Introduction

A thorough understanding of fluvial sediment transport is essential for addressing key environmental issues such as aquatic habitat degradation, flooding, excess nutrients, and challenges with river restoration. Fluvial sediment samples are valuable for addressing these concerns, but their collection is often impractical across all rivers and timeframes of interest. In addition, previously used analytical and numerical methods (Gray and Simões, 2008; Ellison and others, 2016) have not allowed for the transfer of knowledge from sites that have data to sites that do not have data. To overcome this limitation, the U.S. Geological Survey (USGS) developed machine learning (ML) models to predict suspended-sediment concentrations (SSCs) and bedload transport (BL) in Minnesota rivers that lack physical sediment data.

Objective

The USGS's primary goal of developing ML models was to enhance sediment transport predictions and provide insights into sediment transport dynamics in Minnesota rivers. These ML models were trained using physical sediment samples, streamflow time series, and geospatial datasets describing watershed, catchment, near-channel, and channel characteristics.

Machine Learning Models for Fluvial Sediment Prediction

The USGS trained and validated the ML models with approximately 1,300 SSC samples from 56 sites and 600 BL samples from 43 sites across Minnesota (Lund and Groten, 2022; Lund and others, 2022). The ML models incorporate key features such as streamflow, watershed and catchment characteristics, and rate of change in streamflow (slope), which help explain sediment transport processes. The ML models explained approximately 70 percent of the variability in the SSC and BL samples (Lund and Groten, 2022; Lund and others, 2022). These ML models improved sediment transport predictions for rivers and streams with little or no physical sediment data by leveraging the ability of ML to learn complex nonlinear relations and transfer knowledge from sites with data to sites without data available.

StreamStats Integration

In response to the complexity of these ML models and the challenges of applying them, the USGS collaborated with the Minnesota Pollution Control Agency to integrate the ML models into the USGS StreamStats web application. This integration is the first time ML models for sediment prediction have been incorporated into StreamStats, making them accessible by way of a user-friendly interface.

Using StreamStats Fluvial Sediment Prediction Tool

StreamStats users can now access predictions of SSC and BL at Minnesota rivers that lack physical sediment samples. The tool provides three options for streamflow data input used to run the ML models:

• • Select a USGS streamgage—Use existing streamflow data from a USGS streamgage.

• • Upload streamflow data—Upload a comma-separated values file with daily or 15-minute streamflow data. Additional streamflow data can be found on the Minnesota Department of Natural Resources Cooperative Stream Gaging website: https://www.dnr.state.mn.us/waters/csg/index.html.

• • Estimate streamflow using flow duration curve transfer method—Estimate streamflow when data are unavailable.

Once users complete the required steps to run the ML models in StreamStats, the features used and the corresponding outputs will be available in graphical and comma-separated values formats.

Sediment Monitoring in Minnesota

Despite the critical need for fluvial sediment data, funding for monitoring programs has fluctuated throughout the years, affecting the continuity and scope of sampling efforts. After a decline in the 1990s (Ellison and others, 2014), funding for sediment monitoring in Minnesota surged with the approval of the Clean Water, Land and Legacy Amendment in 2008, which established the Clean Water Fund (Minnesota’s Legacy and Trust Funds, 2025) enabling the expansion of the USGS sediment-monitoring network in collaboration with the Minnesota Pollution Control Agency and the Minnesota Department of Natural Resources. However, as funding priorities shifted toward restoration in recent years, the number of monitoring sites decreased, and by 2025, no active USGS sediment-monitoring sites remain in the State. Despite these challenges, the long-term records collected by the USGS remain essential for understanding trends in sediment transport, informing water-quality standards, and supporting river restoration efforts. Continued commitment to fluvial sediment sampling is crucial to maintaining these ML models and advancing effective management strategies.

Summary

The incorporation of ML models into the USGS StreamStats application represents an important innovation in predicting fluvial sediment transport. These ML models provide a powerful tool for researchers, water managers, and policymakers, enabling more efficient decision making regarding water quality, habitat preservation, and river restoration in Minnesota.

For More Information