Between 1985 and 2023, across 175 monitored U.S. rivers, the median number of days required to carry 90 percent of the annual suspended-sediment load fell from 69 to 50. For most of those rivers the annual total had not risen. The delivery window had narrowed.
That result comes from Nirajan Sigdel and Andrew Husic, published July 23 in Communications Earth & Environment after clearing peer review on July 16. What they document is a compression of the sediment calendar: a measurable tightening, over nearly four decades, of the period during which rivers move most of their particulate load.
Suspended sediment is the material a river carries in its water column. Silt, clay, organic particles, whatever the current keeps aloft. It is what turns a river the color of coffee after a storm, and it is what a drinking-water treatment plant has to remove before the water reaches anyone's faucet. Across a substantial fraction of the monitored U.S. river network, that material now arrives in more concentrated pulses than it did in the mid-1980s.
For a plant drawing from one of these rivers, compression means the peak days carry more sediment per hour. The chemical feeds, settling basins, and filters inside those plants were sized against historical peaks. When the peak sharpens, the distance between what the plant can process and what the river delivers on its worst day gets shorter.
Where the ground changed first
The compression is uneven. Of 175 sites, 58 showed a statistically significant increase in temporal concentration, and the strongest signals came from small catchments. Sites showing both rising sediment volume and increasing burstiness had a median drainage area of 20 square kilometers, against 268 square kilometers for sites showing neither trend. Small watersheds respond fast to what happens on the surface. Larger basins have floodplains and channel storage that spread a pulse across more days before it reaches a gauge.
The study's clearest named example is Brushy Fork Creek near Loganville, Georgia, a small stream on the expanding edge of metropolitan Atlanta. Between 1989 and 2019, the modeled number of days carrying 90 percent of annual sediment fell from 178 to 61. Over the same three decades, urban land cover in the catchment rose from 21 percent to 68 percent, pavement and rooftops replacing soil that had absorbed runoff and released it slowly. The creek changed because the ground changed. Impervious surface took a diffuse sediment supply and concentrated it into sharper, shorter pulses. Precipitation on very wet days also rose over that period, from 261 to 332 millimeters annually, but across the study's full network, urbanization was the dominant modeled predictor of compression, not rainfall intensity.
Across the 58 sites with increasing temporal concentration, the attribution model assigned an average of 44 percent of the change to urbanization and 29 percent to forest-cover loss. Precipitation intensity pushed toward greater compression at 88 percent of those sites but dominated at only 10 percent. The authors call this predictor attribution rather than causal proof, and they list what the model leaves out: channel engineering, bank stabilization, wildfire history, agricultural management. The peer-review record shows the authors narrowing their geographic claims during revision, specifying that the conclusions apply most directly to sensor-monitored, predominantly eastern systems under human impact. Western rivers, many of them moving sediment on a snowmelt schedule, are largely absent from the dataset.
Inside the plant when the margin disappears
Brushy Fork Creek does not feed a drinking-water intake. Gwinnett County's plants draw from Lake Lanier, outside the Brushy Fork basin. But the operational question the study raises has an answer on the public record, from a system that draws straight from a river.
In May 2025, heavy rain drove turbidity up and alkalinity down in Virginia's James River. Turbidity measures how much particulate material is suspended in the water. Alkalinity is the water's capacity to buffer acids, which determines how well treatment chemicals do their job. Three plants drawing from the James faced the same deteriorating raw water. Two handled it. Richmond's did not.
The Virginia Department of Health investigation names Department of Public Utilities Director Scott Morris and reconstructs the sequence in unusual operational detail.
Higher turbidity and lower alkalinity made coagulation harder. Coagulation is the chemical step where operators add substances that cause suspended particles to clump together so they will settle out. Plate settlers, angled surfaces designed to catch settling particles, were already carrying accumulated sludge from earlier treatment cycles. They clogged. Settling performance dropped, and more particles moved downstream toward the filters.
Operators shut the plant to clean the plate settlers and empty the sedimentation basins. On restart they brought flow up through individual filters gradually, watching pressure buildup and the cloudiness of water leaving each unit, trying not to overwhelm filters already under strain. They delayed backwashing, the process of reversing flow through a filter to flush out trapped particles, because a backwash consumes finished water and the clearwells and distribution tanks were draining.
