Abstract: Based on the Lagrangian Hybrid Single-Particle Integrated Trajectory (HYSPLIT) model, this study quantitatively analyzed the spatiotemporal diffusion patterns of PM2.5 emitted from a forest fire that occurred in Muli Tibetan Autonomous County, Liangshan Yi Autonomous Prefecture, Sichuan Province, using three fire emission inventories: the Global Fire Emissions Database (GFED), the Quick Fire Emissions Dataset (QFED), and the Global Fire Assimilation System (GFAS), together with meteorological data from March 29 to April 3, 2020. The simulation results were validated against Sentinel-2 satellite imagery, and the simulated PM2.5 concentrations from different inventories were compared with ground-based observations to evaluate the effectiveness and accuracy of these inventories in modeling forest fire events. The results showed that the PM2.5 diffusion patterns simulated by HYSPLIT were consistent with the smoke-affected areas visually identified in Sentinel-2 imagery, with a Sørensen index of 0.682, indicating substantial spatial agreement. The total PM2.5 emissions varied considerably among the different inventories, following the order GFED > QFED > GFAS. Ground-based PM2.5 measurements were positively correlated with all simulated values, with correlation strength ranked as GFAS > QFED > GFED. This study demonstrates the applicability of different global fire emission inventories in small-scale forest fire cases and provides technical support for the assessment and prevention of wildfire-related pollution hazards.