Publication Date

5-2025

Date of Final Oral Examination (Defense)

2-25-2025

Type of Culminating Activity

Dissertation

Degree Title

Doctor of Philosophy in Geophysics

Department

Geosciences

Supervisory Committee Chair

Ellyn Enderlin, Ph.D.

Supervisory Committee Member

Hans-Peter Marshall, Ph.D.

Supervisory Committee Member

Shad O'Neel, Ph.D.

Supervisory Committee Member

Alejandro Flores, Ph.D.

Abstract

Warming air temperatures in recent years have led to declines in seasonal snow pack across western North America. These changes have profoundly impacted communities that depend on seasonal snow melt for water resources, as well as glaciers that act as critical long-term reservoirs in the region. While global snow cover extent products are available, they often fail to reliably distinguish between snow, glacier ice, and clouds, which are prevalent in western North America. Additionally, water managers in the region typically rely on point-based observations of snow mass for decision-making, despite the pressing need for comprehensive, watershed-scale estimates. To address these challenges, my dissertation focuses on advancing optical remote sensing methods for snow monitoring and analysis.

I first developed an automated snow cover mapping workflow that uses Landsat 8/9, PlanetScope, and Sentinel-2 images for glaciers across western North America. The workflow produces ~weekly estimates since 2013 of key glacier mass balance indicators: the snow-covered area, transient accumulation area ratio (fraction of the glacier covered by snow), and snowline altitude. To develop the classifiers, I manually labelled over 10,000 image pixels as snow, ice/firn, rock, or water at the U.S. Geological Survey's benchmark glaciers, then used the labeled pixels to to train, test, and validate several machine learning models. The optimal models for each image product––Nearest Neighbors for Landsat 8/9 and PlanetScope and Support Vector Machine for Sentinel-2––achieved overall accuracies of 92–98%. While intermittent cloud and shadow cover introduce errors in individual classifications, the time series effectively capture weekly trends in seasonal snow cover. These observations allow for precise identification of the timing and extent of snow cover minima, offering improved metrics for assessing glacier mass balance. Additionally, monitoring snowline shifts throughout the season may inform major melt events, providing insights for flood risk assessments and local water resource management. With its demonstrated accuracy and scalability, this workflow has the potential to be applied across broad spatial scales, advancing regional efforts to monitor glacier snow dynamics and their downstream impacts.

Next, I applied the snow cover mapping workflow to 200 glaciers spanning a wide range of geographic, terrain, and climate characteristics in western North America to assess potential biases in current glacier monitoring and modeling approaches. The aggregated time series revealed that the timing of minimum snow cover extent varied by up to three months across the region. This range in timing contradicts the common approach of using late-summer observations to estimate glacier snow minima, particularly at the most northerly sites. By comparing the snow cover time series to a global glacier model, I show that the model typically overestimates seasonal snow melt, potentially leading to biased mass balance projections. I argue that integration of the snow cover mapping workflow across the full melt season can help future field work planning, improve the accuracy of glacier mass loss assessments, and assist with model calibration frameworks for improved glacier mass projections.

Finally, I explored the capacity of SkySat triplet stereo imagery to estimate and model snow depth using terrain characteristics. The SkySat constellation from Planet Labs can be tasked to acquire images on demand, making it a potential tool for repeat snow depth mapping in remote regions. In this study, SkySat imagery was collected in three watersheds in southeast Idaho in May and October 2024. Digital elevation models (DEMs) were constructed from the imagery using the Ames Stereo Pipeline and a modified version of the SkySat Stereo pipeline. Snow depth was estimated by differencing a snow-free DEM from the SkySat DEMs. To investigate the potential for spatial extrapolation of snow depth estimates, I trained and tested Random Forest regression models using terrain characteristics derived from the snow-free DEM. Elevation errors on stable surfaces of the resulting DEMs ranged from 0.01 to 1.23 m across all sites and DEMs, indicating that snow depths below these thresholds may not be captured. Additionally, the SkySat-derived snow depths about half to several meters lower than the local snow telemetry observations and near-coincident airborne lidar surveys. Thus, future work will focus on improving the DEM and snow depth map construction workflow. With improvements, SkySat DEMs hold potential for estimating and extrapolating snow depth observations at watershed scales, filling critical gaps in snow depth observations.

The advancements outlined in this dissertation will help to address the long-term goal of standardized, global, and regularly updated snow extent and snow water equivalent (SWE) products for large-scale assessments of snow water resources. Achieving this vision will likely require integrating observations from multiple platforms. For example, future work could incorporate the glacier snow cover maps detailed here with snow melt maps derived from radar sensors like Sentinel-1, providing validation and deeper insights into seasonal melt dynamics and glacier mass balance. Similarly, combining radar, laser altimetry, and/or very-high-resolution optical imagery, such as SkySat, could address challenges in SWE estimation across complex terrain, particularly where radar struggles with wet snow conditions. Finally, data assimilation techniques that merge observations from optical and radar platforms with snow pack evolution models offer a promising path forward for improving the spatial and temporal coverage of global SWE estimates.

Comments

ORCID, 0000-0002-8497-4253

DOI

https://doi.org/10.18122/td.2340.boisestate

Download

Files over 30MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS