Simulating ground temperatures in mountainous terrain with remote sensing observations and modeling

The warming of high mountain regions caused by climate change leads to decreasing snow cover and thawing permafrost, which has far-reaching effects on ecosystems and societies. In this study, we used historical observations of land surface temperature (LST) derived from Landsat satellite data in conjunction with 1-D thermal modeling to simulate the annual evolution of ground temperatures. Our model includes precipitation and the evolution of snow cover, which exerts important control on ground temperatures. We tested this approach in the European Alps, where snow depth observations from high-elevation sites in Switzerland allowed us to evaluate the performance of different precipitation data sets and a simple scheme for snow removal by avalanches. We assessed our model results by comparing them with existing temperature measurements at boreholes and with Landsat-derived snow cover frequencies. All our analysis is based on daily conditions, but averaged over multiple years, hence neglecting interannual variability. Preliminary results indicate generally good agreement between modelled and observed values at weather stations, depending on the used precipitation data set and the sensitivity of the simple snow avalanche scheme. Typical root-mean-square errors (RMSE) are (1) ~2.7 K for daily ground surface temperature, (2) ~50 cm for daily snow depth, and (3) ~25% for monthly snow cover frequency. Typical RMSE for borehole temperatures are ~2 K at depths >1 m, and somewhat higher, at ~4 K for shallow depths (<1 m). It is worth noting that the global statistics is unevenly distributed, with some sites showing much larger errors than others. Besides deficiencies in the modeling, this could also be related to steep spatial gradients in LST, not well captured by the coarser resolution of the Landsat series (60-120 m) in comparison with the ground observations. Unless the used precipitation data sets exhibit regional bias or the snow avalanche model requires regional tuning, the presented approach is independent from ground-truth measurements and can in principle be applied anywhere on Earth. The model can be used to infer ground temperatures and its likely changes as a function of changes in LST and snow cover.