Glacier Retreat Monitoring

Summary

This report assesses glacier retreat across ten globally distributed sites from 2017 to 2024 using Sentinel-2 Level-2A imagery. It compares two complementary delineation approaches: NDSI thresholding with automated Otsu segmentation, and a U-Net model trained on manually outlined glacier boundaries. The analysis quantifies areal change, annual retreat rates, uncertainty, and agreement between methods across contrasting glacier settings. Results indicate measurable glacier area loss across the sample, with stronger retreat at some sites than others and strong segmentation performance from the deep-learning workflow. The study is designed as a repeatable framework for large-scale glacier monitoring under changing climatic conditions.