Refining debris-covered glacier outlines using land surface temperature data

Accurate mapping of debris-covered glaciers (DCGs) is essential for understanding their dynamics, contribution to sea-level rise and role in regional hydrology. Existing glacier mapping efforts suffer from inaccuracies at the local scale, especially for DCGs. Current approaches often fail to distinguish debris from surrounding terrain due to spectral similarities. In many cases, manual delineation is required, which is time consuming and not suitable for large scale mapping. This thesis developed a methodology to refine DCG outlines using land surface temperature (LST) data, based on the assumption that glacial debris shows a cooler thermal signature influenced by the underlying ice. The Approach Integrates thermal and optical data from Landsat 5 and 8, topographic information from three digital elevation models (Copernicus DEM, NASADEM, and SwissAlti3D), and outlines of existing glacier inventories, including the Swiss Glacier Inventories (SGI2010 and 2016) and the Randolph Glacier Inventory version 7 (RGI 7.0). A novel index combining LST and near infrared (NIR) data was integrated with topographic information and spectral indices in a random forest classifier. The transferability and robustness of the classifier across different regions was enhanced by using multiple combinations of input variables, which were subsequently integrated into a final classification. The method was calibrated on Zmuttgletscher in the Swiss Alps and extended to Oberaletschgletscher and Unteraargletscher. Temporal analyses for 2003, 2010 and 2016 assessed its robustness over time, while additional applications on Belvedere Glacier (Italy) and Satopanth Glacier (India, Himalayas) evaluated its adaptability to different climatic regions. Statistical comparison with the SGI and RGI inventories yielded validation accuracies between 0.743 and 0.929, often exceeding 0.85. While challenges remain in terms of accuracy and robustness, the method effectively delineates DCG outlines across different glaciers and time periods, providing a scalable and automated approach to DCG mapping. The automated nature of the approach and the use of open- source data make the method a valuable tool for DCG mapping on a global scale. With upcoming satellite missions such as TRISHNA, SBG and LSTM expected to provide higher resolution thermal data, the methodology is poised for further refinement and wider application. This research advances the use of LST in glaciology and provides a practical tool for improving global glacier inventories.