<p>Landscape trees are one of the most important urban carbon sinks, and their carbon sequestration capacity is influenced by the unique microclimates of urban built environments. However, the mechanisms underlying this influence remain complex and not fully understood. Here, we combined year-long field measurements of the net photosynthetic rate <i>(Pn)</i> of landscape trees with ENVI-met microclimate simulation and machine learning to develop a predictive framework for a subtropical city. Our key findings revealed that the direct impact of urban microclimatic factors on <i>Pn</i> decreased in the order of solar radiation (<i>SR</i>) &gt; relative humidity (<i>RH</i>) &gt; carbon dioxide concentration (<i>Ca</i>) &gt; air temperature (<i>Ta</i>). The random forest (RF) model demonstrated high predictive accuracy, yielding an <i>R</i>^<i>2</i> of 0.86 for daily carbon sequestration and 0.98 for annual carbon sequestration. The simulation results show significant variations in tree <i>Pn</i> across different species in urban environments with organized building and road layouts. Furthermore, building height and tree placement introduced substantial spatial variability in carbon sequestration rates among trees of the same species. Differences in annual carbon sequestration between individual trees reached up to 37.67%. This mechanism-driven approach provides urban planners with a robust tool for optimizing green space design to enhance carbon sequestration, offering a practical strategy for supporting urban carbon neutrality goals.</p>