The boat is moving. The river is moving. Which motion did you measure?
An acoustic Doppler current profiler can collect thousands of velocity observations while a boat crosses a river. Yet a smooth discharge number can still be wrong if the bed moves, the boat-motion reference fails, or the unmeasured edges are poorly represented. More observations do not remove a shared reference error.
This guide explains how water velocity relative to an instrument becomes velocity relative to the Earth, then discharge through a cross-section. The original examples isolate moving-bed bias, edge estimates, reciprocal transects and changing flow so that each source of uncertainty has a visible consequence.
This is a guide to interpreting evidence, not instructions for operating a boat in a river. Field work requires trained personnel, approved procedures and a site-specific safety assessment. The existing Hydroacoustics and ocean ADCP articles cover acoustic foundations; this article addresses the distinct problem of measuring river discharge from a moving platform.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Understand the measured quantity · Chapters 1–2
Begin with discharge through a section, not speed at a point
Understand the Doppler observation and its coordinate transformations
Choose a defensible motion reference · Chapters 3–5
Recover water motion using an independent boat-motion reference
See why a downstream-moving bed can bias discharge low
Use moving-bed tests to discriminate, not merely reassure
Account for the whole section · Chapters 6–8
Recognise the top, bottom and edge regions that are not measured directly
Treat extrapolation as a model with testable consequences
Estimate edges using geometry as well as nearby velocity
Test repeatability and bias · Chapters 9–10
Read reciprocal transects as a diagnostic sequence
Separate measurement uncertainty from scatter and processing choice
Practise and communicate · Chapters 11–12
Practise three changes without mixing their corrections
Make the final discharge traceable to its conditions
CHAPTER 1 OF 12
1. Begin with discharge through a section, not speed at a point
Discharge Q is signed water volume crossing a defined section per unit time. A simplified expression is Q = ΣAᵢuᵢ, where uᵢ is the velocity component normal to each area element. A fast current parallel to the section does not contribute the same through-flow as one crossing it directly.
For three fictional parts of a section, let areas be 8, 20 and 12 m² and normal mean velocities 0.25, 0.90 and 0.40 m/s. Contributions are 2.0, 18.0 and 4.8 m³/s, giving Q = 24.8 m³/s. Total area is 40 m², so the cross-section mean velocity is 0.62 m/s. The unweighted mean of the three velocities would not produce the same result.
A moving-boat ADCP samples water cells along a crossing. Processing combines velocity, depth and the boat trajectory to estimate the volume passing through the sampled curtain. The water does not need to be uniformly fast, and the track does not need to be a perfectly straight line. It does need an appropriate geometric and motion-reference treatment.
Keep the reporting sign clear. Downstream is often chosen positive, but tidal or locally reversing conditions require signed values. Taking absolute velocities can manufacture discharge in an eddy. Before examining instrument settings, draw the section, the positive direction and the water paths that count as crossing it.
CHAPTER 2 OF 12
2. Understand the Doppler observation and its coordinate transformations
An acoustic Doppler instrument sends sound and interprets frequency changes in returning echoes from scattering material moving with the water. The immediate velocity information is relative to the instrument and initially resolved along acoustic beams. Conversion to useful water velocity involves instrument geometry and coordinate transformations.
Tilt, heading and sound speed influence that transformation. A heading error can rotate velocity components into the wrong direction; an incorrect sound-speed assumption can affect acoustic calculations. Good-looking numbers are not a substitute for checking configuration, sensor diagnostics and the conditions experienced during measurement.
There is also a sampling volume. The instrument does not observe one infinitely small point, and cell size trades spatial resolution against other measurement characteristics. Adjacent observations may not be independent. A dense image of velocity should not be read as a photograph of every parcel crossing the river.
A useful first-principles check is a stationary-water thought experiment. If an instrument moves east through still water, it should observe water moving west relative to itself. To recover Earth-referenced zero water velocity, the processing must add the boat motion correctly. The arithmetic of that correction is central to the next chapters, even though the complete field calculation is vector-valued and more detailed.
CHAPTER 3 OF 12
3. Recover water motion using an independent boat-motion reference
In a simplified one-dimensional illustration, water velocity relative to Earth equals water velocity relative to the boat plus boat velocity relative to Earth. Let the boat move downstream at 0.20 m/s while the water moves downstream at 0.80 m/s. The instrument observes +0.60 m/s relative water motion; adding +0.20 restores +0.80 m/s.
