Hydro-thermal properties of permafrost and its distribution are sensitive to climate changes and human activities. Accurate and reasonable prediction on aforementioned information is important for eco-environment construction and vital infrastructures development. To model the current and future states of permafrost, it is a key challenge to effectively determine the upper hydro-thermal boundary conditions for permafrost models under changing climate and different underlying surfaces at proper spatial and temporal scales. An approach, combined regional climate downscaling method with model output statistics method, was developed to produce a time series of air temperature, surface temperatures, and surface unfrozen water contents for different underlying surfaces. It provided various climate and surface parameters at a spatial scale on the order of 102 m2 for engineering designs, which was used to predict boundary conditions under possible climate scenarios. The predicted and simulated models were calibrated and validated by the monitored data at an experimental site in Chumar, China, close to the Qinghai-Tibet Railway and the Qinghai-Tibet Highway. Results show that the multiple linear regression model (MLRM) can predict the current states and future changes of upper hydro-thermal boundary conditions for permafrost while the original states of natural surface are modified by natural or human factors on the condition of complicated climatic and complex topography regions. The statistical regression model (SRM) based on the outputs of regional climate model (RCM) and MLRM provides a simple method for the convenience of numerical calculation. These results also indicate the possible applications to other areas and situations.