A series of model sensitivity simulations are carried out to calibrate and improve the Weather Research and Forecasting Model coupled with a one-dimensional lake model (WRF-Lake) based on observations over Lake Nam Co. Using the default lake model parameters, the solution of WRF-Lake exhibits significant biases in both the lake thermodynamics and regional climatology, i.e., higher lake surface temperature (LST), earlier onset of summer thermal stratification, and overestimated near-surface air temperature and precipitation induced by the lake’s excessive warming and moistening impacts. The performance of WRF-Lake is improved through adjusting the initial lake temperature profile, the temperature of maximum water density (Tdmax), the surface roughness length, and the light extinction coefficient. Results show that initializing the water temperature with spring observation mitigates the LST overestimation and reduces the timing error of the onset of thermal stratification. By further adjusting Tdmax from 4 °C to the observed value of 3.5 °C, the LST increase from June to mid-July is enhanced and the buildup of thermal stratification is more accurately predicted. Through incorporating the parameterized surface roughness length and decreasing the light extinction coefficient, the model better reproduces the observed daily evolution of LST and vertical lake temperature profile. The calibrated WRF-Lake effectively mitigates the overestimation of over-lake air temperature at 2 m height and precipitation over regions downwind the lake. This suggests that an improved lake scheme within the coupled WRF-Lake is essential for realistically simulating the lake–air interactions and the regional climate over the lake-rich Tibetan Plateau.