【作者】 顾清；

【导师】 王珂；

【作者基本信息】 浙江大学 ， 农业遥感与信息技术， 2014， 博士

【摘要】 由于降水的时空不均和河流水质的严重恶化,水库已成为浙江省最重要的饮用水源。近年来,随着经济的发展和城市建设的推进,不少水库呈现富营养化趋势,水库水质有显著下降趋势,因此,开展水库水质及流域生态环境研究对更好地保护水库水资源显得十分必要。本研究以遥感、GIS、数据挖掘等技术为手段,从水库水质及营养状况、流域土地利用变化和生态环境安全性等方面开展水库水质及风险评价研究,以期为建立集成地面常规监测、遥感监测、常规统计信息评价的全省饮用水水库综合监管系统、及时掌握饮用水水库的水质演变趋势和风险情况提供理论支持。主要研究内容与结论如下：(1)首先对浙江省30座大中型饮用水水库2001年至2013年的水质和营养状态进行分析,结论如下：2001年至2013年间浙江省水库的水质总体上发生了下降。2001年全部水库达到饮用水Ⅲ类水质要求,至2013年已经有10座水库未达到Ⅲ类水质等级。2001至2013年水库的营养水平总体上从贫营养向富营养演变,富营养化趋势显著。采用2010年TM遥感数据对水库的水质进行定性分析,发现遥感信息的总体变化趋势与地面监测的水质等级变化情况相一致,随着水质等级的下降,第一波段和第四波段的DN值有上升的趋势。(2)使用TM遥感数据和千岛湖实测水质数据,选用常规的统计回归模型和RBF人工神经网络模型两种方法对水体中的叶绿素a,透明度,总氮,总磷含量进行模拟和预测。在回归方法建模中,以丰水期数据为例,得出以(b3+b4)/b2为参数的对数模型为丰水期chl-a浓度的最佳遥感反演模型,以b1/b4为参数的幂函数模型为丰水期透明度的最佳遥感反演模型,以b1*b4为参数的多项式模型为丰水期总氮的最佳遥感反演模型,以(bl+b2+b3)/b4为参数的多项式模型为丰水期总磷的最佳遥感反演模型。在RBF人工神经网络模型建模中,采用Cfs(Correlation-based Feature Selection)方法对遥感参数进行最佳参数集合的选择,以选出的最佳参数为自变量用RBF人工神经网络模型对水质指标进行模拟和预测。对两种模型的精度进行比较后得出,RBF人工神经网络模型相比传统的统计回归模型具有更好的反演精度。(3)对水库流域1995、2000、2005和2010四个时期的土地利用变化情况进行分析。结果显示,1995-2010年间,水库流域内建设用地显著增加,水域小幅增加,林地面积基本保持平衡,而耕地面积有所减少。建设用地扩展速度较快的水库流域主要位于省域范围的北部和东部沿海区域,而相对较慢的水库流域多数位于西南部山林地区。研究期内建设用地的占用主要集中于低海拔和较平坦区域。人口增长和经济发展为建设用地扩张的重要驱动因素,其中人口和GDP同建设用地的相关性较高,工业总产值的作用相比这两个因素较小。2010年人口对建设用地的影响相比1995、2000和2005年明显下降,但是GDP和工业总产值的影响有显著上升,特别是工业产值的影响大幅上升。人口是耕地的最重要影响因子,GDP同耕地面积的相关性较低,工业总产值和耕地面积的相关性不明显。比较水库流域内耕地比例与建设用地比例与水库水质之间的关系,发现随着水质等级的下降,流域平均建设用地比例有升高的趋势,说明建设用地的增加将导致水库水质的下降。虽然流域建设用地扩张速度相对比较缓慢,但水库水质对流域内建设用地增长十分敏感,因此需严格控制流域内建设用地的增加。(4)研究选择包括土地利用、社会经济、水库特征和气候等方面的16个参数作为自变量,以2010年浙江省73座饮用水水源地水库的水质等级作为目标变量,用决策树方法进行建模,并对水库水质进行预测。模型的训练精度达到94.23%,预测精度达到80.95%。人为因素包括工业废水排放、工业产值、GDP、人口密度和土地利用方式是与浙江省饮用水水库水质联系最紧密的影响因子。对参数重要性分析的结果显示工业废水的排放是造成水库水质变化的最重要因素。(5)最后结合模糊层次分析法(FAHP)和集对分析理论,借助遥感和GIS手段,在压力-状态-响应(PSR)框架下对30座水库流域1995年和2010年两个时期的生态环境安全性进行了评价和分析。评价结果显示,从1995年至2010年水库流域生态环境安全性整体上发生了下降,由安全向潜在危险和危险转变。从分指标的评价结果中可知,下降的主要为压力和状态指标,而响应指标略有上升。在空间分布上来看,西南部的安全性整体上要高于东北部地区,安全性发生显著下降的流域主要位于浙江省的中部和东北部地区。更多还原

【Abstract】 Unbalanced precipitation and seriously deteriorated quality of river water determine the most important role that reservoirs played as the drinking water resource in Zhejiang Province in China. Protection of reservoirs water became an urgent task relating to the province’s drinking water security and sustainable economic development. Recently, with the process of economic development and urbanization, a number of reservoirs have been confronted with the problem of water eutrophication and quality deterioration. Thus, it is quite necessary to carry out the researches on reservoir water quality and watershed ecological environment for better protecting the reservoir water. This study evaluates the reservoir water environment from aspects of reservoir water quality, eutrophication status, land use change of watershed and ecological security of reservoir environment, by integrating remote sensing, GIS as well as data mining technology. The main objectives and results are as follows:Firstly, the changes in water quality levels and eutrophication status of30large/medium-sized drinking reservoirs in Zhejiang Province from2001to2013were analyzed. It is concluded that the water quality of these reservoirs declined in general during this period, In2001, all these estimated reservoirs reached class Ⅲ standard for drinking water quality, while there were10reservoirs that were not up to class Ⅲ standard in2013. Generally, the eutrophication status of these reservoirs exhibited the transformation from oligotrophic to eutrophic, with significance in water eutrophication during our estimated period. The results of qualitative analysis to the reservoirs water with TM images (2010) were consistent with the changes of water quality. It is shown that as the decline of water quality levels, the DN value of band1and band4exhibited a rising trend.Based on TM images and the water monitoring data of Qiandao Lake, this study adopted traditional statistical regression model and RBF artificial neural network model to simulate and predict chlorophyll a, transparency, total nitrogen and total phosphorus of the reservoirs water, In the regression model, take the high water period as an example, it is concluded that logarithmic model with parameter of (b3+b4)/b2was the best inverse model for chlorophyll a with remote sensing, exponential function model with parameter of bl/b4was the best inverse model for transparency, exponential function model with parameter of b1*b4and exponential function model with parameter of (bl+b2+b3)/b4were the best inverse model respectively for total nitrogen and total phosphorus In the RBF artificial neural network model, it used Cfs (Correlation-based Feature Selection) technology to select the optimal dataset of remote sensing parameters, and with this optimal dataset as variables to simulate and predict indicators of water quality. Through comparison between the accuracies of these two models, it is concluded that the RBF artificial neural network model was estimated to have better performance.According to analysis of the land use change in the reservoir watershed in the year of1995,2000,2005and2010, it revealed that built-up land increased significantly, water bodies increased slightly, forest area was balanced and cultivated land decreased during this15-year period. Reservoir watersheds that experiencing relatively rapid expansion in built-up lands were observed in the northern and eastern coastal regions of Zhejiang Province, while reservoir watersheds that experiencing relatively slow built-up lands were observed in southwestern Zhejiang Province. Built-up lands occupation mainly occurred in flat areas with low altitude. Population and economy were estimated as two main driving factors of built-up land expansion for the strong correlation between population/GDP and the area of built-up lands. Total industrial output played lighter role compared with population and GDP. The Influence of population in2010visibly decreased compared with1995,2000and2005, but the effects of GDP and total industrial output has increased significantly. Population is detected as the most important driving factors of cultivated land decrease. However, low correlation appeared between area of cultivated land and GDP, and no obvious correlation was detected between total industrial output and area of cultivated land. The comparison between the proportions of the cultivated land/construction land and the level of reservoir water quality showed that with the declined water quality level, the average proportion of construction land in reservoir watershed presented a trend of increase.We adopted CART decision tree method to predict the water quality of reservoirs in Zhejiang Province by selecting16parameters including land use, socio-economic indicators, reservoir characteristics and climate as the independent variables and the water quality level of73drinking reservoirs of2010as the dependent variables. The classification accuracy of the model reached94.23%. After validation with the test dataset, the accuracy of the model reached80.95%. The CART rules revealed that human factors, including industrial wastewater emissions, industrial output, GDP, population density and land use, were the main factors in determining the quality level of reservoir water in Zhejiang province. The analysis of the importance of parameters showed that industrial wastewater emission was the most important factor causing water quality change.Finally, we used the PSR framework to evaluate and analyze the ecological environmental security of30large reservoir watersheds in Zhejiang Province from1995to2010by integrating RS, GIS, fuzzy analytic hierarchy process (FAHP) and set pair theory. The results showed that the ecological environment security of the reservoir watersheds had decreased in general from1995to2010. The pressure and status sub-indices mainly contributed to the decrease, while the response sub-index had a slight increase. For the spatial distribution, the security level of southwest is higher than the northeast in general. And regions where significant decreased security occurred were mainly concentrated in the central and northeast areas of Zhejiang Province.更多还原