快速的社会经济发展导致城市出现以PM<sub>2.5</sub>为首要污染物的空气污染问题,PM<sub>2.5</sub>污染严重危害人群健康。因此,厘清PM<sub>2.5</sub>时空分布特征并估算其带来的健康影响,对于PM<sub>2.5</sub>的区域联防联控具有重要意义。现有研究中,为弥补地面监测数据的不足,借助机器学习算法估算PM<sub>2.5</sub>浓度成为研究热点,此外,基于流行病学研究结果的健康效应模型也被广泛用于评估PM<sub>2.5</sub>健康影响的研究中。利用珠江三角洲地区2014-2018年56个空气质量监测站的PM<sub>2.5</sub>实时监测数据、气象数据、社会经济数据和归一化植被指数,构建随机森林模型,多要素联合估算2000-2018年监测站点的PM<sub>2.5</sub>浓度,并采用克里金插值方法获得PM<sub>2.5</sub>浓度的空间分布,在此基础上应用全球暴露死亡(GEMM)模型,评估珠三角地区的PM<sub>2.5</sub>健康效应。结果表明:(1)2000-2018年期间,珠三角地区的PM<sub>2.5</sub>算术年均浓度维持在35μg/m<sup>3</sup>左右,呈现"西北-东南"递减空间分异;降水量、温度、风速和水汽压等气象因子对PM<sub>2.5</sub>浓度具有负向影响,GDP和人口密度等社会经济因子对PM<sub>2.5</sub>浓度具有正向影响。(2)2000-2018年期间,珠三角地区PM<sub>2.5</sub>人口加权年均浓度均低于PM<sub>2.5</sub>算术年均浓度,表明珠三角地区人口密度和PM<sub>2.5</sub>浓度未呈现明显的空间匹配关系,例如肇庆PM<sub>2.5</sub>浓度较高但人口密度较低,深圳PM<sub>2.5</sub>浓度较低但人口密度较高。(3)2000-2018年期间,珠三角地区PM<sub>2.5</sub>污染对于缺血性心脏病和中风的健康影响较显著,而对下呼吸道感染的健康影响较弱。区域PM<sub>2.5</sub>相关过早死亡人数逐渐增多,主要集中在PM<sub>2.5</sub>浓度和人口密度较高的地区,例如珠三角中心地区,以广州中心城区表现明显。本研究建议珠三角地区加大空气污染治理力度,提高医疗服务水平,同时关注城市人口结构,引导城市人口有序流动迁移,以缓解PM<sub>2.5</sub>带来的健康影响,实现城市化的健康发展。;Rapid urbanization has led to severe air pollution in China cities, especially fine particulate matter (PM<sub>2.5</sub>) pollution, which threatens human health seriously and becomes one of the main air pollutants. The clarification of the spatiotemporal distribution of PM<sub>2.5</sub> and estimating its health impact are essential for joint prevention and control of PM<sub>2.5</sub> pollution. Machine-learning with the merits of the estimation of PM<sub>2.5</sub> concentration has become a research hotspot to fulfill the deficiency of ground monitoring data. Based on the results of epidemiological studies, health effect models have been widely used in PM<sub>2.5</sub> health impact estimation research. In this study, we utilized the real-time PM<sub>2.5</sub> monitoring data, meteorological data, socio-economic data and Normalized Differential Vegetation Index (NDVI) data from 56 air quality monitoring stations in the Pearl River Delta (PRD) of China during 2014-2018, and constructed the random forest model to estimate the PM<sub>2.5</sub> concentration in the PRD during 2000-2018. Then, the Global Exposure Mortality Model (GEMM) model was adopted to estimate the long-term variations of the PM<sub>2.5</sub>-related premature mortality in the PRD during 2000-2018. The main results of this study are as follow: (1) the PM<sub>2.5</sub> concentration in the PRD has maintained at about 35 μg/m<sup>3</sup> during 2000-2018, showing a spatial differentiation that declined from northwest to southeast. Precipitation, temperature, wind speed and vapor pressure had a negative effect on PM<sub>2.5</sub> concentration, while GDP and population density had a positive effect on PM<sub>2.5</sub> concentration. (2) The population-weighted average PM<sub>2.5</sub> concentration was lower than the arithmetic average PM<sub>2.5</sub> concentration, indicating that there was no obviously spatial matching relationship between population density and PM<sub>2.5</sub> concentration. For instance, Zhaoqing had a high level of PM<sub>2.5</sub> concentration and a low level of population density, while Shenzhen had a low level of PM<sub>2.5</sub> concentration and a high population density. (3) PM<sub>2.5</sub> pollution in the PRD had a significant impact on ischemic heart disease and stroke, but a weak impact on low respiratory infections during 2000-2018. The number of PM<sub>2.5</sub>-related premature mortality has increased gradually and mainly concentrated in the center of the PRD, especially the central district of Guangzhou. This study suggests that the government should not only strengthen air pollution control efforts and improve the medical service level in city, but also pay more attention to the urban population structure and guide population migration orderly, which will facilitate the alleviation of the health impact of PM<sub>2.5</sub> pollution<sub></sub> and promote healthy urbanization.