Quantifying the Effects of Volcanic Snow Cover on InSAR Coherence Using a Computationally Inexpensive Neural Network

• Published in: Journal of geophysical research. Machine learning and computation Vol. 2; no. 3
• Main Authors: Bussard, R. C., Dufek, J., Wauthier, C., Townsend, M.
• Format: Journal Article
• Language: English
• Published: 01.09.2025
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2024JH000553

Interferometric Synthetic Aperture Radar (InSAR) measurements constrain ground movement across volcanic systems where the signal coherence remains high enough to retrieve quality phase information. A common phenomenon that degrades this coherence at volcanoes is snow cover. To study snow cover's effect on InSAR signal coherence at a volcanic system, we train a computationally inexpensive neural network using an optical image data set composed of 29 Landsat‐8 images spanning 2014–2021 to identify snow cover across the Mount St. Helens region through time. Across the slopes of Mt. St. Helens, snow cover remains high throughout most of the year excluding June through September, while the lower altitude and less vegetated pumice plain emplaced in 1980 only seasonally experienced snow cover. We compare identified snow cover with selected coherence images from a Sentinel‐1 InSAR data set spanning 2014–2020 and find that average coherence can decrease up to 70% when snow cover is present. Forward modeling using an inflating magma reservoir simulated as a point source in an elastic half‐space shows that most centimeter or above scale simulated ground movement is hidden by snow, as commonly present from December to March. This indicates potential difficulties for obtaining enough signal to constrain subsurface processes at Mount St. Helens and other volcanoes that experience consistent snow cover. Satellites are often used to monitor volcanoes in preparation for eruptive events. When snow covers a volcano and the surrounding area, it can become difficult for satellites to image the ground beneath the snow, making these volcanoes harder to monitor. By using a simple neural network to determine the extent of snow cover at Mt. St. Helens during different periods of the year, we can understand which seasons are the most difficult to retrieve signal from Mt. St. Helens and which patches of the volcano and its nearby area experience the most consistent snow cover. Understanding the extent of snow cover at Mt. St. Helens allows us to simulate ground movement related to magmatic activity and observe how much of the ground movement satellites would be able to measure with and without snow present. Snow cover drastically reduces Interferometric Synthetic Aperture Radar coherence values of bare rock areas around Mt. St. Helens from 1.0 to 0.2–0.3, masking deformation Snow cover at Mt. St. Helens can reduce centimeter scale deformation from an inflating point source by 40%–90% Machine learning analysis applicable to the geosciences can be scaled to use minimal computing resources promoting accessibility