With current remote sensing technologies, it is not possible to directly infer the thermal state of the ground from spaceborne platforms. We demonstrate that such limitations can be overcome by combining the information content of several remote sensing products in a data fusion approach: time series of remotely sensed land surface temperature, as well as snow cover and snow water equivalent, are employed to force ground thermal models which deliver ground temperatures and thaw depths. First, we present a semi-empirical model approach based on remotely sensed land surface temperatures and reanalysis products from which mean annual ground temperatures (MAGT) can be estimated at a spatial resolution of 1 km at continental scales. The approach is tested for the unglacierized land areas in the North Atlantic region, an area of more than 5 million km2. The results are compared to in-situ temperature measurements in more than 100 boreholes from which the accuracy of the scheme is estimated to approximately 2.5 °C. Furthermore, we explore transient modeling of ground temperatures driven by remotely sensed land surface temperature, snow cover and snow water equivalent. The permafrost model CryoGrid 2 is applied to the Lena River Delta in NE Siberia ( 25000 km2) at 1km spatial and weekly time resolution for the period 2000–2014. A comparison to in-situ measurements suggests a possible accuracy of around 1 °C for annual average ground temperatures, and around 0.1 m for thaw depth. However, information on subsurface stratigraphies including the distribution of ground ice is required to achieve this accuracy which is currently not available from remote sensing products alone. Finally, we discuss the potential and limitations of schemes using a combination of remote sensing data and thermal permafrost models to assess the thermal state of the ground, and give a feasibility assessment for both mountain and lowland permafrost regions.