Large reservoirs play a key role in regional hydrological cycles as well as in modulating the local climate. However, it’s still not well studied how the thermal characters of large reservoirs will evolve under climate change because previous studies almost always focused on natural lakes. In this study, the lake model in the Weather Research and Forecasting (WRF) system was equipped with an inflow/outflow module (the new model called WRF-RLake hereafter) to reflect the strong impact of throughflow to reservoir thermal structures. The WRF-RLake model was evaluated at the 200 m deep Nuozhadu Reservoir in the Tibetan Plateau Region and produced water temperature profiles in good agreement with measurements. Sensitivity analysis were carried out based on 30-year historical meteorological data and illustrated that the water temperatures in the surface 10 m were sensitive to air temperature and wind speed, and deeper water temperatures were more sensitive to inflow temperature and outflow level. The WRF-RLake model was then driven by different climate scenarios of the 21^st century to predict the dynamics of water temperatures at the Nuozhadu Reservoir. The dynamics of the water temperatures at different depths were quite different as the surface water warmed up significantly while the deep water was almost muted to warming. This resulted in stronger stratification and reduced heat diffusion which in return strengthened the different dynamics of the surface and deep water temperatures. This study also urges that the difference between forcing data produced by diverse climate models also exerts significant impact on lake model simulations.