水文过程受众多自然和人为因素影响，决定了其变化的极端复杂性，表现为确定性的动态规律与不确定性的统计规律共存。开展水文系统的不确定性研究，对进一步加强水文变化规律的认识具有十分重要的意义。同时，开展不确定性研究，对水资源实行科学规划、管理与决策，以及对环境实施有效保护控制措施，意义重大。如果没有对模拟或预测结果中的不确定性进行深入分析，必将无法充分提供为决策者做出客观最优决策所需要的信息，同时也难以满足决策者对风险的信息需求。本文以黄河中游为例，选择了桥头水文站为目标观测站点，分别以时间序列方法和Bayesian理论对其1956年至1987年的日径流数据开展了不确定性研究。同时对比分析了两种方法在进行不确定性分析时各自的优劣。主要的研究内容及结论归结如下：（1）在对国内外相关文献进行综述的基础上，首先对不确定性的定义进行了总结，同时提出了本文所研究的不确定性的定义，即认为度量参数与预测值的不确定性即是确定参数与预测值在一定信度下置信区间的大小；其次归纳总结了水文不确定性的形成原因及分类；再次概括了国内外进行水文不确定性分析所使用的主要理论与方法；最后提出了当今水文不确定性研究还存在的问题。（2）对黄河流域的水文、气象、地形进行了概括描述，总结了黄河流域的水文特征。（3）考虑到不确定性分析事实上是对方差进行估计，因此选用了时间序列模型中能够精确估计方差的GARCH模型进行建模。首先，对序列进行了剔除季节因素的处理；其次，对处理后的序列建立了传统的ARMA模型；再次，在ARMA模型的基础上，建立了GARCH模型对残差的方差进行了修正；最后，以桥头水文站1956年至1987年日径流数据为例进行了应用验证；研究结果表明，与传统ARMA模型相比，在不影响信度的情况下，GARCH模型能够更加精确的预测置信区间，从而为不确定性分析和风险分析提供更加可靠的基础。（4）水文过程并不是一个独立的过程，目标站点流量与其周边站点流量显然有一定联系，因此通过使用Bayesian方法希望将周边站点的信息提取出来，用于目标站点的不确定性分析中。本文以每年1月31日流量为例进行分析。首先将目标站序列与Weibull分布进行了拟合，得出序列符合Weibull分布的结论；然后再利用周边站点的序列，拟合得到了Weibull分布参数的概率密度函数；其次以Bayesian MCMC方法为基础，采用Metropolis-Hastings算法计算了目标站点的后验概率分布函数；再次将由Bayesian方法得到的结果与MLE方法的结果进行了对比分析；最后将该方法应用于全年，得到了目标站点全年的不确定性分析结果。（5）将GARCH模型与Bayesian方法进行了对比分析，结果显示GARCH模型的预测区间在汛期较大，而Bayesian方法的结果有一定程度的飘移。

The hydrological system is extremely complex. The hydrological process in a region is mainly determined by natural conditions and human activities on the region. Both deterministic and stochastic hydrological processes often coexist in the system. Carrying out research on the uncertainty of hydrological system is very important for further understanding the hydrological variation. And it is useful for Water resources scientific planning, management and decision making. Also, it is important for environmental protection. If there is no thorough analysis on the simulation or predict result, the necessary information for decision maker to making objective optimum decision can’t be provided. And it will be difficult to meet the information needs of decision-making on the risk.This paper takes the Yellow River as an example. The Qiaotou hydrological station is chosen as the target observation station. The time series method and Bayesian method are used to carry out the research on the daily runoff data. And this paper made a comparative analysis of the two methods in the uncertainty analysis of the advantages and disadvantages of each. The main research contents and conclusions summarized as follows: (1) On the basic of summarizing the domestic and foreign literatures, the definition of the uncertainty is summarized. And the definition of uncertainty in this paper is provided. In this paper, meriting the uncertainty of the parameters and the predictive value is to determine the confidence interval size of the parameters and the predictive value, under a certain confidence level. Then the causes and classification of the hydrological uncertainty is summarized. Then the main theory and method on hydrological uncertainty analysis are summarized. At last, the existing problem is presented. (2) The hydrological, meteorological and topographic of the Yellow River basin are summarized.(3) Considering Uncertainty analysis is in fact the variance estimation, so the GARCH model, which can estimate the variance accurately, is used. First, the seasonal factors in the sequence are removed. Second, the traditional ARMA model is established. Then, the GARCH model is used to correct the residual. At last, the daily runoff data in 1956-1987 of Qiaotou Hydrological Station is taken to be an example. The result shows that, compared to the traditional ARMA model, GARCH model have the ability to predict more accurate confidence intervals under the same confidence level. (4) Hydrological process is not an independent process, and the runoff of the target station and its surrounding station apparently linked to a certain extent. So for extracting the information of the surrounding station, the Bayesian method is used. In this paper, the annual runoff in January 31 is taken an example. First, the series are fitted as Weibull distribution, and the result shows the series follows the Weibull distribution. Then thought the series of the surrounding station, the probability density function of the parameters of the Weibull distribution is fitted. Second, on the basic of Bayesian MCMC method, choosing the Metropolis-Hastings algorithm, the posterior probability distribution function of the target station is calculated. Third, the comparison and analysis are taken between the result of Bayesian and the result of MLE. At last, this method is applied to the year, and the result of the uncertainty analysis of the whole year is got. (5) The GARCH model and Bayesian method are discussed. The result shows GARCH model prediction intervals in flood season is larger, and Bayesian method results in some degree of drift.