Date of Award:

12-2025

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Computer Science

Committee Chair(s)

Soukaina Filali Boubrahimi

Committee

Soukaina Filali Boubrahimi

Committee

Shah Muhammad Hamdi

Committee

John Edwards

Abstract

Forecasting river flow is essential for managing water supplies, reducing flood risk, and supporting healthy ecosystems. In the Upper Colorado River Basin, much of the yearly water comes from melting snow. However, many traditional models struggle to capture how snowpack and river flow interact, especially across such a large and complex region.

This study uses a modern machine learning approach called a Spatio-Temporal Graph Neural Network (STGNN) to improve streamflow prediction. The model uses Snow Water Equivalent (SWE)—a measure of how much water is stored in the snowpack—along with river flow data. By treating each river gauge as part of a connected network, the STGNN learns how conditions in one part of the basin affect flow in other areas downstream.

Using 30 years of historical data, the results show that adding SWE information significantly improves prediction accuracy by about 12.8%. The model is especially effective during spring and early summer, when melting snow drives most of the river’s flow. The greatest improvements appear at mid- and high-elevation stations, where snowpack changes strongly influence when and how much water enters the river.

Overall, this work demonstrates that combining snowpack data with advanced network-based learning provides a powerful tool for forecasting streamflow in snow-dependent regions like the Upper Colorado River Basin. This approach can help water managers make better decisions as climate change and declining snowpack continue to challenge water availability in the western United States.

Download

Share

COinS

Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .