Towards physical habitat characterisation in the Antarctic Sor Rondane Mountains using satellite remote sensing

Ice-free areas occupy less than 0.25% of the Antarctic surface, and mainly occur along coastlines, or as inland nunataks protruding from the extensive ice sheet. Their extreme environment and geographical isolation have contributed to the evolution of highly adapted, and largely endemic terrestrial biota. Physical habitat mapping is important to identify the main drivers of spatial variation in soil biodiversity, to predict its response to climate change and to help conservation planning. In this paper we retrieved remotely sensed Land Surface Temperature (LST) and Digital Surface Models (DSM) for the Sor Rondane Mountains in East Antarctica from respectively the Thermal InfraRed Sensor (TIRS) on Landsat 8 and the Pl eiades constellation of high resolution optical imagers. Satellite data were combined with ground truth temperature and elevation measurements with the aim to assess the performance of these remotely sensed data. Over a 2 year period, satellite derived LST corresponded to in situ temperature with Mean Average Difference (MAD) of 2.5 K, and Root Mean Squared Difference (RMSD) of - 6.3 K. Lower biases were observed for periods when the data loggers were frozen or snow covered (MAD 1.8K) compared to intervals where the devices were not frozen (MAD 4 K), with larger scatter (RMSD 7 K) being observed during the latter. These larger MAD and RMSD are caused by the differential heating of the top of the gravel/rocks as observed by the satellite compared to their underside, where the loggers were installed, and are also impacted by the time step of the in situ logging (3 h) and the non-linear heating of the surface. In addition, for the different study sites different biases were observed as a result of the spatial resolution of the TIRS, depending on the composition, structure, geomorphology, and the surroundings of the logger position. Sites on a narrow rock outcrop (e.g. the Perlebandet and Utsteinen nunataks) show a negative bias due to the surrounding ice fields decreasing the satellite pixel average temperature compared to the in situ measured temperatures. For sites with a more extensive rock and gravel composition further away from the ice sheet (Dry and Yuboku Valleys), a positive bias was found as a result of the temperature differential between the exposed top and covered bottom of the rocks and gravel. The DSM derived from Pl ' eiades (without bias correction) showed in general a bias of - 6 m (MAD) and scatter of - 9 m (RMSD) when compared to the Reference Elevation Model of Antarctica and IceSAT-1 LIDAR data. Lowest errors were also found here for the extensive Dry and Yuboku Valley sites (respectively - 5 and 3 m MAD, 7 and 4 m RMSD). Correspondence with in situ logged GPS positions was slightly worse, with a positive bias of 12 m (MAD) and scatter of 15 m (RMSD) across the sites. An analysis on the basis of these LST and DSM datasets, show a clear separability of sites by their average temperature and solar exposition (in terms of slope and aspect) at the time of Landsat overpass. Due to the Landsat overpass times, the LST datasets are however strongly biased, and further work is needed for a better understanding of the light and temperature climate in the ice-free regions. This could for example be achieved by coupling the DSM to a solar irradiance and terrain shadowing model