As a key driver of land-atmosphere interactions, land surface temperature (LST) has been widely applied across various geoscientific disciplines. In mountainous regions, complex terrain and frequent cloud cover create a strong demand for LST products with higher spatiotemporal resolution. Previous studies have developed methods for cloud-covered LST reconstruction or for downscaling to alleviate the challenges of cloud obstruction and coarse resolution in thermal infrared observations. However, most of these approaches focus on a single task and seldom account for the unique characteristics of mountainous areas, which limits the applicability in cloudy mountainous regions. This study, therefore, proposes an integrated reconstruction-downscaling framework designed to simultaneously address the limitations of cloud obstruction and coarse resolution in cloudy mountainous regions. The framework was developed based on the annual temperature cycle (ATC) model, which partitions daily LST into a background temperature component estimated from the ATC coefficients and a residual term representing short-term thermal variability. ATC coefficients and daily, all-weather residual term at 30 m scale were separately estimated using Extreme Gradient Boosting (XGBoost) regression. These components were then combined to generate the final LST product. Applied to the Wanglang National Nature Reserve for generating LST data in 2022, the proposed framework demonstrated improved agreement with ground LST measurements relative to a two-step sequential approach of first reconstructing and then downscaling, achieving a reduction in RMSE of 0.16 - 0.73 K. This study holds the potential to provide valuable insights for improving fine-scale, all-weather LST retrieval in cloudy mountainous regions.