【作者】 金树权；

【导师】 吕军；

【作者基本信息】 浙江大学 ， 农业资源利用， 2008， 博士

【摘要】 水是人类赖以生存和发展的物质基础，饮用水安全更是影响人体健康和国计民生的重大问题。在我国南方地区，伴随着社会经济的快速发展和城市化进程的迅速推进，许多河流的中下游河段由于点源和面源污染加剧而丧失了饮用水功能，河源区段的水库已经成为目前最重要的城市集中式饮用水供水水源。因此，对水库水源地的保护展开系统研究，具有深远而现实的意义。水库水源地的水质变化规律的研究是水源地保护的基础，而水质模拟和预测是研究水库水源地的水质变化规律的重要途径，也是建立科学合理的水源地污染物排放控制方案的主要理论依据。本文在总结国内外相关研究的基础上，以浙江省湖州市老虎潭水库流域为研究对象，通过对流域内河流进行周年的水文水质监测，协同当地自然、经济和社会状况的调查统计，围绕水源地水质模拟预测的关键问题，对老虎潭水库水源地现状水环境质量进行分段、分期评价，分析主要污染物的形态组成和各污染指标的时空变异规律；利用SWAT2002对整个老虎潭水库集水区进行子流域分区，分类统计各子流域主要污染物的投（排）放量：在综合考虑点源、面源和环境背景值对水源地河流水质影响的基础上，建立了适合于水库水源地的源头溪流一维水质模型，并利用NSGA-Ⅱ多目标遗传算法率定模型的关键参数-面源污染各污染源不同污染物的入河系数；同时，选择合适的水库水质模型，对不同水文和排污条件组合下的河流-水库系统的水质进行联合模拟预测；最后，利用Monte-Carlo法对源头溪流一维水质模型进行不确定性和敏感性分析，研究模型的输出结果与输入参数之间的响应关系，识别模型不确定性的主要影响因素，为模型的改进和完善提供理论依据。通过本项目的研究，建立了较为完整的水库水源地水质模拟预测和模型不确定性分析的基本理论和方法体系，并对湖州市老虎潭水库源头区河流-水库系统的水质进行了模拟预测分析，为水库和水源地水质保护提供科学的依据和可靠的基础。本文的工作成果和主要结论如下：1)通过对老虎潭水库流域内不同区域河流12个监测断面逐月连续的水文水质监测（2007年1月至12月），协同当地的自然、经济和社会条件的调查统计，采用单因子评价和基于熵权模糊综合评价相结合的方法，对河流水质进行分段、分期的分析和评价，结果表明：老虎潭水库流域河流水质总体情况良好且水质浓度相对稳定，目前流域内所有河流断面都符合或优于Ⅲ类水质标准，符合集中式生活饮用水地表水水源地二级保护区标准。2)流域河流水质的主要污染因子是总氮（TN）、总磷（TP）、氨氮（NH₃-N）和有机物污染综合指标高锰酸钾指数(COD_Mn)，营养物质污染和有机污染是该流域应重点关注的问题。农村生活、畜禽养殖和农业面源等面源污染源是该流域的主要污染源，来自面源污染源的污染物投（排）放量量占污染物总投（排）放量95％以上。其中，农田施肥和畜禽养殖是总氮总磷前二位的污染来源，两者占到总氮总磷排放量的80％以上；畜禽养殖、农村生活源是氨氮最主要的污染来源，两者分别占到氨氮总排放量的45．52％，30．70％；有机物污染的主要来源则是畜禽养殖和农业面源，两者分别占总排放量的36．57％，34．98％。3)根据源头溪流的污染特点和河流特征，在国内外河流水质模型和面源污染研究的基础上，综合考虑点源、面源和环境值对河流水质的共同影响，建立了源头溪流一维水质模型：4)系统研究了河流污染物综合降解系数、污染物环境背景值、不同污染源月入河量分配等源头溪流水质模型关键参数的取值和率定方法；建立了针对求解面源污染各污染源不同污染物的入河系数的多目标优化模型，基于Matlab利用NSGA-Ⅱ多目标遗传算法求解多目标优化模型的Pareto解集，提出了根据各污染源的产物特点和不同流域之间的差异对Pareto解集进行选择的基本原则和方法。应用这一套方法，对老虎潭水库各子流域的面源污染各污染源不同污染物的入河系数进行了实际求解，取得了令人满意的结果。5)应用本文所建立的源头溪流一维水质模型和参数确定的系统方法，对老虎潭水库流域河流水质进行了模拟验证，结果表明模拟值与实测值吻合良好，平均误差基本控制在±20％之内。6)综合应用源头溪流一维水质模型、完全混合湖库水质模型和Dillon模型对河流-水库系统进行了不同水文条件和排污状况组合下的水质模拟预测。模拟预测结果表明，在现状排污条件下，河流和水库的氨氮和COD_Mn浓度均为Ⅰ-Ⅱ类水质；在点源污染和面源各增加一倍的排污条件下，除大陈水和埭溪两个子流域在枯水年部分的枯水月COD_Mn浓度达到了Ⅳ类水质外，其余时期为Ⅰ-Ⅲ类水质，而水库的氨氮和COD_Mn浓度可保持在Ⅰ-Ⅱ类水质水平。在现状条件下，河流TP浓度能保证Ⅱ-Ⅲ类水质水平，但由于水库氮磷水质标准严于河流相应标准，水库的TN和TP浓度在丰水年和平水年为Ⅲ类水质，而在枯水年将达到Ⅳ类水质标准水平。如果能同时削减50％的点源污染负荷和26％的面源污染负荷，水库的TN和TP浓度在各典型水文年条件下都能达到Ⅲ类水质水平。7)利用Monte-Carlo法对源头溪流一维水质模型在埭溪2007年各水期TN的模拟结果进行了不确定分析，结果表明综合降解系数k和点源污染负荷q_i对模型输出结果的不确定性影响较小，河流流量Q_e和污染物环境背景值C_b对模型输出结果的不确定性影响相对较大，而模型结构不确定性对模型输出结果的不确定性影响最大。鉴于模型结构不确定性是源头溪流一维水质模型最大的不确定性来源，为了降低模型输出结果的不确定性应重点加强面源污染月入河量分配的研究。在现状条件下，埭溪2007年各水期的TN模拟输出结果呈正态分布，其90％的输出结果基本落在均值±20％范围之内。通过降低模型各方面不确定性因素，可以有效降低模型模拟结果的不确定性，如果能将现有条件下模型各不确定性因素降低50％，那么模型90％输出结果的相对误差将从＜±20％降低到＜±10％。本文的主要创新点为：1)将信息学中“熵”的概念应用于模糊综合水质评价中权重的确定，改进了水质模糊综合评价方法。基于熵权模糊综合评价法，一方面避免了单因子水质评价的片面性，利用模糊数学隶属度的概念体现了不同评价因子对水质的综合影响，另一方面引入“熵”的概念使得权重的设计更为科学合理同时，体现了多个评价对象之间的相互联系。2)根据源头溪流的污染特点和河流特征，在国内外河流水质模型和面源污染研究的基础上，综合考虑点源、面源和环境背景值对河流水质的共同影响，提出并建立了源头溪流一维水质模型。该模型克服了简单河流水质模型单纯考虑点源污染影响和大型流域综合水质模型需要大量基础数据与参数的缺点，为以面源污染为主的水源地河流的水质模拟预测提供了新的理论体系和方法。3)通过在同一流域不同时段建立针对求解不同面源污染入河系数的多目标优化模型，利用NSGA-Ⅱ多目标遗传算法求解得到模型的Pareto解集，并提出了根据各污染源的产污特点和不同流域之间的差异对Pareto解集进行入河系数选择的基本原则和方法。此方法克服了面源污染入河系数难以确定的瓶颈问题，为面源污染的研究提供了新思路和新方法。4)结合源头溪流一维水质模型，完全混合湖库水质模型和Dillon模型，实现了不同水文条件和不同排污条件组合下河流-水库系统水质的联合模拟预测，并提出了相对合理的流域污染物控制对策和方案。5)将模型不确定性来源分为参数不确定性、输入数据不确定性和模型结构不确定性的基础上，利用Monte-Carlo法对建立的源头溪流一维水质模型进行了不确定性分析研究，分析了输出结果的取值范围与概率分布，并利用”龙卷风图”识别模型不确定性的主要因素，为模型的进一步改进提供理论依据。更多还原

