A riverbed can make a signal without giving you a sediment mass
A gravel particle strikes a plate in a riverbed. The plate vibrates, a sensor records a waveform, and software counts selected features. The record may reveal when coarse sediment is moving. Turning it into kilograms transported is a further step, and that step depends on particles, hydraulics, sensor behaviour and calibration.
This guide explains that conversion without treating every pulse as one stone. It distinguishes physical impact, recorded vibration, detected event and inferred mass. Those distinctions become especially useful during floods, when transport can be intermittent, signals can overlap and the bed around the sensor may change.
The numerical examples are fictional checks on interpretation, not a device calibration or installation manual. A plate geophone, a waterborne hydrophone and a remote seismic instrument do not observe identical signals. We compare their reasoning where useful without pretending that one equation or threshold can be copied between them.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1: Follow the signal chain · Chapters 1–3
Route 2: Understand missed and merged events · Chapters 4–5
Route 3: Convert a surrogate to mass · Chapters 6–8
Calibration connects a defined signal to a defined reference
Route 4: Check time and cross-section scaling · Chapters 9–10
Route 5: Test uncertainty and transfer · Chapters 11–12
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Follow the signal chain · Chapters 1–3
Begin with bedload, not all river sediment
Follow the path from contact to waveform
Particle size and impact speed both matter
Understand missed and merged events · Chapters 4–5
A detection threshold changes what is visible
Overlap, clipping and missing records are different problems
Convert a surrogate to mass · Chapters 6–8
Calibration connects a defined signal to a defined reference
A calibration mixture can conceal grain-size bias
Sensor coupling and the surrounding bed can change
Check time and cross-section scaling · Chapters 9–10
Transport pulses make time matching essential
A local sensor does not automatically span the channel
Test uncertainty and transfer · Chapters 11–12
Assess uncertainty without hiding calibration structure
Check transfer with changed conditions, not a repeated fit
CHAPTER 1 OF 12
1. Begin with bedload, not all river sediment
Bedload is sediment transported with frequent or sustained contact with the bed, including rolling, sliding and saltating motion. Suspended sediment is carried within the flow for appreciable periods. The distinction concerns transport behaviour, not an immutable label attached to each grain. A particle can behave differently as hydraulic conditions change.
An impact sensor is particularly connected to interactions near the bed. It does not automatically measure the suspended load or the whole sediment budget. A strong bed-impact signal and a weak turbidity signal can coexist because the two observations respond to different processes. Neither should be forced to reproduce the other without a physical explanation.
Keep the target quantity explicit: transport activity, detected impacts, grain-size information, mass rate per unit width, or total mass crossing a section. These are related but distinct outputs. A monitoring system may support one more directly than another. Begin a review by naming the output actually required rather than assuming that any sediment signal supplies all five.
CHAPTER 2 OF 12
2. Follow the path from contact to waveform
In a plate-based system, a moving grain transfers momentum and energy during contact, the structure responds, and a sensor converts some part of that response into an electrical record. Contact location, mounting, plate properties and the motion of the grain can affect the waveform. The detector therefore observes a coupled grain–structure–sensor system.
One physical impact can make a decaying oscillation with several peaks. Counting each peak as a separate particle would confuse the structure’s ringing with the number of grains. Conversely, several particles arriving close together may form an overlapping waveform that is difficult to separate. Event definitions belong to the measurement method, not to the river itself.
Wyss and colleagues’ primary flume study of the Swiss plate geophone investigated how particle size and transport conditions affect registered signals. Read its results as evidence for conditional response, not for a universal conversion from electrical impulses to gravel mass. The specific device and processing definition remain part of what was calibrated.
CHAPTER 3 OF 12
3. Particle size and impact speed both matter
For geometrically similar particles of equal density, mass scales with diameter cubed. Doubling diameter multiplies mass by eight. This simple geometry already explains why counting particles cannot determine total mass unless their size distribution is known. One hundred small grains and one hundred large grains are equal counts but very different transported masses.
Kinetic energy scales as one half of mass times speed squared. At fixed mass, doubling impact speed multiplies that energy by four. Only part becomes the recorded signal, with losses and response depending on the contact and sensor system. Do not equate voltage directly with total kinetic energy without an independently supported response relationship.
