The environmental effects of increasing CO2 emissions from road transport in port landside area (PLSA) have attracted significant research attention. This study aims to identify the spatial effects of CO2 emissions in PLSA by introducing a machine learning approach. A high spatiotemporal precision emissions inventory is constructed using one month of heavy-duty vehicles GPS trajectory data. We develop an eXtreme Gradient Boosting model to assess the spatial effects of CO2 emissions. Our findings indicate that the spatial effects of CO2 emissions in PLSA exhibit non-linear characteristics, identified as four effect ranges in space with corresponding thresholds of 30 km, 150 km, 300 km, and 1200 km. The interaction effects between the distance to port and other variables reveals the spatial thresholds leading to negative effects on CO2 emissions. We propose the concept of a four-tier Landside Emission Control Area and suggest four policy recommendations to promote carbon reduction in PLSA.