When the same river level carries different flows, measure another state
A river gauge records the same water level on two afternoons. On the first, water moves steadily downstream. On the second, a high downstream water level slows the current almost to rest. A stage-only rating cannot distinguish those states if it assigns one discharge to each level.
The index-velocity method separates the problem into two relationships: stage tells us the wetted cross-sectional area, and a continuously measured velocity proxy helps estimate the mean velocity through that area. Discharge follows from their product. The distinction sounds simple; maintaining both relationships through changing channels and reversing flow requires careful evidence.
This article develops that full measurement chain with original calibration, timing, reversal and uncertainty examples. It continues the existing rating-curve hysteresis article without repeating its overview. The examples are educational, not operating rules for a live gauge, flood warning or navigation decision.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Understand the two-rating idea · Chapters 1–2
Identify why stage alone has stopped identifying discharge
Separate observed index velocity from inferred mean velocity
Build geometry and calibration · Chapters 3–5
Build a stage-area relationship at a defined standard section
Calibrate using paired observations that represent the same time
Read residuals for missing physics rather than chasing a high fit score
Test changing and reversing conditions · Chapters 6–8
Treat reversal as a calibration domain, not a negative sign shortcut
Synchronise stage and velocity before multiplying them
Use geometry and hydraulic events to detect rating changes
Protect the continuous record · Chapters 9–10
Propagate uncertainty without assuming the two ratings are independent
Preserve gaps, qualifications and net-exchange meaning
Practise and document · Chapters 11–12
Practise a backwater comparison with a geometry change
Document a continuous record as a maintained inference
CHAPTER 1 OF 12
1. Identify why stage alone has stopped identifying discharge
Stage is water-surface elevation relative to a reference. Discharge is signed volume crossing a section per unit time. Their relationship can be approximately single-valued at a suitable stable control, but downstream backwater, tides, storage and unsteady flow can make the same stage accompany different velocities.
Imagine a channel with wetted area 30 m² at stage 1.8 m. If mean downstream velocity is 0.80 m/s, discharge is 24 m³/s. If downstream backwater reduces mean velocity to 0.20 m/s at the same stage, discharge is 6 m³/s. A stage-only calculation that returns 24 in both cases misses a real hydraulic difference.
The answer is not necessarily to make the stage-only curve more complicated. If the input does not uniquely identify the output, a more flexible curve can fit noise without observing the missing state. Add a physically informative measurement or use an appropriate alternative method.
The index-velocity method is particularly useful when a representative velocity measurement adds information absent from stage. It does not imply every site requires it, or that any conveniently located velocity sensor will be representative. Site selection, calibration and ongoing validation determine whether the additional observation genuinely improves the discharge record.
CHAPTER 2 OF 12
2. Separate observed index velocity from inferred mean velocity
A fixed acoustic velocity instrument observes a selected part or path of the channel. Its index velocity vᵢ is a proxy, not automatically the area-weighted mean velocity V of the entire standard cross-section. A calibrated relationship translates between them.
A simple teaching rating is V = a + bvᵢ. If a = 0.02 m/s, b = 0.90 and vᵢ = 0.60 m/s, inferred mean velocity is 0.56 m/s. With area 30 m², discharge is 16.8 m³/s. Multiplying 30 by the raw 0.60 would instead give 18 m³/s and silently assume a different relationship.
Some sites require additional predictors such as stage or cross-stream velocity. The USGS index-velocity manual explains that the suitable relation depends on the site. More predictors are not automatically better: they require enough representative calibration data and independent validation to support their effects.
Keep units and coordinate definitions explicit. A beam-axis velocity, a transformed streamwise component and a speed magnitude are not interchangeable. The chosen index may be signed. A change in sensor orientation or processing can change its meaning even when the variable retains the same name in a spreadsheet.
CHAPTER 3 OF 12
3. Build a stage-area relationship at a defined standard section
The stage-area rating A(h) describes wetted cross-sectional area as a function of stage h at a surveyed standard section. Survey geometry and vertical reference are essential. This relation concerns area, not discharge, so it should not be inferred solely from a convenient stage–flow curve.