Real crossings involve two horizontal components, changing heading and an integration over the trajectory. Do not apply the one-dimensional example as an operational correction formula. Its purpose is to reveal the dependency: a biased boat velocity can bias every Earth-referenced water cell even when the relative observations are consistent.
Bottom tracking estimates boat movement relative to acoustic targets on the bed. With a stationary bed and adequate returns, that can provide the required motion reference. But stationary is a physical assumption. Sediment moving downstream near the bed can make the acoustic reference itself move.
Satellite navigation can provide a different motion reference, but not an automatically perfect one. Position or velocity quality, reception, timing, heading and coordinate consistency all need review. Changing reference is a change in the measurement chain, not a button that removes every other uncertainty. Preserve which reference was used and why it was accepted for each measurement.
CHAPTER 4 OF 12
4. See why a downstream-moving bed can bias discharge low
Imagine the boat is held approximately fixed relative to Earth while a mobile acoustic bed target moves downstream at 0.06 m/s. Relative to that target, the boat appears to move upstream. If processing assumes the target is fixed, it can assign the boat an upstream velocity and subtract too much from the observed downstream water motion.
Use a deliberately simplified uniform section with area 50 m² and true mean water velocity 0.70 m/s. True discharge is 35 m³/s. If a moving-bed reference produces an effective velocity underestimate of 0.06 m/s, the apparent mean becomes 0.64 m/s and discharge becomes 32 m³/s. The bias is −3 m³/s, about −8.6% of the true value.
The numerical shortcut is illustrative only. Moving-bed effects in a real transect depend on spatial variation, geometry, navigation reference and processing. A single local moving-bed speed should not be casually added to the reported total without an approved correction method and supporting evidence.
The important misconception is that a moving-bed problem necessarily looks noisy. It can create a stable directional bias. Several repeat crossings can agree closely while sharing that bias. A smooth bed track and a small standard deviation therefore do not establish that the reference was stationary.
CHAPTER 5 OF 12
5. Use moving-bed tests to discriminate, not merely reassure
USGS moving-boat guidance treats moving-bed assessment and boat-motion reference quality as important parts of measurement review. Stationary and loop-style tests provide different evidence under their applicable procedures. Their interpretation depends on navigation, instrument configuration and flow conditions, not simply whether the plotted path looks tidy.
A test made before a storm peak may not represent conditions during the peak. Bed mobility can change with sediment transport and flow. Record the time, location and conditions of the test alongside those of the discharge transects, and consider whether they represent the same hydraulic state.
If satellite-referenced and bottom-track-referenced discharges disagree, investigate both chains. A downstream-moving bed is one explanation; poor satellite data or heading-related problems are alternatives. Independent checks should discriminate among them. Selecting whichever number seems more plausible without documenting the evidence is not quality assurance.
Suppose bottom-track processing gives 32 m³/s and a validated satellite reference gives 35 m³/s, while a suitable moving-bed test supports downstream bed movement. That combination strengthens the low-bias interpretation. If the satellite record has repeated dropouts and no reliable heading, the disagreement remains unresolved. Different numbers are a reason for diagnosis, not permission to choose the larger one.
CHAPTER 6 OF 12
6. Recognise the top, bottom and edge regions that are not measured directly
An ADCP does not directly measure every part of the cross-section. Instrument draft and blanking leave a near-surface region; boundary interference limits usable cells near the bed; shallow banks leave edge regions. These contributions are estimated using supported methods and appropriate site information.
In a fictional result, measured cells contribute 72 m³/s, the surface estimate contributes 8 m³/s, the bottom estimate 10 m³/s, and both edges together 5 m³/s. Total discharge is 95 m³/s. Directly measured cells account for about 75.8% of that total, not 100%. This is not automatically unacceptable, but it tells the reviewer where assumptions matter.
The fraction outside direct sampling depends on depth, geometry, equipment and settings. A deep central channel and a broad shallow channel with the same total discharge can have very different unmeasured proportions. Copying confidence from one setting to the other ignores a major change in the observation footprint.
Ask what supports each estimate: near-surface velocity shape, near-bed profile, edge distance and edge shape. An uncertainty statement should reflect these dependencies. Quoting the instrument precision alone misses the uncertainty from reconstructing the parts of the section it did not observe.
