非点源污染不仅是当今我国面临的一个严峻的生态环境问题，也是国内外学者十分关注的热点领域。山东省是我国的经济大省，随着国民经济的迅速发展和城镇化进程不断推进，非点源污染及其所造成的环境影响已严重威胁居民健康生活，也成为了山东省现代生态农业健康发展和城镇经济可持续发展的瓶颈之一，其中淮河流域已被列为国家的非点源污染重点治理区域。面对严峻的生态环境形势，亟需对全省水环境非点源污染现状，区域空间分布特征等方面进行综合研究和分析。因此本研究提出水文模型和人工神经模型与输出系数模型相互耦合的方式丰富优化了输出系数模型的功能结构，构建了非点源总氮径流污染模型，以山东省为大区域尺度典型研究区域，在分析其年降雨量和土地利用时空演变的基础上，多尺度(时间尺度和空间尺度)，多角度(网格单元、控制单元和土地利用类型)视角下，综合分析山东省的非点源总氮径流流失污染空间分布特征，并建立控制单元监测断面总氮浓度预测模型。本研究以期为基于控制单元污染的非点源污染研究提供一些参考，同时为山东省在非点源总氮污染控制区域以及治理措施提供一些理论支持，其主要研究结论如下： (1)本研究采用了Mann-Kendall统计检验方法和小波变换方法等手段解析了年降雨量和土地利用的时空演变趋势。结果表明：在时间序列上，1980-2017年山东省年降水量表现为略微上升的趋势(Z=0.83<1.65, α>0.1)。山东省年降水量有明显的准5年，准11年和准25年三重嵌套的震荡周期现象，以此分析在未来25年里可能会在处于降雨量偏少的气候背景下，同时接下来的5年将进入干旱期，尤其是潍坊市和青岛市的干旱状况可能更加严重。在空间格局上，山东省的多年平均降水量呈南高北低的空间分布。MK趋势检验结果表明：山东省中部区域的年降水量变化呈不显著上升趋势，而西南和东北部分区域呈极显著下降趋势，其中潍坊市和青岛市近五年的平均降水量下降最为明显。另一方面，从土地利用转移情况来看，2011至2017年山东省农田用地和林地面积有小幅度的减少(10%以下)，草地和未利用土地缩减明显(30%以上)，农田与农村居民区之间具有相互流动的特性。同时，城镇化进程较快，城镇用地面积增加了近一倍。 (2)本研究提出将水文模型和人工神经网络与输出系数模型相互耦合的方式丰富优化输出系数模型，构建了非点源总氮径流污染模拟模型，阐明了污染物模拟过程和模型整体框架。本研究将所构建的非点源总氮径流污染模型应用于2011-2017年山东省非点源总氮污染径流流失量的模拟，结果表明：2011-2017年山东省非点源总氮流失负荷呈波动上升的趋势。对于不同土地利用而言，山东省的总氮径流流失量排序状况为农村居民区>城镇居民区>农田>林地等自然土地利用，其中容纳畜牧养殖和生活污水两大污染源的农村居民区的多年平均总氮径流流失量占了总氮径流流失总量的67.3%。对于不同控制单元而言，山东省的总氮径流流失量在空间分布上具有较大差异，其中胶东半岛附近控制单元的总氮径流流失量低于内陆地区，南四湖地区附近的控制单元的农业污染状况比其他地区更严重。 (3)以控制单元作为单独汇水区域，非点源总氮径流污染模型考虑了动力因子(降雨径流模块)，负荷因子(输出系数模块)，环境因子(林地面积占比和草地面积占比)和地形因子(坡度和水流长度)建立的人工神经网络模块能够较好地模拟控制单元的2011至2017年内出水口监测断面的总氮浓度。研究以2011-2016年为模型的率定阶段(R2=0.86, RMSE=1.95)，以2017年为验证阶段(R2=0.8, RMSE=1.86)以及有效性判定阶段(R2=0.74, RMSE=3.11)。模拟良好的预测结果说明了整个非点源总氮污染模拟模型在具有一定的物理机制上，不仅避免一些机理模型过于复杂的数据准备和参数选取而且又能够达到较高的模拟精度(总体预测R2=0.82, RMSE=2.14)，证实了非点源总氮污染模拟模型的科学性和可行性。

Non-point source pollution is not only a serious ecological environment problem of China today, but also a hot spot that is highly concerned by scholars around the world. The economy of Shandong Province is very important in China's economy. With the rapid development of the national economy and the continuous progress of urbanization, non-point source pollution and its environmental impact have seriously threatened the healthy life of residents. Non-point source has also become one of the bottlenecks for the healthy development of modern ecological agriculture and the sustainable development of urban economy in Shandong Province. The Huaihe River Basin has been listed as a key non-point source pollution area in the country. Faced with the severe ecological environment, it is urgent to conduct comprehensive research and analysis on the status of non-point source pollution and regional spatial distribution characteristics of the province's water environment. Therefore, this study proposes a hydrological model, the artificial neural model and the Export Coefficient Model are coupled to each other to optimize the functional structure of the Export Coefficient Model, which constructed the non-point source total nitrogen runoff pollution model. This study took Shandong province as a typical regional study area, based on the analysis of the annual rainfall and the spatial and temporal evolution of land use, multi-scale (time scale and spatial scale), multi-angle (grid unit, control unit and land use type) perspective, to analysis the spatial distribution characteristics of total nitrogen runoff loss pollution in Shandong province, and established the prediction model of total nitrogen concentration in the monitoring section of the control unit. The research results are intended to provide some reference for non-point source pollution research based on control unit pollution, and provide some theoretical support for Shandong Province in non-point source total nitrogen pollution control areas and control measures. The main conclusions of this paper are as follows: (1) Mann-Kendall test and wavelet transform were used to understand the temporal and spatial evolution