Abstract: Using ERA5 single-level reanalysis data during 2010–2019 and meteorological observations from a PBL (planetary boundary layer) tower in a mountainous forest in Huainan from 1 May 2016 to 30 April 2017 as forcing data, the authors evaluated the performance of the CLM4.5 (Community Land Model Version 4.5) against observed canopy fluxes and micrometeorological data at this site. The effects of these two forcing datasets on model simulation were compared. Additionally, three simulation experiments were conducted to assess the influence of soil texture on soil moisture and improve model performance. The results showed that CLM4.5 performs well in simulating the land–atmosphere exchange processes in the Huainan mountainous forest, with simulations driven by PBL tower data outperforming those driven by ERA5 reanalysis data. Both datasets performed well in radiation simulation, with the PBL-driven simulation showing particularly strong performance, where the correlation coefficient exceeded 0.97 and RMSE (root mean square error) was below 25.056 W m⁻². The ERA5 reanalysis data forced simulation showed a slightly lower correlation coefficient also reaching 0.92, with a RMSE below 29.94 W m⁻². In soil temperature simulation, correlation coefficients were all above 0.98. The correlation coefficients of soil moisture simulation were above 0.86, but the values were systematically higher. The annual average correlation coefficients for sensible heat flux simulations were 0.72 and 0.78, respectively. The simulated results of the soil moisture with the measured three layers of soil texture combined with the deep given soil texture data were closest to the observations. Thus, accurate description of soil texture could greatly improve simulation results of soil moisture. This study demonstrates that ERA5 single-layer reanalysis data is reliable for simulating land–atmosphere exchange processes in Huainan’s forest and can be applied to similar regions in the north–south climate transition zone.