【Abstract】 Water is the material foundation of mankind’s survival and development, and thesafety of drinking water is even more important to people’s health and livelihood. Insouthern China, with rapid socio-economic development and urbanization, water inthe middle and lower reaches of many rivers has lost its function for drinking due toincreasing point-source pollution and non-point-source （NPS） pollution, andreservoirs became the most important municipal centralized water supply sources.Therefore, systematically research on protection of headwater area of reservoir hasprofound and significant meanings.Variation of water quality is the foundation for the water quality protection inheadwater area of reservoir. Modeling and prediction of water quality is key approachfor the protection research, and the predicted results are also the main theoretical basisfor creating scientific and reasonable emission control project. The aims of the presentstudy were to establish a methodological and technical system for modeling andprediction of water quality in headwater area of Laohutan reservoir watershed inHuzhou, Zhejiang Province, Southern China. Water quality throughout the Laohutanreservoir watershed were monthly measured continuously for the whole year, whichcombined with investigation of natural, social and economic conditions. Based on theanalysis of overall datum, water environmental quality in Laohutan reservoirwatershed was evaluated in each reach and period, and spatio-temporal variations andthe composition analysis of main pollutants were also studied. Sub-watersheds wereseparated by means of SWAT2002 in Laohutan reservoir, and each produce flux ofmain pollutant were gained in each sub-watershed. Based on integrative considerationof the effects of point-source pollution, NPS pollution and environmental backgroundvalues on river water quality in headwater area, and one-dimensional river waterquality model for headwater area （1-D RWQHA model） was founded, and the keycoefficient （export coefficient of NPS pollution） in the model was calibrated usingmulti-objective genetic algorithm （MOGA） based on NSGA-Ⅱ. As a case study, 1-DRWQHA model, Vollenweider model and Dillon model were used to predict the variations of water quality in the river-reservoir system in Laohutan reservoirwatershed. Finally, based on the response relations between input parameters andoutputs, uncertainty and sensitivity on 1-D RWQHA mode was analyzed by usingMonte-Carlo method, and the main uncertainty factors were identified.The main conclusions of the dissertation included:1) Water quality at 12 sampling sites throughout the Laohutan reservoir watershedwere monthly measured continuously for the whole year from Jan 2007 to Dec 2007,which combined with investigation of natural, social and economic conditions. Basedon single index method and fuzzy synthetic method, river water quality wereevaluated, and the results showed that the water quality in all sampling sites were orbetter than GradeⅢI.2) Total nitrogen （TN）, total phosphorus （TP）, ammonia-nitrogen （NH₃-N） andCOD_Mn were the main pollution factors