These dependencies create ambiguity. A changed amplitude distribution could reflect changed particle sizes, impact velocities or coupling rather than a simple mass-flux increase. Frequency information can provide additional constraints in suitable systems, but it is not a context-free diameter ruler. Multiple signal features are helpful when their physical interpretation and calibration are demonstrated.
CHAPTER 4 OF 12
4. A detection threshold changes what is visible
Software often distinguishes a candidate event from background by applying a threshold or another detection rule. Raising the threshold can suppress noise while also discarding weaker sediment signals. Lowering it can retain small signals while admitting more contamination. There is no interpretation-independent threshold that simultaneously detects every relevant grain and rejects every disturbance.
If small particles or low-energy contacts fall below detection, a zero count means no accepted events under that rule. It does not prove that the bed was completely immobile. The missing fraction may also change with flow conditions or particle mixture. A calibration should identify the range of transport it represents, including the effective lower sensitivity limit.
Threshold definitions must remain versioned. An apparent change in transport following a software update may partly reflect a new acceptance rule. Preserve raw waveforms or sufficient diagnostic summaries where feasible so the effect can be examined. Comparing two count series is meaningful only after confirming that the word count refers to the same processing quantity.
CHAPTER 5 OF 12
5. Overlap, clipping and missing records are different problems
At high activity, individual impact responses can overlap. The system may report fewer separable events even as true transport increases. Electrical clipping is different: the recorded waveform reaches an instrument limit and loses amplitude information. A data gap is different again: there is no usable observation for part of the interval. None is safely handled by calling the count zero.
A simple hypothetical dead-time model illustrates the danger of saturation. If an ideal detector accepts one event and is then unavailable for τ seconds, a specified non-paralysable model gives observed rate n = m/(1 + mτ), where m is the true event rate. Real bedload waveforms need not follow that model; it is a sensitivity exercise only.
Inverting the teaching model gives m = n/(1 − nτ), provided nτ is below one. As that product approaches one, the inferred rate becomes highly sensitive to small errors. This is not an invitation to correct any field count with the formula. It shows why extrapolating a low-rate calibration into unresolved high-activity signals can become unstable.
CHAPTER 6 OF 12
6. Calibration connects a defined signal to a defined reference
Suppose a calibration interval contains I accepted impulses and an independently estimated transported mass M over matching support. A simple relation I = kM defines k in impulses per kilogram. The coefficient depends on the signal definition and conditions. It is not a physical constant of sediment, and the same grain mixture need not produce the same k at a different site.
If 2,400 impulses correspond to 120 kg in a fictional matched calibration, k is 20 impulses/kg. A later compatible record of 3,000 impulses would imply 150 kg under that relation. The division is easy. Establishing compatibility of particle mixture, transport conditions, sensor state and sampling support is the substantive part of the estimate.
The companion field application by Wyss and colleagues compared flume-based relationships with direct bedload samples. Its importance here is the need to test transfer against independent material observations. A fitted calibration should not be presented as validated merely because it reproduces the same observations used to fit its coefficient.
CHAPTER 7 OF 12
7. A calibration mixture can conceal grain-size bias
Imagine two grain classes with different impulse yields per kilogram. The aggregate coefficient depends on their mass fractions. If those fractions change during a flood, a single fixed coefficient can drift even when the electronics remain stable. This is physical calibration drift: the relationship changes because the transported material and interactions change.
In a teaching mixture, class A produces 40 impulses/kg and class B produces 10 impulses/kg. An equal-mass mixture gives 25 impulses/kg. If the mixture later becomes 20% A and 80% B by mass, the coefficient becomes 16 impulses/kg. Applying the old 25 would underestimate mass for a given accepted impulse total.
The example does not assign those values to real grain sizes or a real sensor. It exposes the mixture dependence. A useful monitoring interpretation retains size information from independent samples or validated signal features, and tests whether the chosen calibration spans the mixture being inferred. More total counts cannot by themselves identify an unobserved mixture change.
