High-latitude areas are very sensitive to global warming, which has significant impacts on soil temperatures and associated processes governing permafrost evolution. This study aims to improve first-layer soil temperature retrievals during winter. This key surface state variable is strongly affected by snow's geophysical properties and their associated uncertainties (e.g., thermal conductivity) in land surface climate models. We used infrared MODIS land-surface temperatures (LST) and Advanced Microwave Scanning Radiometer for EOS (AMSR-E) brightness temperatures (Tb) at 10.7 and 18.7 GHz to constrain the Canadian Land Surface Scheme (CLASS), driven by meteorological reanalysis data and coupled with a simple radiative transfer model. The Tb polarization ratio (horizontal/vertical) at 10.7 GHz was selected to improve snowpack density, which is linked to the thermal conductivity representation in the model. Referencing meteorological station soil temperature measurements, we validated the approach at four different sites in the North American tundra over a period of up to 8 years. Results show that the proposed method improves simulations of the soil temperature under snow (Tg) by 64% when using remote sensing (RS) data to constrain the model, compared to model outputs without satellite data information. The root mean square error (RMSE) between measured and simulated Tg under the snow ranges from 1.8 to 3.5 K when using RS data. Improved temporal monitoring of the soil thermal state, along with changes in snow properties, will improve our understanding of the various processes governing soil biological, hydrological, and permafrost evolution.

The distribution of shallow frozen ground is paramount to research in cold regions, and is subject to temporal and spatial changes influenced by climate, landscape disturbance and ecosystem succession. Remote sensing from airborne and satellite platforms is increasing our understanding of landscape-scale permafrost distribution, but typically lacks the resolution to characterise finer-scale processes and phenomena, which are better captured by integrated surface geophysical methods. Here, we demonstrate the use of electrical resistivity imaging (ERI), electromagnetic induction (EMI), ground penetrating radar (GPR) and infrared imaging over multiple summer field seasons around the highly dynamic Twelvemile Lake, Yukon Flats, central Alaska, USA. Twelvemile Lake has generally receded in the past 30yr, allowing permafrost aggradation in the receded margins, resulting in a mosaic of transient frozen ground adjacent to thick, older permafrost outside the original lakebed. ERI and EMI best evaluated the thickness of shallow, thin permafrost aggradation, which was not clear from frost probing or GPR surveys. GPR most precisely estimated the depth of the active layer, which forward electrical resistivity modelling indicated to be a difficult target for electrical methods, but could be more tractable in time-lapse mode. Infrared imaging of freshly dug soil pit walls captured active-layer thermal gradients at unprecedented resolution, which may be useful in calibrating emerging numerical models. GPR and EMI were able to cover landscape scales (several kilometres) efficiently, and new analysis software showcased here yields calibrated EMI data that reveal the complicated distribution of shallow permafrost in a transitional landscape. Copyright (c) 2016 John Wiley & Sons, Ltd.