in this watershed, and organic and nutrientpollution was the main pollution problem in this headwaters area. Pollutant produceflux from NPS pollution （rural domestic waste pollution, livestock-poultry wastepollution, agricultural NPS pollution） was the main pollution source, and the produceflux from NPS pollution occupied above 95% in total amount. In all pollution sources,agricultural NPS pollution and livestock- poultry waste pollution were the mainsources of TN and TP, which occupied 80% of total produce flux of TN and TP.Livestock- poultry waste pollution and rural domestic waste pollution were the mainsources of NH₃-N, which occupied 45.52%, 30.70% of total produce flux of NH₃-N,respectively. Livestock-poultry waste pollution and agricultural non-point pollutionwere the main sources of organic matter, which occupied 36.57%, 34.98% of totalproduce flux of organic matter, respectively.3) According to the pollution characteristics and river feature in headwater area ofreservoir, 1-D RWQHA model was founded. The model integrative considered theeffects of point-source pollution, NPS pollution, and environmental backgroundvalues on river water quality. The denotation of 1-D RWQHA model as follow: 4) Pollutant degradation coefficients, pollutant environmental background values,and the distribution of monthly pollutant export amount from different pollutionsources were calibrated and validated. Multi-objective optimization model wasestablished to solve export coefficient in each sub-watershed based on 1-D RWQHAmodel. Using Matlab the Pareto solution set of multi-objective optimization modelwas solved by means of NSGA-Ⅱ. According to the difference of sub-watershedcharacteristics and the pollution sources feature, proper export coefficients were chosein the Pareto solution set.5) After the calibration of all parameters and input data of 1-D RWQHA model,the model was validated. The validation result showed that the predicted valuesagreed well with the measured values, and the average errors basically within±20%.6) Combining with 1-D RWQHA model, Vollenweider model and Dillon model,water quality of river-reservoir system was synchronized predicted under differenthydrology and emission conditions. The results showed that, in current emissioncondition, the predicted concentrations of NH₃-N and COD_Mn in rivers and reservoirwere all between gradeⅠ-Ⅱwater quality levels. If the emission of point-sourcepollution and NPS pollution were doubled, the predicted concentrations of COD_Mnwere in gradeⅣwater quality for some tributaries in dry hydrologic year, theremainders were in gradeⅠ-Ⅲwater quality, while the predicted concentrations ofNH₃-N and COD_Mn in reservoir maintained gradeⅠ-Ⅱwater quality. In currentemission condition, the predicted concentrations of TP in river s were in gradeⅡ-Ⅲwater quality, and the predicted concentrations of TN and TP in reservoir would be ingradeⅢin flooding and average hydrologic year, but would be in gradeⅣin dryhydrologic year due to the water quality standard of nutrient in reservoir is morerigorous than that in river. If 50% point-source pollution emission and 26% NPSpollution emission