CHAPTER 8 OF 12
8. Sensor coupling and the surrounding bed can change
The same impact can produce a different recorded waveform if structural response or transmission changes. Mounting condition, surrounding sediment and the state of the measurement surface can matter. Changes following a flood should therefore prompt both a transport interpretation and a sensor-condition review. A smooth time series is not evidence that the physical response stayed fixed.
Burial can prevent grains from contacting a plate as before; scour can alter the approach and contact conditions. Nearby machinery, traffic or other vibrations may enter the record without representing transported gravel. These possibilities call for diagnostic evidence and site-specific checks, not an assumption that every non-sediment influence has one easily subtracted frequency.
A pre-event and post-event calibration comparison can help distinguish a stable response from a changed one, within the evidence available. Record inspections and configuration changes with the data. If a response change cannot be resolved, flag the affected period and bound its interpretation rather than silently applying the earlier coefficient throughout.
CHAPTER 9 OF 12
9. Transport pulses make time matching essential
Bedload can move in bursts. A short direct sample may catch a pulse or miss it, while a continuous sensor integrates a different interval. Comparing unmatched windows can make a reasonable surrogate appear poor or a poor surrogate appear convincing by accident. Time stamps, durations and averaging rules are part of the calibration evidence.
For example, 60 kg moved in the first minute and 6 kg in each of the next nine minutes gives a ten-minute total of 114 kg. Extrapolating the first minute alone would yield 600 kg. Extrapolating one of the quiet minutes would yield 60 kg. Neither is the event total merely because it was measured precisely during its own interval.
Integrate rates with their actual time support. A kilogram-per-second estimate multiplied by seconds gives kilograms; an impulse total should not be multiplied by time again. Gaps during peak transport deserve particular attention because they may remove a disproportionate share of the load. A complete-looking daily number can conceal an unresolved flood interval.
CHAPTER 10 OF 12
10. A local sensor does not automatically span the channel
An impact plate samples a limited part of the bed. Transport may concentrate in particular corridors and shift laterally as the river changes. Multiplying the local estimate by total channel width assumes an appropriate transverse distribution. The site-specific evidence for that distribution is separate from the calibration of the sensor itself.
If a representative strip carries 2 kg/s per metre across 5 m, the corresponding section rate is 10 kg/s. But if only a 1 m corridor carries 2 kg/s per metre and the remaining 4 m carries 0.2, the total is 2.8 kg/s. The same energetic local reading can imply very different section totals.
A waterborne hydrophone or a remote seismic sensor has a different spatial weighting from a contact plate. Petrut and colleagues’ passive-acoustic study models grain-size information from self-generated noise; it is not evidence that one microphone records all grains crossing a known width. Ask how the spatial response enters each method before comparing their reported mass rates.
CHAPTER 11 OF 12
11. Assess uncertainty without hiding calibration structure
Uncertainty can enter through event detection, reference sampling, the conversion relationship, temporal gaps and cross-section scaling. Some components are correlated: a particle-size change can alter both signal response and sampler efficiency. Treating all errors as independent small percentages can understate the uncertainty of the final load.
For a simple illustrative calculation M = I/k, if independent relative standard uncertainties are 5% in I and 20% in k, first-order propagation gives about sqrt(0.05² + 0.20²), or 20.6%, in M. This excludes model mismatch and support error. A narrow counting uncertainty should not be presented as the complete precision of the transported mass.
Detection limits and saturation may create asymmetric uncertainty. Fine material can be systematically missed; large overlapping events may not be recoverable through symmetric error bars. Explain these limits directly. A result can still be useful as a coarse-transport indicator while remaining unsuitable as a complete sediment budget. The label should match the defensible output.
CHAPTER 12 OF 12
12. Check transfer with changed conditions, not a repeated fit
Hold the sensor and impulse total fixed, then change the grain mixture. Hold the mixture fixed, then change the detection rule. Hold processing fixed, then change the active transport corridor. These are independent transfer checks because each challenges a different link in the chain. They are more informative than performing the same division several times.
Keep calibration observations separate from independent checks wherever the data permit. Examine residuals by flow condition, particle mixture and time since installation or disturbance, rather than relying only on one overall agreement statistic. A relation that performs well on average can still fail during the events that dominate annual coarse-sediment transport.
A defensible conclusion names the measured signal, the conditions supporting conversion and the unresolved limits. For example, a sensor may document increased detectable coarse-particle activity without establishing an exact total mass. That is a useful finding. The goal is to make the evidence transferable, not to force every instrument record into a stronger claim than it can support.
Independent worked case and evidence checks
Use a fictional impact-plate record with a fixed event definition and stable sensor condition. During a matched calibration interval, an independent reference assigns 120 kg to the same spatial and temporal support, while the processor reports 2,400 impulses. The simple coefficient is 20 impulses/kg. These numbers define the teaching relationship rather than a real device specification.
A later compatible interval produces 3,000 impulses. Using M = I/k gives 150 kg. Before accepting it, list what compatible means: the same event definition, relevant particle-size mixture, comparable impact conditions, an unchanged sensor response and the same interpretation of sampled width. The arithmetic cannot verify those assumptions.
Now keep the impulse total fixed but use a physically justified new coefficient of 15 impulses/kg after an observed response change. The inferred mass becomes 200 kg. Applying the old coefficient would give a value 25% lower than 200 kg. The count series can remain continuous while the mass conversion changes.
To isolate particle mixture, assign class A 40 impulses/kg and class B 10 impulses/kg. With 60 kg of each class, the total is 2,400 + 600 = 3,000 impulses from 120 kg. The aggregate coefficient is 25 impulses/kg. Confirm that this differs from the first calibration: they are separate fictional scenarios, not interchangeable observations.
Change the 120 kg mixture to 24 kg of A and 96 kg of B. The predicted impulse total is 960 + 960 = 1,920, giving 16 impulses/kg. If the old 25 coefficient were used, the estimated mass would be 76.8 kg. That is 36% below the true 120 kg specified in this example.
For a pure geometry check, consider equal-density spherical particles of diameters d and 2d. Their mass ratio is eight. Ten large particles therefore equal the mass of eighty small particles under these assumptions. Equal particle counts cannot stand for equal mass. Real shapes require their own size–mass information rather than automatic spherical treatment.
At fixed mass, increase impact speed from v to 2v. Kinetic energy increases by four. Do not infer that sensor voltage must also increase fourfold: energy transfer and structural response intervene. The exercise is useful precisely because it separates a valid mechanics calculation from an unsupported electrical conversion.
Next define two processor settings. The lower threshold accepts 100 true sediment events and 20 noise events; the higher accepts 70 true sediment events and 2 noise events. Precision among accepted events improves from 100/120 to 70/72, but recovery of the specified true events falls. The setting that looks cleaner is not automatically the least biased for mass.
For an ideal non-paralysable dead-time illustration, set τ = 0.002 s and observed n = 100 events/s. The assumed model gives m = 100/(1 − 0.2) = 125 events/s. This recovers a model event rate, not kilograms. Do not apply it to a real plate without evidence that its overlapping-signal behaviour follows the same model.
Increase the observed rate to 450 events/s under that hypothetical model. The inferred m becomes 4,500 events/s because the denominator is only 0.1. If observed n were 460, the result would be 5,750 events/s. A small observed-rate change now produces a large inferred change. Near saturation, model and measurement errors matter intensely.
For time integration, let a calibrated mass rate be 0.4 kg/s for 600 seconds and 1.2 kg/s for the next 300 seconds. The total is 240 + 360 = 600 kg. Averaging the two rates without accounting for their unequal durations would give 0.8 kg/s and a misleading 720 kg over the full 900 seconds.
Suppose the 300-second high-rate interval was actually missing. The supported observed total would be 240 kg, with an unresolved gap. Filling that gap using the preceding 0.4 kg/s would produce 360 kg overall, but that is an imputation based on an assumption. Keep it distinguishable from measured and independently supported load.
For cross-section scaling, compare a uniform 5 m bed carrying 2 kg/s per metre with a channel where only 1 m carries that rate and 4 m carries 0.2. The totals are 10 and 2.8 kg/s. If the sensor sits in the energetic corridor, its local record alone cannot distinguish these two transverse patterns.
