A fully coupled three-dimensional surface and subsurface land model is developed and applied to a site along the Columbia River to simulate three-way interactions among river water, groundwater, and land surface processes. The model features the coupling of the Community Land Model version 4.5 (CLM4.5) and a massively-parallel multi-physics reactive transport model (PFLOTRAN). The coupled model, named PFLOTRAN_CLM v1.0, is applied to a 400 m × 400 m study domain instrumented with groundwater monitoring wells along the Columbia River shoreline. PFLOTRAN_CLM v1.0 simulations are performed at three spatial resolutions over a five-year period to evaluate the impact of hydro-climatic conditions and spatial resolution on simulated variables. Results show that the coupled model is capable of simulating groundwater-river water interactions driven by river stage variability along managed river reaches, which are of global significance as a result of over 30,000 dams constructed worldwide during the past half century. Our numerical experiments suggest that the land-surface energy partitioning is strongly modulated by groundwater-river water interactions through expanding the periodically inundated fraction of the riparian zone, and enhancing moisture availability in the vadose zone via capillary rise in response to the river stage change. Furthermore, spatial resolution is found to impact significantly the accuracy of estimated the mass exchange rates at the boundaries of the aquifer, and it becomes critical when surface and subsurface become more tightly coupled with groundwater table within six to seven meters below the surface. Inclusion of lateral subsurface flow impacted both the surface energy budget and subsurface transport processes. The coupled model developed in this study can be used for improving mechanistic understanding of ecosystem functioning, biogeochemical cycling, and land-atmosphere interactions along river corridors under historical and future hydro-climatic changes. The dataset presented in this study can also serve as a good benchmarking case for testing other integrated models.