128-17 From Reflectance to Sediment: Satellite-Based SSC Estimation Across Six Northeastern U.S. Rivers

Abstract:

Suspended sediment concentration (SSC) is a key indicator of watershed erosion, water quality, and aquatic habitat health, but it is expensive to measure directly, requiring repeat water sampling and lab analysis at each site. Despite seven decades of USGS water-quality monitoring across the Northeast (1956–2026, ~38,700 samples at 181 sites), SSC sampling remains sparse relative to the storm-driven, episodic nature of sediment transport: even sampled rivers average fewer than 3 samples per site per year, with coverage concentrated in New York and Pennsylvania and thin elsewhere. To address this, I pair Sentinel-2 satellite reflectance — how much light of different wavelengths bounces off the water's surface — with discharge-based SSC estimates across six Northeastern U.S. rivers of varying size, testing whether satellite data can extend sediment monitoring beyond what sparse sampling allows. Starting from 1,174 candidate USGS gages, I filtered to sites with drainage areas >100 mi² (a proxy for river width usable by 10m Sentinel-2 imagery), rating-curve model performance of R² > 0.6, and records extending past 2016. Applying these criteria, plus a Particulate Organic Carbon (POC) threshold  >30 samples and SSC >30 samples, narrowed the pool to six sites: the Ohio River, Mohawk River, Susquehanna River, Cattaraugus Creek, Swatara Creek, and White River. I advanced a Google Earth Engine pipeline to extract Sentinel-2 reflectance at each site. I then fit a regression model for each river using visible and near-infrared reflectance to predict SSC. These site-specific prediction models were all statistically significant and generally successful, with  R² 0.37–0.78. Their predictive power was strongest at large mainstem rivers (Ohio River, R² = 0.78; Mohawk River, R² = 0.69) and weakest at smaller streams (Swatara Creek, R² = 0.49; White River, R² = 0.37), suggesting channel size and land cover affect how well reflectance may capture sediment signal. Per-site models explained substantially more variance than a single model fit to combined data from all six rivers. These results indicate that further work is required to create  a Sentinel-2-based SSC algorithm that can be generalized across diverse river systems. However, site-specific calibration meaningfully improves satellite-based SSC estimation, offering a path to fill the sparse ground-truth record with continuous satellite-derived estimates and extending sediment monitoring to where direct sampling is rare.

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.