道路交通排放已成为影响我国大城市空气质量的关键因素。精准解析道路交通排放特征、挖掘“交通-浓度”的动态响应关系并进行精细化交通排放监管，已成为改善城市空气质量、降低公众健康风险的关键工作。本研究聚焦机动车道路排放特征精准解析及空气质量影响动态表征两大目标，从数据基础到技术手段全面提升机动车污染研究的精细化、智能化和动态化水平。首先对中国高分辨率机动车排放模型进行了重要更新，引入了轻型车启动排放模块并完善了尾气排放温度修正模块。进而基于机器学习方法构建了面向不同交通数据基础的城市全路网交通流及排放清单动态模拟方法，并解析了人口、地理、城市建设等指标对成都和洛杉矶地区交通排放空间分布的影响规律。最后基于机器学习方法高效、准确地表征了两地气象、交通与污染物浓度之间的动态响应关系。环境温度显著影响机动车排放水平，我国重点车队排放因子呈现明显的区域和季节变化特征：黑龙江省汽油小客车和柴油重货的NOX年均排放因子分别为海南省的1.3和1.9倍；北京市柴油重货1月的NOX排放因子是7月的2.5倍。环境温度对排放的影响幅度可能高于交通拥堵，在机动车排放研究及空气质量模拟中需予以重视。成都市作为典型同心圆模式的城市，呈现城市核心区聚集、向外辐射状削减的交通排放分布特征；洛杉矶地区是多中心城市结构，呈现向城际高速路网聚集的交通排放分布特征。人口密度、土地利用、重要兴趣点分布等要素可解释导致两地区交通排放空间异质性的大部分因素（R2>0.6）。交通排放强度与人口密度的关系非线性、非单调，需谨慎采用以人口密度作为权重的交通排放空间分配方法。实时交通流数据的引入显著提升了机器学习空气质量模拟方法的动态性和准确性，可以高效、准确地表征气象、交通等要素对污染物浓度的影响规律（R2基本达到0.8）。如NOX排放贡献更高的货车流量降低会导致成都市和洛杉矶地区大部分站点O3浓度增加，体现了NO滴定效应对VOC控制区的影响。排除气象因素的影响，新冠疫情期间交通削减导致洛杉矶地区NO2和PM2.5浓度下降了2.9 ppb（-30.1%）和1.1 μg/m3（-17.5%），MDA8 O3浓度上升了2.1 ppb（5.7%），与基于化学传输模型的研究高度吻合。机器学习空气质量模拟方法的模型灵活度和计算效率相比传统物理化学传输模型显著提升，在未来的实时、精细化交通及空气质量管理中可发挥重要作用。

On-road vehicle emission is one of the most important factors affecting air quality in megacities of China. Mapping on-road vehicle emissions and quantifying traffic-air quality interrelations, thus supporting accurate vehicle emission control, are fundamental tasks for improving urban air quality and alleviating public health risks.The major goal of this study is to improve the accuracy, intelligence and dynamics of vehicle emission studies thoroughly. Two major objectives are included, which are accurate vehicle emission characterizing and dynamic air quality impacts quantifying. Firstly, this study updates the Chinese vehicle emission model by introducing start emission sub-model and modifying temperature correction module. Then, high-resolution vehicle emission mapping methods are developed based on various traffic data and machine learning (ML) methods. Furthermore, nonlinear relationships between emissions, demographic and land use variables were analyzed. Finally, the dynamic responses between meteorology, traffic and air quality are described based on ML-based methods. Emission factors of key fleets in China show significant regional and seasonal variations affected by ambient temperature. The updated emission model shows that the annual average NOX emission factors (EF) of light-duty gasoline vehicles (LDGVs) and heavy-duty diesel trucks (HDDTs) in Heilongjiang province are 1.3 and 1.9 times as large as those in Hainan province, respectively. While the fleet average NOX EF of LDGVs and HDDTs in Beijing in January are 1.5 and 2.5 times as large as those in July. Temperature can pose larger impacts on traffic emission than congestion driving conditions, needed to be carefully considered in vehicle emission studies and air quality simulations. As a city following the typical concentric zone model, on-road traffic emissions in Chengdu tend to concentrate in city center and disperse radially. As a city following the multiple nuclei model, on-road traffic emissions in Los Angeles (LA) basin tend to gather along intercity highways. Population density, land use and important points of interest (POIs) account for most of the heterogeneity in spatial distribution of traffic emissions (R2>0.6). The population density-traffic emissions relationships vary in different cities with nonlinear and non-monotonic patterns. Therefore, the spatial allocation of traffic emissions based on population density should be carefully adopted.The introduction of real-time traffic data significantly improves the dynamics and accuracy of ML-based air quality simulation methods, enable the method to quantify the dynamic responses between meteorology, traffic and pollutant concentrations with high efficiency and accuracy (with cross validation R2 up to 0.8). For example, MDA8 O3 at most AQ monitoring sites in Chengdu and LA basin increases with the reduction of truck activities serving as the major contributor for on-road NOX emissions, reflecting the NO titration effects in VOC-limited regimes. Excluding impacts of meteorological factors, the decreases in traffic activities during COVID-19 pandemic reduced NO2 and PM2.5 concentrations in LA basin by 2.9 ppb (-30.1%) and 1.1 μg/m3 (-17.5%), while O?3 concentration increased by 2.1 ppb (5.7%). The conclusions above are consistent with the studies based on chemical transport models. The ML technique has more flexibility in leveraging real-world data and possesses higher computational efficiency, promising to support high-resolution, real-time air quality regulation in the near future.