Generation of an improved land surface temperature time series to support permafrost modelling in the northern high latitudes

Abstract

Northern high latitudes have experienced pronounced warming throughout the last decades with particularly high temperatures during winter and spring. Due to Arctic Amplification, the Arctic region is warming thrice as fast as anywhere else. The warming affects the sensible ecosystem, vegetation dynamics and the cryosphere (sea ice, snow and permafrost). Permafrost, which is a crucial component of arctic ecosystems, is particularly sensitive to increasing air temperatures and changes in the snow regime. Climate change has a high impact in these regions because thawing affects the stability of the bedrock, damages infrastructures and releases massive quantities of organic carbon. Permafrost cannot directly be observed from space, but permafrost models link physical surface variables such as land surface temperature (LST) to the thermal ground regime. Models are an important addition to boreholes to monitor the status of the permafrost at hemispheric scale. On a global scale, observation of LST is only available from very few in-situ stations or climate models with coarse spatial resolution. Both data sources are not sufficient to model fine-scaled features. In contrast, LST information retrieved from satellite data has high spatiotemporal coverage.To compute LST on a hemispheric scale, we use the Advanced Very High Resolution Radiometer (AVHRR) Global Area Coverage (GAC) data set starting in 1981. The AVHRR on board the NOAA and MetOp satellite series now covers more than four decades. AVHRR’s two thermal infrared channels allow applying the split-window (SW) method to reduce the atmospheric effect and retrieve LST. Split-window algorithms (SWA) performances depend on the quality of SW coefficients. These are empirical coefficients, which are retrieved by fitting the SWA to a calibration database. Here, the calibration data is generated by running a radiative transfer (RT) model. The input profiles of the RT are selected to cover typical atmospheric conditions occurring in permafrost regions. The coefficients are adjusted for different water vapour and satellite viewing conditions. Cloud and water masks as well as fractional snow cover information from the ESA CCI snow project and emissivity data are included in the final LST retrieval algorithm. Besides, a machine learning algorithm was applied to improve the spatial resolution of the GAC data to generate a 40-year time series with a spatial resolution of 1km. The first validation results of the LST time series are shown.