Abstract: An accurate prediction of the regional crop yield is often required to simulate the plant growth process. It is very necessary to effectively capture the various environmental factors. Particularly, the soil moisture dynamics can be shaped by the rainfall variability over different seasons. Traditional crop models, like the Environmental Policy Integrated Climate (EPIC) model, have been widely used to assess crop growth, nutrient cycling, and yield formation. However, it is still lacking to represent the watershed-scale hydrological processes. The high accuracy is limited in the regions, where the water availability is highly variable and uncertain. Especially, over 80% of the apple orchards rely entirely on the natural precipitation in dryland on the Loess Plateau in Northern China. Therefore, the soil moisture availability driven by rainfall can be the major constraint on the orchard productivity and sustainability in these water-limited environments. In this study, a coupled modelling approach was developed to integrate the Soil and Water Assessment Tool (SWAT) with the EPIC model. A framework was constructed to simulate both hydrological processes and crop growth at regional scales. Extended Fourier Amplitude Sensitivity Test (E-FAST) was carried out to verify the improved model. The key parameters were then optimized, including the crop water use, water stress response, and soil carbon dynamics. The performance of the model was improved for the overall reliability of the simulation. The coupled SWAT-EPIC model was applied to a typical apple production area in Dali County, northern Shaanxi Province, China. The results revealed that the optimized model reduced the simulation error in the rainfed apple yield prediction by 38.98%, with a root mean square error (RMSE) of 2.56% and a relative RMSE (RRMSE) of approximately 9.8%. The apple yields were simulated under rainfed conditions. The interannual variations in the apple yield were closely linked to the fluctuations in the precipitation and soil moisture. Furthermore, the shallow soil moisture declined significantly in years with less than 700 mm of rainfall. The deep soil layers (6–10 m depth) increased the dry behavior, leading to the water stress that reduced the yield. In contrast, higher water productivity and more stable yields depended on moderate rainfall and better soil moisture balance. Once the cropland was converted into apple orchards, the soil organic carbon storage was enhanced by about 14.85%. A great contribution was then gained to improve soil health and climate mitigation after carbon sequestration. The findings also highlighted the long-term risk of deep soil destruction resulting from excessive water extraction by deep-rooted perennial trees. The water resource strategies were also given, including supplemental irrigation during drought years. Particularly when the annual precipitation was below 700 mm, the available soil moisture was sustained in the shallow root zones. Furthermore, mulching practices and soil moisture conservation should be employed to optimize the orchard water use efficiency for overall fruit productivity. Overall, the SWAT-EPIC coupled model can serve as an effective tool to simulate the complex interactions among rainfall, soil moisture, and apple productivity across large regions. As such, the hydrological and crop growth were integrated to optimize the water use for the long-term sustainability of the apple orchards in water-scarce environments. The finding can also provide valuable insights to assess the regional apple yield and soil carbon dynamics in the Loess Plateau.