空气污染对人体的健康损害使得污染物健康暴露研究成为领域热点。土地利用回归（Land Use Regression, LUR）模型存在时间分辨率较低的缺陷，而兴趣点（Point of Interests, POI）由于其功能特征在不同时刻会导致人群聚集的差异。本研究旨在实现高分辨率的人群动态暴露评估，在传统LUR模型预测变量的基础上加入POI数据从而提升模型的时间分辨率，针对不同季节的典型时刻，建立小时尺度细颗粒物（PM2.5）和二氧化氮（NO2）模型，并研究了各类POI在不同时间情景下的污染特征。为了弥补暴露研究中静态人口数据的缺陷，本研究使用共享单车大数据分析了北京城区人群出行的时空特征，并通过线性回归模型研究了不同情景下单车出行与区域POI的影响关系，在城区出行热度差异的基础上对静态人口进行重新分配，实现了高分辨率的动态人群分布。将相同尺度的污染物数据与人群动态分布数据匹配，得到高分辨率北京城区人群动态暴露水平，该方法对流行病学研究以及区域健康风险防控具有重要意义。主要创新成果如下：

         （1）和以往研究中年均、季均尺度的LUR模型相比，本研究实现了小时尺度上的建模，通过十折交叉验证以及同以往研究的模型参数对比显示模型拟合效果良好，R2最高达到0.955。本研究证实了POI对提升模型的时间分辨率具有重要作用，并且在工作日或非工作日的不同时刻，不同类型的POI对污染物的贡献程度具有差异性。此外，气象条件和道路变量对细颗粒物和二氧化氮两类污染物模型的解释程度也比较大。不同季节、时刻以及污染物类别的模型结果存在明显差异。各类POI的污染特征具有规律性，餐饮服务、公司企业、交通设施、生活服务以及金融银行类POI长期处于细颗粒物和二氧化氮均较高的污染水平，这对高风险功能区防控以及城市布局优化具有重要意义。

         （2）共享单车骑行具有较强的时空特征，时间上主要体现在工作日和非工作日以及不同时间段的骑行热度差异。从空间上来看，工作日骑行的高热度区主要聚集在几个典型商务区和产业园区，而非工作日骑行高热度区比较分散。长期骑行热度较高的位置主要集中在各个地铁站、交通枢纽以及国贸、金融街、望京、中关村等商圈和园区。逐步回归结果显示，每个网格内全天的骑行人次和各个时间段的骑行人次均与各类POI数量有显著关联，交通服务、商务住宅和休闲娱乐场所对骑行热度的影响最为显著。工作日和非工作日的不同时间段，与骑行热度相关的POI类别和数量存在明显差异。

         （3）根据共享单车骑行时空特征对静态人口重新分配，可以实现高分辨率的人群动态分布。使用动态人群和静态人口数据的污染物暴露水平结果存在显著性差异（p < 0.01）。两类污染物在不同时刻暴露分布情况与人群分布趋势较为一致，整体人群污染物暴露浓度长期较高的区域主要聚集在城区二环和四环环线附近，东城区和朝阳区交界附近是任何情景下污染物暴露浓度均较高的区域。在同一位置但不同时间的人群暴露水平也存在明显差异。基于人口加权的污染暴露评价更能反映对人群的真实影响，两类污染物加权暴露浓度比区域污染物平均浓度高出1.67 %-60.8 %，证明城市人群主要聚集在污染物的高风险区域。

        Pollutant’s health exposure research becomes hot spots due to the damage to human health caused by air pollution. Land Use Regression (LUR) models have the defect of low time resolution. Point of Interests (POI) can cause differences in crowd gathering at different moments due to its functional characteristics. This study aimed to achieve a high-resolution population dynamic exposure assessment. Based on the predictive variables of the traditional LUR models, POI data was added to improve the temporal resolution of the model. This study established models of fine particulate matter (PM2.5) and nitrogen dioxide (NO2) at hourly scale of typical moments in different seasons. Pollution characteristics of various POIs were also analyzed. To make up for the defect in exposure studies using static population data, this study used shared cycling big data and analyzedtemporal and spatial characteristics of Beijing city, based on which the static population was redistributed, and high-resolution dynamic population distribution was realized. Linear regression models were used to study the relationship between cycling characteristics and regional POIs. By matching high-resolution pollutant data with exposure level of the urban population in the same scale, high-resolution dynamic exposure levels of Beijing urban population were obtained. This study is of great significance for epidemiological research and regional health risk prevention and control. The main innovative results are as follows:

          (1) Compared with previous annual and seasonal LUR models, this study realized modeling on hourly scale, and the results of 10-fold cross validation and comparison with previous studies showed that the model fitting effect was good, and the maximum R2of the model reached 0.955. This study confirmed that POI plays an important role in improving the temporal resolution of the model, and different types of POI contribute to pollutants to different degrees at different moments of working days or non-working days.In addition, the temporal variables such as meteorological conditions and roads can explain the models to a greater extent. There were significant differences in the model results among different seasons, time, and types of pollutants. The pollution characteristics of all kinds of POI were regular, and POI of catering services, companies, transportation facilities, life services and financial banks had been at a high pollution level of fine particulate matter and nitrogen dioxide for a long time, which is of great significance for the prevention and control of high-risk functional areas and the optimization of urban layout.

         (2) Shared bike riding has strong spatial and temporal characteristics, which is mainly reflected in the frequency of riding on weekdays and non-weekdays, as well as the comparison between morning and evening peak hours and other periods. From the perspective of spatial distribution, the high heat areas of weekday cycling were mainly concentrated in several typical business districts and industrial parks, while the high heat areas of non-weekday cycling were relatively dispersed. The locations with high long-term cycling fever were mainly concentrated in subway stations, transportation hubs, business districts and parks such as World Trade Center, Financial Street, Wangjing and Zhongguancun areas. Stepward regression results showed that the number of riders in each grid all day and in each time period were significantly correlated with various POI categories, and transportation services, business residences and leisure and entertainment places had the most significant impact on the cycling heat. There were significant differences in the types and quantities of POI related to cycling heat at different working days and non-working days.

        (3) According to the spatiotemporal characteristics of shared bike riding, the static population redistribution can realize the high-resolution dynamic population distribution. There were significant differences in pollutant exposure levels between dynamic and static population (P < 0.01). The exposure distribution of the two types of pollutants at different times was consistent with the distribution trend of the population. The areas with long-term high pollutant exposure concentration of the whole population were mainly concentrated in the vicinity of the Second Ring Road and the Fourth Ring Road in the urban area. The area near the junction of Dongcheng District and Chaoyang District was the area with high pollutant exposure concentration under any situation. There were also significant differences in exposure levels among people in the same location but at different times. The pollution exposure assessment based on population weighting can better reflect the real impact on the population. The weighted exposure concentrations of the two pollutants were 1.67 %-60.8 % higher than the average concentration of regional pollutants, which proved that urban people mainly gather in the high-risk areas of pollutants.