The constraint migrated. It began as a raw-water quality problem, became a settling-capacity problem, then a filter problem, then a storage problem. One distribution zone dropped below 20 pounds per square inch, which triggered a boil-water advisory. Restoring pressure was not enough to lift it; the utility also had to collect two rounds of bacteriological samples sixteen hours apart and wait for state concurrence.
Morris acknowledged publicly that both deferred maintenance and poor raw-water quality contributed. The state investigation confirmed that and sharpened it: ordinary high-turbidity operations should have been manageable with timely maintenance, closer process monitoring, and better chemical-feed adjustment. The two upstream plants demonstrated that the same raw water did not require a system to fail. Richmond's breakdown came out of the interaction between source-water stress and maintenance that had already been postponed, in a plant whose margins were thinner than the river was about to require.
That interaction is where compression matters operationally. A compressed calendar does not produce treatment failures on its own; it shortens the distance between normal operations and the edge of capacity. A plant that could absorb a turbidity surge spread over five days may not absorb the same mass arriving in two. Chemical inventory sized for historical peak-load days runs short when the peaks are higher and closer together. And a maintenance schedule that defers plate-settler cleaning until next month works until the month's sediment arrives in a week.
No reviewed source connects the James River specifically to the study's long-term compression trend. Richmond is not evidence that compression caused the May 2025 event. What the Morris account and the state investigation provide together is a detailed record of what the edge of capacity looks like from inside a plant, and of how three facilities reading the same river made different operational choices with different outcomes.
Who sees the pulse coming
The study reconstructed daily sediment records with machine-learning models trained on high-frequency turbidity sensors, discharge, weather, and land-cover data. They had to. Continuous four-decade sediment observations do not exist for most rivers, and the authors note that abrupt sediment pulses are hard for conventional monitoring to capture at all.
Federal rules require conventional plants to monitor each individual filter continuously and record turbidity at least every 15 minutes. Those rules govern filter effluent, the water leaving the treatment process. They confirm that treatment is working. They do not require anyone to watch the raw-water pulse coming down the river.
A plant can be in full regulatory compliance and still have no advance warning, no time to adjust chemical feeds, call in staff, clear settling capacity, or close an intake before the sediment arrives.
Some systems have built the warning themselves. On Oregon's North Santiam River, a cooperative USGS network records water quality at 15-minute intervals and transmits multiple times a day, giving Salem's operators lead time to prepare or shut the intake before highly turbid water reaches their slow-sand filters. During a December 2025 storm, Salem kept its intake closed for 13 days and bought roughly 25.3 million gallons from neighboring Keizer for $57,738, according to a damage assessment the city filed with its council in February 2026. Closing the door and buying water is an expensive option, and having it requires both the sensors and the neighbor.
Other systems run with less visibility. Maryland's source-water assessment for the Bloomington Water Plant, which draws from Savage River Reservoir, notes that during a September 1996 storm, reservoir turbidity exceeded 300 NTU (nephelometric turbidity units, the standard measure) and the plant shut down for two days. The same assessment records that turbidity readings were not taken every day at Bloomington, though periodic turbidity spikes were described as common. When the sampling interval is wider than the pulse, the event passes unrecorded, and the record that would justify tighter monitoring never accumulates.
A 2024 Water Resources Research analysis found one weekly sample sufficient to keep long-term load estimates within roughly 10 percent on a relatively stable river like the Rhine, while more variable systems needed higher frequency or storm-targeted sampling. USGS guidance on automatic samplers recommends triggering collection off turbidity or conductance rather than the clock, because a calendar-based schedule can miss an urban stream's first sediment pulse entirely. Greater temporal concentration raises the odds that fixed-interval sampling misses the days carrying most of the load. How badly depends on the river's flashiness, the monitoring objective, and the interval in use.
Uncertainty in the reconstruction
The 19-day contraction is a point estimate for the network median. When the authors propagated prediction uncertainty through a bootstrap analysis, the estimated contraction came out at 10 days. Smaller, same direction.
