ABSTRACT:

Code accompanying Percich et al., 2026. "Upscaling sediment source prediction for watershed management"

Authors: Abigal Percich (1), Allen Gellis (2), James Fox (3), and Admin Husic (1)*

(1) Department of Civil and Environmental Engineering, Virginia Tech
(2) Department of Atmospheric, Oceanic & Earth Sciences, George Mason University
(3) Department of Civil Engineering, University of Kentucky

*Admin Husic, [husic@vt.edu], 9408 Prince William St., Occoquan Watershed Monitoring Laboratory, Virginia Tech, Manassas, VA 20110

This HydroShare repository contains the complete dataset, modeling code, and regional application results for the study "Upscaling sediment source prediction for watershed management" (Percich et al., 2026). The collection provides a global meta-analysis synthesis and a Multivariate Random Forest (MVRF) machine learning framework designed to predict the proportional contributions of four primary catchment sediment sources: Subsurface, Cultivated, Non-Cultivated, and Infrastructure.

Repository Structure:
1. Data Preparation: Delineates watersheds used in the model and analyzes meta-analysis data.

2. Model Development: Trains multivariate random forest (MVRF) model and conducts Shapley feature importance.

3. Model Application: Evaluates model applicability in new basins and applies the model.

4. Supplemental Excel Files: Tabulated source fingerprinting metadata including study sampling techniques, mixture classification, tracer types, reported catchment areas, and DOI bibliographic reference tables.