Soil moisture plays an essential role in the terrestrial water cycle. It is very important to obtain soil moisture for many applications. Soil moisture acquired by remote sensing, land surface modeling, and data assimilation must be evaluated against in-situ measurements before being used, but a procedure should be performed to upscale the point-scale measurements to the grid-scale or footprint-scale. In this study, a new upscaling algorithm is developed by introducing MODIS-derived apparent thermal inertia (ATI). First, a functional relationship between the station-averaged soil moisture and the pixel-averaged ATI is constructed. Second, this relationship is used to calculate the representative soil moisture time series at a certain spatial scale. Last, the Bayesian linear regression is applied to obtain the upscaled area-averaged soil moisture by using in-situ measurements as independent variables. The algorithm is evaluated using a network of in-situ moisture sensors in the central Tibetan Plateau. The results indicate that it can effectively obtain the area-averaged soil moisture, reducing the root mean square error (RMSE) from before upscaling to after upscaling. Finally, the algorithm is implemented to the grid box where the network is installed, and the temporal pattern of the upscaled soil moisture agrees with the hydro-meteorological knowledge of this region.