中文摘要：

将点源作为未知参数,结合一种新的马尔科夫链蒙特卡罗（MCMC）算法——延迟拒绝适应性Metropolis算法（DRAM）,对钱塘江支流东阳江许村至义东桥河段的化学需氧量、氨氮、饱和溶解氧等3种指标的水质模型进行了贝叶斯参数估计.DRAM算法兼有延迟拒绝算法和适应性Metropolis算法的优点,且稳定收敛速度更快.基于抽样得到的马尔科夫链,对参数和模型误差项的后验分布进行了量化,并实现了点源的不确定性反演.用这个不确定性模型对污染物质量浓度的后验分布进行模拟,表现了良好的拟合效果.基于马尔科夫链,可对各类情景（如不同的水温、流量或点源排放情况）下的污染物超标风险进行直观的分析和预测,也易于实现敏感性分析.研究结果能帮助管理者制定不同水期的减排和调水风险决策,为钱塘江流域的水污染风险管理提供支持.

英文摘要：

By regarding point sources as unknown parameters, a new Markov chain Monte Carlo （MCMC） algorithm--delayed rejection and adaptive Metropolis （DRAM） was used in Bayesian estimation of the chemical oxygen demand,ammonia nitrogen and dissolved oxygen coupled model in the reach from Xucun to Yidong Bridge, Dongyang River, one tributary of Qiantang River. DRAM has the advantages of both delayed rejection and adaptive Metropolis, and shows more efficiency while guaranteeing stable convergence. Based on the sampled Markov chains, the uncertainties of parameters and model errors were quantified by posterior distributions, and the point sources were successfully inversed. The coupled stochastic model shows a good fit when used in modeling the posterior distributions of pollutant concentration. With some controllable and hydrodynamic variables such as pollution discharge, flow, or water temperature varying within certain ranges, the risk of pollutant concentration exceeding national water quality standards under a variety of scenarios could be calculated, which made the sensitivity analysis easily implementable. The final results can provide multiple reduction or water diversion plans in different seasons to reduce water pollution risk, and promote the risk-based manag＇ement of water pollution in Qiantang River Basin.