Model agreement with discrete sediment-concentration observations was moderate: a median R-squared of 0.43 for concentration and 0.69 for flux. (R-squared measures how much of the variation in the observed data a model reproduces; 1.0 would be exact.) The authors argue that sensor saturation at extreme concentrations, plus the smoothing of sub-hourly readings into daily values, likely flattens the peaks, which would make their burstiness result a floor rather than a ceiling. That is their reading of their own instrument, not an independently validated correction.
The omitted variables they name could produce or mask compression signals through mechanisms the model never sees. A channelized reach moves sediment faster. A stabilized bank releases less between storms. A reservoir traps load that would otherwise register at a downstream gauge. Whether accounting for these would strengthen or weaken the urbanization signal is untested. The paper is recent enough that no published counterargument exists yet, so the disagreement on the table is internal: between the 19-day point estimate and the 10-day bootstrap, and between a full-period trend prevalence of 33 percent and much weaker results when the record is split into subperiods. The authors attribute the subperiod difference to low statistical power in short records of a highly variable process. That is a reasonable explanation, and it also means the strongest version of the finding rests on the full modeled reconstruction rather than on the shorter sensor-era record alone.
A reservoir losing storage responds to total incoming sediment mass over years. A treatment plant responds to peak concentration on the worst day — which is what compression intensifies.
Only 15 percent of sites showed significant increases in both sediment volume and temporal concentration, which is why the same paper reads differently depending on which piece of infrastructure you are responsible for. A reservoir operator watching storage capacity disappear over decades and a plant superintendent watching a coagulant dose fail on a Tuesday are reading the same river against different clocks.
An earlier piece in this publication, "Soil Cannot Read a Bucket," followed a version of this distinction through precipitation and soil: annual totals matter less than the rate of arrival, because the receiving system has a capacity per unit of time, not per unit of year. The sediment study extends the same logic downstream. A coagulant feed rate, a settling-basin volume, a filter's hydraulic loading capacity — all of them are engineered against a peak. Compress the same annual mass into fewer days and the peak rises, whether or not the annual total moved at all.
The questions that stay local
The study does not tell a utility what to do. It says that the temporal pattern of sediment delivery in many U.S. rivers has shifted measurably over four decades, that urbanization is the strongest modeled predictor of the shift, and that the effect is most pronounced in small, rapidly developing catchments. Whether any specific source river shares the trend takes local data the network median cannot substitute for: gauge records, land-cover history, plant logs.
The questions that follow are operational. Is the chemical inventory sized against the historical peak-load day or the current one? Is the maintenance schedule calibrated to the old sediment calendar? Does the monitoring see the pulse before it arrives, or only after it is inside the plant? If the intake has to close, how many days of alternative supply exist, and at what price per million gallons?
Nineteen days over 38 years works out to about half a day a year. Each season's sediment calendar resembles the one before it closely enough that nothing in a single year prompts anyone to re-size a basin. The shift only exists in the record, and only for the rivers where someone has been keeping one. Richmond's two upstream neighbors answered the same surge without losing production, on the same day, from the same water. Whether that difference holds as the peaks sharpen is a question each utility gets to answer with its own numbers.
- Rain totals hiding drought: A May Nature study found that concentrating annual rainfall into fewer, heavier events reduced terrestrial water storage across most climates even when total precipitation was unchanged, because heavier downpours partition more water into runoff and evaporation rather than infiltration.
- Rivers staying hot longer: A July Nature Geoscience reconstruction of 796 river basins found that riverine heatwaves lasted an average of 2.67 days longer than the atmospheric heatwaves that triggered them, with minimum air temperature as the most important climatic predictor of the gap.
- Fire weather arriving in waves: A July study defined persistent periods of warm, dry, windy conditions as "fire weather waves" and found they accounted for 26 percent of burned area globally despite occurring on only 4 percent of days, with frequency increasing across most burnable lands during 1979–2024.
- Coastal floods at predictable hours: A July Nature Communications study found strong intraday clustering of recurrent coastal floods at tide-dominated U.S. and U.K. sites, with Boston events clustering around noon and midnight and Southern California events clustering in the morning.