CHAPTER 7 OF 12
7. Treat extrapolation as a model with testable consequences
Extrapolation extends information from measured water into an unmeasured region. Different supported velocity-profile assumptions can produce different discharge totals. Inspect the measured profile and its suitability for the chosen method; do not select a familiar option without examining the evidence.
Return to the 95 m³/s example. Hold measured cells and edges fixed. One plausible profile combination gives surface plus bottom contributions of 18 m³/s; another gives 22 m³/s. The total changes from 95 to 99 m³/s. The 4 m³/s spread is a sensitivity to model choice, not a formal confidence interval for the whole measurement.
A consistent profile fit in the centre does not ensure that every subsection behaves identically. Secondary circulation, bed forms, vegetation, rough banks or flow separation can alter local structure. Review whether the extrapolation represents the discharge-weighted section rather than a visually selected patch.
The changed-context test is to imagine an identical measured core with twice as much shallow unmeasured margin. Confidence should generally fall unless extra observations constrain that margin. The instrument has not become worse; the inference has become more dependent on assumptions. Reporting the measured and estimated fractions makes this dependency understandable to a reader.
CHAPTER 8 OF 12
8. Estimate edges using geometry as well as nearby velocity
An edge estimate needs the distance to the bank, water depth, velocity evidence and an appropriate edge-shape assumption. The last valid cell is not the bank, and water near a rough or recirculating margin may not behave like the central channel. A photograph can support geometry but cannot supply the missing velocity by itself.
Consider a simplified triangular edge extending 4 m horizontally from the last measured vertical to the bank, with depth 1.2 m at that vertical. Area is approximately 4 × 1.2 ÷ 2 = 2.4 m². If its representative normal velocity is 0.25 m/s, the teaching estimate is 0.60 m³/s.
If the distance is actually 7 m, the same triangular assumption gives 4.2 m² and 1.05 m³/s. The difference is 0.45 m³/s for that edge alone. In a 100 m³/s river this may be relatively small; in a 5 m³/s stream it is substantial. The acceptable consequence depends on the total and intended use, not the edge distance in isolation.
Do not assume every edge is triangular. A vertical wall, inundated bench or vegetated floodplain changes the geometry and flow distribution. Preserve notes that justify the edge model. If an edge is inaccessible or unsafe to observe, label the limitation instead of inventing a precise bank location.
CHAPTER 9 OF 12
9. Read reciprocal transects as a diagnostic sequence
Repeated crossings in opposite directions help reveal consistency, direction-dependent problems and temporal changes. They should be evaluated as an ordered set, not merely pooled until a reassuring average appears. The appropriate number and acceptance criteria follow current agency procedures and site conditions.
Suppose downstream-positive discharges from alternating crossings are 98, 102, 99 and 101 m³/s. Their mean is 100 m³/s and sample standard deviation is about 1.83 m³/s. The spread indicates repeatability under those conditions. It does not include an unrecognised common datum, moving-bed or edge bias.
Now consider 98, 104, 110 and 116 m³/s with time. The sequence may reflect a rising river rather than poor repeatability. Averaging gives 107 m³/s, but a single mean hides a real trend. Review stage and timing before treating the departures as random error.
A directional pattern is different again: left-to-right crossings give 96 and 97, while right-to-left give 104 and 105 m³/s without an overall time trend. Investigate heading, navigation, track geometry, edge treatment and configuration. Opposite-direction agreement is useful evidence, but disagreement should be explained rather than cancelled by arithmetic.
CHAPTER 10 OF 12
10. Separate measurement uncertainty from scatter and processing choice
Measurement uncertainty combines more than variation among transects. Relevant components can include navigation reference, depth, velocity transformation, unmeasured regions, invalid cells, environmental nonstationarity and judgement in processing. Some errors are shared across all crossings and do not shrink as the number of transects grows.
Keep raw data and processing history so a reviewer can compare alternatives. If removing a short invalid interval changes total discharge substantially, document why the replacement is defensible. A software output is an implementation of assumptions and quality checks, not independent proof of them.
Compare methods when possible. An independent discharge measurement, a stable rated control or a continuity check can reveal discrepancies, but each comparison has its own limitations. Two nearby sections need not carry identical instantaneous flow during changing storage, tributary input or tidal conditions.
