Why Are Some Watersheds More Sediment-Productive Than Others? An Explainable AI Approach

The Sediment Delivery Ratio (SDR) represents the proportion of eroded sediment within a watershed that ultimately reaches its outlet. Sediment yield (SY) is defined as the amount of sediment exported from a watershed outlet per unit time, normalized by drainage area. Quantifying the spatial variability of both metrics is critical for sediment management and water quality protection, yet the drivers of SDR remain poorly constrained at continental scales. Here, we develop a data-driven framework to model SDR and SY across the contiguous United States (CONUS) by integrating high-frequency aquatic sensing, sediment load estimation, and explainable machine learning. SY and SDR were quantified at 134 U.S. Geological Survey (USGS) stations and modeled using a random forest algorithm trained on 20 basin attributes, spanning climate, land cover, soils and geology, and topography. The model performed well in the estimation of SY and SDR, yielding R2 values of 0.59 and 0.60, respectively. Model interpretation using Shapley values revealed that anthropogenic factors—including urban area, pasture area, and road density—exert stronger controls on SDR than natural drivers such as slope or precipitation, while the opposite relationship was observed for SY, where natural drivers prevailed over anthropogenic factors. We extended the model to the spatial variability of SY and SDR in two basins of major interest (the Upper Mississippi River Basin and the Chesapeake Bay Basin), identifying highly connected and productive basins to be targeted for management intervention. This study provides a large-scale, explainable framework for predicting SDR and SY using remote sensing and watershed attributes, offering new insights into the spatial controls of sediment delivery and supporting the design of targeted sediment mitigation strategies across heterogeneous landscapes.

Name thesis:degree_name

Master of Science

Level thesis:degree_level

masters

Discipline thesis:degree_discipline

Civil Engineering

Department dc:contributor.department

Civil and Environmental Engineering

Grantor dc:publisher

Virginia Tech

Year dc:date.issued

2026

Author dc:creator

• Shrestha, Sugam

Chair dc:contributor.committeechair

• Husic, Admin

Committee members dc:contributor.committeemember

• Strom, Kyle Brent
• Stewart, Ryan D.

• Sediment yield
• Sediment delivery ratio
• High frequency sensors
• Turbidity
• explainable AI
• targeted mitigation

Statement dc:rights

• In Copyright

Licence dc:rights.uri

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language.iso

en

Dc Identifier Other

vt_gsexam:45436

OAI identifier oai:identifier

oai:vtechworks.lib.vt.edu:10919/140899