trend of annual rainfall and land use change in the study area. In terms of time, the annual rainfall in Shandong Province from 1980 to 2017 showed a slight upward trend (Z=0.83<1.65, α>0.1). Judging from the annual rainfall of Period, with an obvious triple nested oscillation Period of quasi-5 years, quasi-11 years and quasi-25 years, Shandong Province will be in a relatively dry climate in the next 25 years. Spatially, the annual rainfall in the middle region of Shandong Province showed an insignificant upward trend from 2011 to 2017, while some regions in southwest and northeast showed a significant downward trend. According to the Period and spatial changes of annual rainfall, the drought situation in Weifang and Qingdao may become more and more serious in the future. On the other hand, according to the land use transfer situation, from 2011 to 2017, the area of agricultural land and forest land in Shandong Province decreased slightly (less than 10%), and the reduction of grassland and unused land was the most obvious (more than 30%). What’s more, farmland and rural settlements have mutual flow characteristics. At the same time, urban land use has nearly doubled, which means that the urbanization process is rapid. (2) Based on remote sensing technology, SCS hydrological model, Export Coefficient Model and BP artificial neural network model were coupled to construct the total nitrogen non-point source pollution estimation model based on control unit division, and the model was applied to estimate the runoff export load of total nitrogen non-point source pollution in Shandong Province from 2011 to 2017. The total nitrogen pollution load from non-point sources showed a fluctuating upward trend from 2011 to 2017 in Shandong. For different land use, the sort r of total nitrogen runoff loss in Shandong Province is Rural > Farm > Town >natural land use such as forest land. The multi-year average total nitrogen runoff loss in rural residential areas, which contain two major sources of livestock farming and domestic sewage, which accounted for 67.3% of the total pollution nitrogen runoff losses load. From 2011 to 2017, rural land use and urban land use have always been the two highest non-point source total nitrogen pollution load per control unit area in Shandong Province. The load of total nitrogen non-point source pollution from each control unit is significantly different in spatial distribution. The total nitrogen pollution load of control units near Jiaodong Peninsula is lower than other of inland areas, and the agricultural pollution of control units near Nansi Lake river basin is more serious than that of other areas. (3) Taking the control unit as a small separate watershed, based on the consideration of dynamic factors (SCS hydrological model), load factors (Export Coefficient Model), environmental factors (forest area ratio and grass area ratio) and terrain factors (slope and water flow length), this model established the artificial neural network module to predict the total nitrogen concentration of the monitoring section in the control unit area, which can better predict the total nitrogen water quality concentration of the monitoring section in the control unit area from 2011 to 2017. The period 2011 -2016 is the model calibration phase (R2=0.86, RMSE=1.95), and the period 2017 is the model validation phase (R2=0.8, RMSE=1.86) and the validity determination phase (R2=0.74, RMSE=3.11). The good prediction results show that the whole coupling model can achieve higher simulation accuracy (overall prediction R2=0.82, RMSE=2.14) on the basis of having a certain physical mechanism while avoiding some overly complicated data preparation and model parameter selection, which proves the feasibility and stability of the non-point source total nitrogen pollution model.