were reduced, the predicted concentrations of TN and TP inreservoir would or better than gradeⅢwater quality in all hydrologic years.7) Uncertainty on predicted concentrations of TN in Daixi river in each hydrologic period in 2007 was analyzed by using Monte-Carlo method based on 1-D RWQHAmodel. The results showed that the effects of degradation coefficient k andpoint-source pollution load q_i on model’s outputs were minor, the effects of river flowQ_e and environmental background value C_b on model’s outputs were medium, and theeffect of structural uncertainty on model’s outputs was major. Thus, to diminish theuncertainty of 1-D RWQHA model’s outputs, the research about allocation of monthlynon-point transport flux should strengthen. In current condition, the predicted valuesshowed normal distribution, and the approximately 90% of values within the range of±20% of mean value. The model’s output uncertainty could diminish via decreasedthe uncertainty of model’s uncertainty sources. If the uncertainty of each sourcedecreased 50%, the approximately 90% of values would within the range of±10% ofmean value.The innovated progress of this dissertation included:1) The concept of "entropy" in informatics was applied in the determination of theweight for evaluating indicators in fuzzy synthetic water quality evaluation. Usingfuzzy synthetic evaluation based on entropy not only avoided the one-sidedness ofsingle index method, but also considered the correlation of different evaluatedsubjects. Thus, the results of fuzzy synthetic evaluation based on entropy were morescientific and reasonable.2) According to the pollution characteristics and river feature in headwater area ofreservoir, and combined with current studies about river water quality model and NPSpollution, 1-D RWQHA model integrative considered the effects of point-sourcepollution, NPS pollution, and environmental background values on river water qualitywas founded. Thus, a methodological and technical system for modeling andprediction of river water quality in headwater area of reservoir which dominated byNPS pollution was created.3) Multi-objective optimization model was created to solve export coefficients ineach sub-watershed, and solved by MOGA based on NSGA-Ⅱ. Proper exportcoefficients were solved according to the difference of sub-watershed characteristic and the pollution sources feature. The export coefficients identification methodprovided new idea and new approach for NPS pollution research.4) Combining with 1-D RWQHA model, Vollenweider model and Dillon model,water quality of river-reservoir system was synchronized predicted under differenthydrology and emission conditions, and reasonable emission control project for wholewatershed was founded.5) Sources of model uncertainty include parameter uncertainty, input datauncertainty and structural uncertainty, and the uncertainty of 1-D RWQHA model wasstudied by using Monte-Carlo method. The range and distribution of outputs wereanalyzed, and the main uncertainty factors were identified using "tornado graphs".The results of uncertainty analysis provided theoretical basis for model furtherimprovement.更多还原