Abstract: Vehicle carbon emissions (i.e., CO₂ emitted from vehicle energy consumption) account for 25% of total carbon emissions and have a significant impact on local ecological environments. However, at present, both the vehicle carbon emission estimation method based on statistical yearbook data and the carbon emission simulation method primarily using numerical models exhibit poor temporal and spatial resolution and accuracy, which need improvement to meet the demands of existing research. Therefore, Shenzhen city was taken as the research area, a traffic carbon emission inversion method that considers temporal and spatial non-stationarity was proposed. The specific steps include: (1) Differentiating roads into urban center roads, suburban roads, and highways; (2) Conducting field experiments during peak and non-peak hours on typical roads to obtain vehicle carbon emission coefficients; (3) Establishing a correlation between vehicle carbon emissions and variables such as the built environment of roads and time tags using the Random Forest method, combined with measured data; (4) Ultimately constructing a high-precision traffic carbon emission dataset within the study area by integrating trajectory data. The temporal and spatial variation characteristics of traffic carbon emissions are explored, and their associations with urban functional zones are uncovered. The results show that: (1) The daily vehicle carbon emissions within the study area range from 2.2×10⁴ to 3.0×10⁴ kg; (2) Vehicle carbon emissions exhibit obvious temporal and spatial differentiation, with average daily emissions of 2.4×10⁴ kg on weekdays, and 2.9×10⁴ kg on weekends; (3) High vehicle carbon emission areas in Shenzhen are located in the southwest of the city, namely Nanshan District, Bao'an District, and Futian District, and these areas show a high-high clustering pattern. The research indicates that traffic trajectory data can provide a data foundation for large-scale vehicle carbon emission calculations, and the traffic carbon emission inversion method that considers temporal and spatial non-stationarity can effectively improve the temporal and spatial accuracy and precision of vehicle carbon emission data.