A useful report distinguishes repeatability, estimated total uncertainty and sensitivity experiments. For example: reciprocal transects average 100 m³/s; profile alternatives shift the total by up to 4 m³/s; motion-reference quality has been independently reviewed. Do not combine these statements into an invented ±4% confidence interval without a justified uncertainty model.
CHAPTER 11 OF 12
11. Practise three changes without mixing their corrections
A fictional measured-cell contribution is 40 m³/s, with top and bottom estimates totalling 7 m³/s and edges totalling 3 m³/s. The initial total is 50 m³/s. A profile sensitivity test increases the top-and-bottom contribution to 9 m³/s, giving 52 m³/s. Only the extrapolated region changed.
Now change the reference. A separately validated motion-reference treatment increases the measured-cell contribution to 43 m³/s, while the currently assumed extrapolation and edges remain 9 and 3 m³/s in this simplified comparison. The total becomes 55 m³/s. In a real reprocessing, several terms may change together; retain the full output rather than adding isolated corrections blindly.
Finally, new bank evidence revises the edge total from 3 to 2 m³/s. Total becomes 54 m³/s. Explain why one change concerns water-motion reference, another velocity-profile extension, and another missing geometry. They are not three interchangeable calibration factors.
For transfer, halve the river depth while leaving the instrument draft unchanged. Predict a larger unmeasured fraction near the surface and bed. Then change from a quiet stable bed to active sediment transport. Predict a stronger need to test the bottom-track assumption. The appropriate checks follow the changed mechanism, not a generic instruction to collect more data.
CHAPTER 12 OF 12
12. Make the final discharge traceable to its conditions
Record the site, section, start and end times, flow direction, stage behaviour, equipment and configuration, navigation reference, moving-bed evidence, edge observations, invalid-data treatment and extrapolation choices. Include transect-level results and diagnostics that explain the accepted measurement.
Preserve the distinction between an individual field measurement and a continuous streamflow record. One ADCP measurement can help validate a rating or index-velocity relation; it does not by itself establish a year of discharge. The companion index-velocity article explains that separate calibration problem.
State which conditions limit transfer. A measurement under stable in-bank flow may not validate overbank conditions, reversing flow, a different sensor mounting or a mobile-bed flood. Reference selection and acceptable extrapolation must be reconsidered when the physical observation system changes.
The practical lesson is not distrust of acoustic instruments. It is respect for the full measurement chain. A useful number is supported by a defensible reference, representative coverage, checked estimates for missing regions and a clear account of what could still bias the result. That is stronger than an unexplained total with many decimal places.
Independent worked case and evidence checks
A fictional team measures an accessible, steady river section using four reciprocal crossings. Bottom-track-referenced totals are 46.8, 47.2, 46.6 and 47.4 m³/s. The mean is 47.0 m³/s. The close agreement initially looks reassuring, but the team keeps the moving-bed question separate from repeatability. Its independently assessed navigation data support reprocessing with a satellite-based reference.
The alternative totals are 50.1, 50.5, 49.9 and 50.3 m³/s, with mean 50.2 m³/s. The difference is 3.2 m³/s, or about 6.4% of the satellite-referenced mean. Suitable moving-bed evidence supports the interpretation that the bottom reference was moving downstream. The team does not average the two means: they represent competing reference assumptions, not two equally valid random samples.
Review of the accepted processing shows 39.5 m³/s from measured cells, 7.2 m³/s from top and bottom extrapolation and 3.5 m³/s from edges. Their sum is 50.2 m³/s. Directly sampled cells account for 78.7% of the total. A supported alternative profile increases the top-and-bottom term to 8.1 m³/s, producing 51.1 m³/s. This 0.9 m³/s difference is documented as profile sensitivity.
An edge-distance review then reveals that a wide vegetated margin was represented too simply. The revised edge estimate is 4.2 m³/s rather than 3.5 m³/s. With the original accepted profile, the total would be 50.9 m³/s; with the alternative profile it would be 51.8 m³/s. These are explicit scenarios. Choosing the largest figure because it seems conservative would not establish a best estimate.
Now change the hydrology. Suppose the four accepted transects were instead collected during a rapid rise and gave 46, 50, 54 and 58 m³/s in time order. The mean is 52 m³/s, but it represents an interval summary, not a stationary discharge. The report should retain timing, stage and the trend; additional uncertainty from changing conditions cannot be erased by adding more crossings after the rise.
