The groundwater is one of the most important basic resources in arid land. The groundwater recharge is easier to be affected by vegetation coverage in the arid and semi-arid areas than the wet areas. Bayin River Basin, a typical alpine and arid endorheic river watershed is located in the northeast of the Qaidam Basin. In the past 20 years, because of the warmer and more humid climate and human activities, the vegetation coverage condition of the watershed was becoming better. Calculation of the groundwater recharge in endorheic watershed is difficult. Soil and Water Assessment Tool (SWAT) is one of the most important tools that can simulate the watershed hydrological processes and its impact factors. However, it has considerable limitations in vegetation coverage and groundwater processes simulation. For the vegetation coverage simulation (leaf area index, LAI), it neglects the precipitation and terrain, which is important in arid land; For the groundwater simulation, it only is based on a simple linear mathematical formula without consideration of the physical processes. To better reveal the impacts of the vegetation coverage increasing on groundwater recharge in arid inland river, this research modified the SWAT model by replacing the LAI module with Global Land Surface Satellite (GLASS) based on LAI. The land use/cover change of the Middle and Lower Reaches of the Bayin River was further considered to better simulate the vegetation dynamics. After that, the modified SWAT model (named Dynamic Vegetation SWAT, DVSWAT) was coupled with the MODFLOW (MODular finite difference groundwater FLOW model), a professional groundwater processes simulation model. The latest GLEAM (Global Land Evaporation Amsterdam Model) v3 based on monthly evapotranspiration data and the observed groundwater level data were used to calibrate the DVSWAT-MODFLOW. We further used the variable-controlling approach to analyze the impacts of the vegetation coverage increasing on groundwater recharge based on the simulation results of the calibrated DVSWAT-MODFLOW. The results showed that the performance of the DVSWAT-MODFLOW in modelling the monthly evapotranspiration was good, with the R² value of over 0.83, NSE value of over 0.68, and PBIAS value within -22%-22% for each subbasin. The performance of the DVSWAT-MODFLOW in modelling the monthly groundwater level was good as well, with the R² value of over 0.95 and absolute bias of below 1 m for each groundwater level observation well. The area of the forest land and grassland in the study area increased 5.41 times and 98.96%, respectively, in 2019 compared to 2001. In addition, the annually average LAI of the study in 2019 increased 28.83% than 2001. The yearly and monthly groundwater recharge reduced about 6.1-26.52 mm and 0-15.03 mm, respectively, under the impacts of the increased vegetation coverage. The influence of the increasing vegetation coverage on the annual groundwater recharge depends on the annual precipitation to a certain extent. The months with strong influence on the monthly groundwater recharge are concentrated in the period of vigorous plant growth and more precipitation.