Canopy interception is an important part of forest ecological hydrology. The SWAT (Soil and Water Assessment Tool) model is relatively rough in its simulation process. In order to describe the canopy interception process more precisely in the hydrological simulation of the watershed, a semi-theoretical canopy interception model (Gash model) is used to couple with the SWAT model, and the forest canopy interception module of the SWAT model is optimized and improved with the Tianshan forest area as the research area. Through comparative analysis of the simulation results before and after the improvement, the results show that:1) the coefficient of determination (R²) of the SWAT model and the SWAT-Gash model are 0.59-0.83 and 0.65-0.86, respectively. The Nash-Sutcliffe efficiency (NSE) values are 0.58-0.82 and 0.63-0.85, respectively. The percent bias (PBIAS) of the two models is 7.2%-17.1%. These results reveal the improved performance of the proposed SWAT-Gash approach over the existing approach for catchment-scale streamflow estimation; 2) Compared with the outfall runoff data, the root-mean-square difference (RMSD) values of the SWAT model and the SWAT-Gash model are 3.49-7.80 m³/s and 3.22-4.68 m³/s. Pearson correlation coefficients of the SWAT-Gash model during the calibration period and the verification period are 0.93 and 0.81, respectively, which are higher than the SWAT model's 0.91 and 0.77; 3) Uncertainty analysis based on quantile regression (QR) shows that the P factors of SWAT model and SWAT-Gash model verification period are 0.93 and 0.96, R factors are 1.26 and 1.19, and the average width of 95% uncertainty confidence interval is 13.50 m³/s and 12.86 m³/s, respectively; 4) The monthly average surface runoff of the SWAT model and the SWAT-Gash model during the verification period are 6.55 m³/s and 8.50 m³/s, respectively, indicating that the original SWAT model would overestimate the canopy interception in this watershed. This paper takes the forest area in the middle section of the northern slope of Tianshan Mountain as an example to simulate the canopy interception of spruce forest. Although the requirements for model input data are increased and the collection of canopy interception data increases the uncertainty of model simulation, the accuracy of hydrological simulation based on physical processes in this study area is improved significantly, and the improved model is more consistent with the measured runoff data at the mountain pass. It can provide a more reliable basis for water resources management in small watersheds in Tianshan forest area.