Return to M = I/k with an inferred 150 kg. With independent relative uncertainties of 5% and 20%, the illustrative combined standard uncertainty is about 30.9 kg. That calculation excludes the transverse ambiguity just described. It would be inappropriate to cite ±30.9 kg as a complete section-load uncertainty when corridor width remains unresolved.
Finally, imagine counts decline after a flood and the plate is found partly buried. Two explanations remain possible: less transport, changed sensor exposure, or both. An unchanged electrical self-check does not establish unchanged grain contact. Additional physical and independent transport evidence is needed before the decline can be attributed entirely to river behaviour.
A calibration reference also has uncertainty. Suppose the same 2,400 impulses are paired with a physical-sample mass estimate that is later revised from 120 to 100 kg after a documented sampling correction. The coefficient changes from 20 to 24 impulses/kg. Under the revised relation, 3,000 impulses imply 125 kg rather than 150. The sensor has not changed; the reference interpretation has. Preserve both versions and the reason for revision.
For a validation check, imagine the coefficient was fitted using several moderate-flow intervals and then tested on a separately sampled high-flow interval. A large residual at high flow could reflect mixture change, overlap, sampling limitations or a different transport corridor. It is evidence of a transfer problem, not a unique diagnosis. Investigate the mechanism before adding a new coefficient solely to make that one point fit.
Change the question from total mass to the onset of detectable coarse transport. A sensor with unresolved mass conversion may still provide useful timing, provided noise, thresholds and clock alignment are understood. Conversely, an approximate event total need not locate the onset precisely if the output was heavily averaged. Judge usefulness against the requested quantity instead of labelling the entire record either perfect or worthless.
An independent answer should recover 150 kg under the first calibration, the mixture coefficients 25 and 16, the 600 kg time-integrated total and the 10 versus 2.8 kg/s section comparison. Then state which quantities are observations, which are model outputs and which require new evidence. That separation is the practical skill the sensor record should teach.
Worked comparison
| Record or change | Defensible first reading | Stronger claim needing evidence |
|---|---|---|
| Accepted impulses increase | More detected signal under the rule | Proportional mass increase |
| Small signals disappear | Below-threshold or absent events | No sediment motion |
| Waveforms overlap | Individual detection is limited | Exact corrected grain count |
| A local plate is active | Transport near that support | Uniform transport across width |
| Counts fall after burial | Signal or exposure changed | River transport declined by the same fraction |
Questions and practical checks
Is one impulse one grain?
Not generally. Ringing can create several threshold crossings from one contact, while overlapping impacts can merge. The processing definition must be stated.
Can the same calibration be used after every flood?
Only with evidence that its relevant conditions remain supported. Changes in bed state, particle mixture, sensor exposure or processing can change the relationship.
Does this measure suspended sediment too?
Not automatically. The impact-based bedload signal and suspended-sediment observations have different sensitivities and should be interpreted separately.
Can frequency identify grain diameter exactly?
Suitable calibrated methods can use frequency information, but source, propagation and sensor response remain part of the inference. It is not a universal one-frequency–one-diameter rule.
What should a useful archive contain?
Time support, waveform or diagnostic data, event rules and versions, sensor condition, calibration samples, support definitions, flags and gap-handling assumptions.
Continue with the connected subjects
Connect the sensor to independently sampled sediment
Bedload sampling, intake efficiency and cross-section scaling
Maximum scour, infill and post-flood survey limits
Distinguish other sediment surrogates
Acoustic sediment indices and particle-size-dependent calibration
Suspended loads, turbidity surrogates and event hysteresis
Sources and scope
Wyss and colleagues (2016), part 2: application to field sites with direct bedload samples
Petrut and colleagues (2018): passive acoustic grain-size inference from self-generated noise
USGS Scientific Investigations Report 2010–5091: bedload-surrogate monitoring technologies
The numerical cases in this guide are original fictional teaching examples, not measurements from a real site. This is an educational explanation, not a field safety instruction, engineering approval or an official certification. For a real project, use qualified practitioners, current product documentation and site-specific evidence. Sources were checked on 30 October 2026.