An independent learner should produce three separate conclusions: the original closely agreeing measurements may share a reference bias; the accepted total depends partly on unmeasured regions; and nonstationarity is a different problem from either. The next action follows the unresolved issue. Check navigation for reference bias, improve edge evidence for margin uncertainty, and use time-aware interpretation for rapidly changing flow.
Another desk exercise tests aggregation. Three subsection areas are 5, 15 and 30 m², with normal mean velocities 0.10, 0.50 and 0.90 m/s. Contributions are 0.5, 7.5 and 27.0 m³/s, giving 35.0 m³/s. Total area is 50 m² and the area-weighted mean is 0.70 m/s. An unweighted average of the three velocities is 0.50 m/s, which would produce only 25 m³/s if multiplied by the full area. A large set of equally weighted velocity observations can therefore be misleading when their represented areas differ.
Now remove the middle subsection because its acoustic data failed. Simply omitting it leaves 27.5 m³/s, which is not an estimate of the original complete section. Filling its 15 m² with the mean of the two remaining velocities happens to give 7.5 m³/s in this specially constructed case, but that agreement is accidental. If its actual velocity were 0.20 rather than 0.50 m/s, the same fill would overestimate its contribution by 4.5 m³/s. Missing-data treatment needs physical and neighbouring evidence, not an arithmetic rule chosen because one example works.
Duration weighting answers a different question. Suppose one valid discharge result represents a stable ten-minute interval at 30 m³/s and another represents a stable thirty-minute interval at 50 m³/s. The time-weighted mean is (10 × 30 + 30 × 50)/40 = 45 m³/s. The unweighted mean is 40 m³/s. Neither weighting should be applied automatically to reciprocal transects; first decide whether they sample one stationary measurement or distinct time intervals. The reporting quantity determines the appropriate aggregation.
A practical review sheet can preserve four columns: observation, interpretation, alternative and discriminating check. For an apparent upstream boat velocity while the vessel is held in place, record the observation; consider bed motion; retain positioning or instrument error as an alternative; and identify independent navigation and a suitable moving-bed assessment as checks. This format prevents a plausible explanation from being promoted into a correction before the evidence is sufficient. It also makes uncertainty useful: each unresolved issue points to a specific next observation rather than a vague instruction to be more careful.
Worked comparison
| Evidence | What it supports | What it does not establish |
|---|---|---|
| Closely agreeing transects | Repeatability during the crossing set | Absence of common motion-reference bias |
| Supported moving-bed test | Assessment of bed-reference suitability | Perfect satellite navigation |
| Profile sensitivity run | Consequence of extrapolation choice | A complete confidence interval |
| Observed rising stage and discharge | Time-dependent flow during measurement | Random scatter around one stationary total |
Questions and practical checks
Why can a moving bed make discharge too low?
Bottom tracking can treat downstream bed motion as apparent upstream boat motion. The resulting Earth-referenced water velocity can be underestimated. Actual corrections require an appropriate validated method.
Does GPS solve the moving-bed problem automatically?
It supplies an alternative boat-motion reference, but navigation quality, timing, heading and transformations still need review. An unreliable alternative does not validate itself.
Why not average all available reference methods?
If one reference is physically biased, averaging retains part of that bias. Diagnose the discrepancy and justify the accepted reference.
Are unmeasured surface and edge regions always a failure?
No. They are recognised parts of the method, but estimates need evidence and sensitivity checks. Their importance depends on their contribution and uncertainty.
Can students repeat these examples by entering a river?
No field entry is needed. Use supplied fictional tables or authorised public datasets. River and boat measurements belong to trained teams following approved safety procedures.
Continue with the connected subjects
Understand the acoustic and reference foundations
Hydroacoustics: echoes, sound speed and observation limits
Ocean ADCP interpretation: depth bins, bottom tracking and heading
Hydrographic Surveying: vessel motion, positioning and depth quality
Connect individual measurements to hydrological records
Rating-curve hysteresis: why the same stage can carry different discharge
Gage datum shifts: protect stage and continuous-record interpretation
Tracer-dilution streamflow: a different measurement chain with different assumptions
From a field measurement to a continuous discharge record
Sources and scope
USGS Hydroacoustics: moving-boat discharge measurement resources and quality guidance
USGS: loop-method investigation of moving-bed effects and discharge bias
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 29 October 2026.