Shean, DavidBrencher, George2026-02-052026-02-052026-02-052025Brencher_washington_0250E_29170.pdfhttps://hdl.handle.net/1773/55182Thesis (Ph.D.)--University of Washington, 2025Mountainous regions store critical water resources and produce devastating natural hazards. As climate change disproportionately impacts mountainous regions, accurate and timely observations are needed for adaptive resource and hazard management and to understand changing cryospheric and geomorphic processes. Synthetic aperture radar (SAR) can provide these observations, but SAR-based measurements are subject to noise and errors that reduce their reliability. Relying on the extensive SAR archive, this dissertation develops workflows that integrate emerging data science approaches, including deep learning, with established geophysical methods to improve SAR-based measurements of surface movement and snow depth in mountainous terrain.In Chapter 1, I used a convolutional neural network (CNN) to remove atmospheric noise from interferometric synthetic aperture radar (InSAR) interferograms. The CNN was trained using thousands of Sentinel-1 interferograms and exploits differences in the spatial and topographic structure of atmospheric noise and deformation signals, without relying on external atmospheric data. This approach outperforms commonly used atmospheric correction methods and reveals previously obscured centimeter-scale deformation of rock glaciers and landslides in the Rocky Mountains. These improvements enable more reliable interpretation of subtle surface kinematics in high-relief terrain. In Chapter 2, I developed a fused InSAR and SAR feature tracking approach to quantify surface displacement of moraines damming glacial lakes. Combining InSAR and feature tracking results in improved displacement time series that are more accurate than those produced using either method alone. Application to the Imja Lake moraine dam in Nepal reveals decimeter-scale cumulative subsidence over a seven-year period and widespread buried ice. I validated these results using very-high-resolution satellite stereo digital elevation models. The observed displacement patterns are consistent with year-round ice flow and warm-season ice melt. These results provide new constraints on the processes contributing to moraine dam degradation and have direct implications for glacial lake outburst flood (GLOF) hazard assessment. In Chapter 3, I extended this approach to the 23 moraine-dammed glacial lakes in Nepal which are the highest priority for monitoring. I used seasonal change in InSAR coherence as a proxy for buried ice presence. I found that most moraine dams contain buried ice that produces surface displacement of centimeters to tens of centimeters per year. Analysis of displacement components indicates that the observed deformation reflects a combination of ice melt and ice flow, with the relative contribution of each process varying between sites. I found evidence for extensive buried ice in several moraine dams previously classified as ice-free, which substantially changes the conclusions of prior hazard assessments. These results can be used to improve GLOF hazard assessments and modelling studies. In Chapter 4, I developed a deep-learning approach for regional snow depth prediction across the Western United States. I trained a U-Net CNN using a large archive of airborne lidar snow depth measurements and multi-modal inputs including SAR backscatter, optical imagery, topography, and coarse-resolution physical model outputs. The final CNN substantially outperforms existing approaches for near real-time prediction of Western U.S. snow depth in accuracy, precision, and resolution. It can be applied to create spatially continuous maps of snow depth over the entire Western U.S. and dense snow depth time series over the past decade. This work establishes a new benchmark for regional snow depth prediction performance, with implications for future operational forecasting.application/pdfen-USCC BYdeep learningglacial lake outburst floodInSARremote sensingrock glaciersseasonal snowRemote sensingGeomorphologyGeophysicsCivil engineeringThesis