Land surface temperature trends derived from Landsat imagery in the Swiss Alps

Abstract. The warming of high mountain regions caused by climate change is leading to glacier retreat, decreasing snow cover, and thawing permafrost, which has far-reaching effects on ecosystems and societies. Landsat Collection 2 provides multi-decadal land surface temperature (LST) data, principally suited for large-scale monitoring at high spatial resolution. In this study, we assess the potential to extract LST trends using Landsat 5, 7, and 8 time series. We conduct a comprehensive comparison of both LST and LST trends with data from 119 ground stations of the IMIS network, located at high elevations in the Swiss Alps. The direct comparison of Landsat and IMIS LST yields robust satellite data with a mean accuracy and precision of 0.26 K and 4.68 K, respectively. For LST trends derived from a 22.6-year record length, as imposed by the IMIS data, we obtain a mean accuracy and precision of -0.02 K yr^-1 and 0.13 K yr^-1, respectively. However, we find that Landsat-LST trends are biased due to unstable diurnal acquisition times, especially for Landsat 5 and 7. Consequently, LST trend maps derived from the 38.5-year Landsat data exhibit systematic variations with topographic slope and aspect that we attribute to changes in direct shortwave radiation between different acquisition times. We discuss the origin of the magnitude and spatial variation of the LST trend bias in comparison with modelled changes in direct shortwave radiation and propose a simple approach to estimate the LST trend bias. After correcting for the LST trend bias, remaining LST trend values average between 0.07 and 0.10 K yr^-1. Further, the comparison of Landsat- and IMIS-derived LST trends suggests the existence of a clear-sky bias, with an average value of 0.027 K yr^-1. Despite these challenges, we conclude that Landsat LST data offer valuable high-resolution records of spatial and temporal LST variations in mountainous terrain, where monitoring is generally difficult. Our study highlights the significance of understanding and addressing biases in LST trends for reliable monitoring in such challenging terrains.