skill in the medium and small river basins.

中小河流洪水频发，灾害损失严重。中小河流站网稀疏，资料匮乏，洪水预报精度低，难以满足防汛实际需求。降水的不确定性、模型参数的不确定性及资料匮乏区参数推衍的不确定性是造成洪水预报精度低的重要原因之一。本项目基于典型流域降雨径流资料和分布式模型模拟，剖析影响洪水预报不确定性的主导因素，研究降水预报、模型参数的误差来源和不确定性降低方法，对中小河流洪水预报的不确定性进行动态定量评估。主要研究包括：基于历史暴雨评估典型定量降水预报产品的精度与不确定性，识别各研究区适用的降水预报模式；探明模型参数的不确定性及时空移植性，研究基于多相似流域同步模拟的共享参数确定方法，提出资料匮乏区参数推衍不确定性控制技术；构建降水和模型参数不确定性联合分布函数，研究洪水预报不确定性动态定量分析方法。实现中小河流洪水预报多元时变不确定性的有效降低和综合评估，提高我国中小河流洪水预报水平，具有重要的科学意义和应用价值。

我国中小河流众多，洪涝灾害频繁发生，已成为我国洪涝灾害防控最突出的短板。洪水预见期短、资料匮乏等问题严重制约了洪水预警预报的水平。极端暴雨的随机性和水文资料匮乏使得中小河流水文模型模拟的不确定性凸显，综合剖析模型输入、模型参数的不确定性对于中小河流洪水预报研究具有十分重要的现实意义。本项目围绕中小河流洪水预报面临的关键问题，构建基于网格的分布式洪水预报模型，剖析预报误差主要来源，探究降低模型输入与模型参数不确定性的有效方法，提升洪水预报的实时性和精准度。.本研究基于网格的分布式洪水预报模型，以不同水文气象分区中小河流为研究区域，构建了示范流域精细化洪水预报方法, 实现了流域内任意网格流量过程的模拟预报，建立了模型产汇流参数与土壤水文常数、土层厚度等流域下垫面特性间的定量关系，实现了模型参数空间分布的估算。基于历史暴雨评估了典型定量降水预报产品的误差分布规律与不确定性，基于QUANT和RQUANT方法进行了降雨产品修正研究，开展了预报降雨产品的区域适用性分析。研究表明，不同流域的降雨预报效果存在一定差异，各产品在修正后大部分检验指标预报精度能够得到提升，在半干旱地区使用ECMWF修正后预报产品，在湿润流域选择NCEP和ECMWF产品和修正方法，能够在一定程序上减少降雨预报误差。基于多目标折衷优化方法和ROPE参数自动率定算法构建了多相似流域模型并行率定及无资料地区参数移植方法，结果表明，该方法能够克服单个流域参数优选过度拟合而空间移植性差的问题，有效提高了无资料地区参数移植稳定性和洪水预报的精度。基于实测降雨信息和不考虑未来降雨、考虑未来降雨、考虑降雨的雨量误差和考虑降雨的时间误差四种实验方法，定量评估了降雨输入不确定性对实时洪水预报的影响，构建了基于贝叶斯理论和极大似然估算方法的中小河流洪水预报不确定性动态分析方法，研究成果能够有效控制洪水预报不确定性，提高预报精度和预见期，为支撑洪水调度决策和减少灾害损失奠定基础。