ABSTRACT:

This data product is related to a journal article by Nishchal Sigdel and Admin Husic entitled "U.S. rivers are transporting more suspended sediment, often in less time". The article is under review as of July 2026.

This resources includes the trained LSTM model weights, the processed turbidity-derived training targets, and the Python scripts used to compute the Gini coefficient and B90 metrics.

Abstract:
Riverine suspended sediment transport is highly episodic, yet how the timing and magnitude of these bursts have shifted under climate and land-use change remains uncertain. Here, we integrate high-frequency turbidity sensing with deep learning to reconstruct nearly four decades of daily sediment flux for 175 U.S. rivers. We apply a temporal inequality framework to quantify multi-decadal trends in sediment timing alongside magnitude. Annual sediment yields have risen at 28% of rivers and export has become time-compressed at 33% of rivers, with the network-wide median days required to deliver 90% of the annual load dropping from 69 in 1985 to 50 in 2023. These trends in magnitude and timing are partially decoupled, as only 15% of sites show both, and they are governed by distinct attributed drivers. Land-use change is the dominant predictor of temporal compression, while intensifying precipitation is the dominant predictor of rising sediment yields. Both trends are most pronounced in small, urbanizing catchments. The result is a narrowing window for monitoring and intervention, and a shift in geomorphic and infrastructure hazards toward rarer, higher-impact events.