Abstract
Ground surface temperature (GST), measured at a depth of around 5 cm below the ground surface, is essential for understanding the climate change impacts in the Earth Critical Zone. Large spatiotemporal variations of GST have been reported in mountain regions due to the heterogeneity of surface cover and topography. This work aims to improve the monitoring of GST using a physical land-surface model driven by satellite-based land surface temperature (LST). In this regard, GST was simulated using the physical GEOtop model at 1500 m elevation in Matsch Valley, north-eastern Italian Alps, from 2014 to 2017 during the phenological cycle, between April and October. The model was forced only by the LST derived from the Terra MODerate resolution Imaging Spectroradiometer (MODIS). The 1-km MODIS LST was first downscaled to a finer spatial resolution of 250-m using data-driven sharpening from random forest algorithm. The simulated GSTs correlate well with the in-situ observations with a Pearson correlation of 0.88 and a coefficient of determination of 0.77. However, the model overestimated the GST for the whole period with a mean bias of 8.72 °C. These overestimations are similar to the differences between in-situ GST and MODIS LST which range from 4.8 to 19 °C with an average of 8.5 °C. They are mainly caused by the low temporal resolution of LST data with only one observation per day which is additionally limited by frequent cloud cover contamination and the low spatial resolution of the MODIS thermal channels. Modelling the damping of the LST signal in the first centimeters of soil to simulate GST in very heterogeneous areas like alpine pastures is still challenging. This is mainly due to the resolution mismatch between ground and remote sensing observations and the poor knowledge of soil and vegetation properties needed to parametrize physical models.