Assessing possible permafrost degradation related to engineering projects, climate change and land use change is of critical importance for protecting the environment and in developing sustainable designs for vital infrastructure in cold regions. A major challenge in modelling the future degradation of permafrost is finding ways to constrain changes in the upper thermal boundary condition over time and space at appropriate scales. Here, we report on an approach designed to predict time series of air, ground surface and shallow ground temperatures at a spatial scale on the order of 102 m2 for engineering design of a railway or highway project. The approach uses a regional-scale atmospheric model to downscale global climate model output, and then stepwise multiple regression to develop an equation that provides a best-fit prediction of site-specific observational data using bilinearly interpolated output from the atmospheric model. This approach bridges the scale difference between atmospheric climate models and permafrost thermal models, and allows for a wider range of factors to be used in predicting the thermal boundary condition. For a research site located in Beiluhe, China, close to the Qinghai-Tibet Railway, a comparison of model predictions with observational data not used in the construction of the model shows that this method can be used with a high degree of accuracy to determine the upper boundary condition for a permafrost thermal model. Once a model is constructed, it can be used to predict future changes in boundary condition parameters under different greenhouse emission scenarios for climate change. Copyright © 2012 John Wiley & Sons, Ltd.