基于黄河流域1979—2018年的ERAInterim再分析气候与水文数据，以及CMIP5中10个气候模式下3种典型排放情景（RCP2.6、RCP4.5、RCP8.5）的全球气候变化数据，采用离散数据的处理方法，建立黄河流域贝叶斯网络模型，推断黄河流域近40余年来气候要素对径流的影响概率，预测黄河流域未来径流量。结果表明：1979—2018年黄河天然径流量呈减小趋势，基于贝叶斯网络分区间概率预测预报的径流量也呈减小趋势；黄河流域的不同区间（低、中、高）径流量对气候的敏感程度不同，但径流始终与降水相关性最高；在RCP2.6情景下，黄河流域未来20年、60年的径流量为585.50亿、588.57亿m3；在RCP4.5情景下，其值为585.42亿、587.53亿m3；在RCP8.5情景下，其值为593.50亿、585.11亿m3。

With the development of social economy,the demand for water resources of the Yellow River basin is increasing.And most areas of the Yellow River basin are in arid and semiarid areas and the ecological environment is fragile,aggravating the sensitivity of the water resources system to climate change.Although many researches have been conducted on the impact of climate factors on hydrological phenomena,the causality and the probability of interaction between climate factors and runoff are still very vague.Besides,in the study of hydrological forecast,the water balance model,the wavelet analysis,the neural network,and the fuzzy inference method can merely provide a deterministic forecast of hydrological process,while they can not quantitatively describe the uncertainty of forecast results.Based on the correlation and uncertainty of climate and hydrological systems,Bayesian network (BN) is then used to quantify the impact of climate factors on runoff and forecast the future runoff in the Yellow River basin. Based on expert knowledge bases and other scholars′ research results on the relationship between climate and runoff in the Yellow River basin,six climate factors including temperature,pressure,wind speed,specific humidity,evaporation,and precipitation were determined to form the variable node of BN with runoff,and the BN model of climate runoff was constructed by the Netica.The ChiMerge method was used to discretize the ERAInterim reanalysis of climate and hydrological data from 1979 to 2018 into three sections.After the determination of network structure and training data set,the conditional probability table of each node can be obtained by the maximum likelihood estimation,and the Bayesian influence probability between variables can be calculated by the variable elimination method.In the BN model for predicting runoff,all the data in the prediction model is divided into twelve intervals to improve the prediction accuracy.The ERA reanalysis data in years of 1979-2018 is used as the training set.The history climatic data of CMIP5 ten climate models in years of 1979-2005 is used as a validation set.The reliability verification of the model is established by comparing the predicted range and trend of runoff in years of 1979-2005 with ERA runoff data over the same period.Finally,the future runoff data of the Yellow River basin is predicted by the climate variables of several typical concentration emission scenarios (RCP 2.6,RCP 45,and RCP 8.5) in years of 2006-2080. From the climate runoff Bayesian network,it is found that any change in the state of any variable can cause other variables to change.For example,as the temperature state from low to high,the probability of high state evaporation and precipitation can increase,and the probability of high state runoff can also increase,indicating temperature,evaporation,and precipitation are positively correlated with runoff.The relationship among other climatic factors and runoff can also be found that specific humidity is positively correlated,pressure and wind speed are negatively correlated,and precipitation has the closest relationship with runoff.From years of 1979-2018,the natural runoff of the Yellow River showed a decreasing trend,and the runoff based on the probability prediction from BN also showed a decreasing trend.In the RCP26 scenario,the Yellow River basin′s runoff in the next twenty and sixty years will be reached 58.55 billion m3 and 58.857 billion m3.While in the RCP4.5 scenario,the Yellow River basin′s runoff in the next twenty and sixty years will be reached 58.542 billion m3 and 58.753 billion m3;and in the RCP8.5 scenario,the associated values will be reached 59.35 billion m3 and 58.511 billion m3. Climate change is of great significance to the change of surface runoff.The climaterunoff BN constructed in this paper is a network of uncertain relationships between climate and hydrological elements.It explores the climatic reasons for the reduction of runoff and conducts midand longterm predictions of future runoff.It is concluded that the main reason for the decrease in runoff is the decrease in precipitation,and the specific humidity is a key climatic element affecting precipitation.The future runoff forecast of the Yellow River in the next sixty years is estimated to be around 58.7 billion m3 by BN.