Lake thermal stratification is important for regulating lake environments and ecosystems and is sensitive to climate change and human activity. However, numerical simulation of coupled hydrodynamics and heat transfer processes in deep lakes using one-dimensional lake models remains challenging because of the insufficient representation of key parameters. In this study, Lake Qiandaohu, a deep and warm monomictic reservoir, was used as an example to investigate thermal stratification via an improved parameterization scheme of the Weather Research and Forecast (WRF)-Lake. A comparison with in situ observations demonstrated that the default WRF-Lake model was able to simulate well the seasonal variation of the lake thermal structure. However, the simulations exhibited cold biases in lake surface water temperature (LSWT) throughout the year while generating weaker stratification in summer, thereby leading to an earlier cooling period in autumn. With an improved parameterization (i.e., via determination of initial lake water temperature profiles, light extinction coefficients, eddy diffusion coefficients and surface roughness lengths), the modified WRF-Lake model was able to better simulate LSWT and thermal stratification. Critically, employing realistic initial conditions for lake water temperature is essential for producing realistic hypolimnetic water temperatures. The use of time-dependent light extinction coefficients resulted in a deep thermocline and warm LSWT. Enlarging eddy diffusivity led to stronger mixing in summer and further influenced autumn cooling. The parameterized surface roughness lengths mitigated the excessive turbulent heat loss at the lake surface, improved the model performance in simulating LSWT, and generated a warm mixed layer. This study provides guidance on model parameterization for simulating the thermal structure of deep lakes and advances our understanding of the strength and revolution of lake thermal stratification under seasonal changes.