For a fictional rectangular channel 12 m wide with bed elevation corresponding to stage 0.50 m, A(h) = 12(h − 0.50) m² while the rectangular assumption applies. At h = 2.00 m, water depth is 1.50 m and area is 18 m². At h = 2.20 m, area is 20.4 m².
A natural cross-section generally needs surveyed points and an appropriate area computation rather than a rectangle. A bench may become inundated above a threshold, causing area to increase more rapidly. Floodplain activation can also alter the velocity distribution, affecting the other rating at the same time.
Maintain stage datum and cross-section stability. Deposition, scour, vegetation and structural changes can alter the relationship. Re-survey and review at a frequency suited to susceptibility and current agency requirements. A stable index instrument does not guarantee stable geometry, and a corrected stage record cannot repair an outdated area relation by itself.
CHAPTER 4 OF 12
4. Calibrate using paired observations that represent the same time
Calibration requires independent discharge measurements and simultaneous stage and index-velocity data. For each reference discharge Qₘ, estimate mean velocity at the standard section as Vₘ = Qₘ/A(h), using geometry and timing appropriate to that reference measurement.
Suppose a reference measurement gives Qₘ = 21 m³/s while A = 35 m². Mean velocity is 0.60 m/s. The simultaneous index reading is 0.70 m/s. This pair informs the relation between index and mean velocity; it does not say that the index instrument is wrong by 0.10 m/s. It samples a different part of the velocity field.
Repeat across the conditions the record must represent: low and high flows, backwater, seasonal changes and both directions if reversal is relevant. Several measurements clustered near one velocity cannot validate a relation at much larger velocities or on the opposite side of zero.
Match averaging windows. An independent discharge crossing lasting several minutes should be paired with a representative index and stage interval, not a single convenient instantaneous reading. During rapidly changing conditions, the choice of window becomes part of the uncertainty. Preserve timestamps and time zones so later reviewers can reconstruct the pairing.
CHAPTER 5 OF 12
5. Read residuals for missing physics rather than chasing a high fit score
Fit a relationship only after inspecting the data and physical setting. In a simple regression, residual is observed mean velocity minus predicted mean velocity. Plot residuals against index velocity, stage, flow direction, time and relevant downstream conditions. A pattern can reveal a missing state or changing relation.
For the fictional rating V = 0.02 + 0.90vᵢ, an index reading 0.50 m/s predicts V = 0.47 m/s. An independent measurement gives 0.53 m/s, so residual is +0.06 m/s. At area 40 m² that discrepancy corresponds to 2.4 m³/s. Its importance depends on measurement uncertainty and the decision, not only the regression statistic.
A high coefficient of determination can coexist with a meaningful low-flow bias if the dataset spans a wide range. Large flows dominate variation while small errors near zero determine the sign of net exchange. Inspect absolute and relative consequences, especially where the intended use depends on low or reversing flow.
Do not add stage, squared velocity or separate branches merely to improve the fit on the same data. Ask whether the extra term has physical support and improves independent predictions. Keep calibration and validation conceptually distinct; a model cannot establish its own performance by reproducing the points used to tune it.
CHAPTER 6 OF 12
6. Treat reversal as a calibration domain, not a negative sign shortcut
A signed index measurement permits negative values, but a rating calibrated only for downstream flow is not automatically valid upstream. Flow structure, recirculation and the instrument sampling footprint may behave differently after reversal. Reference measurements need to test that domain.
Use the same fictional relation V = 0.02 + 0.90vᵢ. At vᵢ = −0.10 m/s, predicted V = −0.07 m/s. If A = 30 m², Q = −2.1 m³/s. At vᵢ = 0, it predicts a small positive V of 0.02 m/s, not zero. An intercept can therefore affect the timing of inferred reversal.
The predicted zero occurs at vᵢ = −0.02/0.90, about −0.022 m/s. That may be physically plausible for a nonuniform section, or it may reveal insufficient near-zero data. Do not force an intercept to zero solely for visual neatness, and do not accept a fitted intercept solely because a regression produced it.
Near reversal, inspect absolute uncertainty. If predicted mean velocity is −0.01 m/s with uncertainty comparable to ±0.03 m/s, the direction may be unresolved. Reporting a huge percentage uncertainty is less informative than stating that the estimate spans both signs. Maintain the distinction between a measured index reversal and a confidently inferred whole-section discharge reversal.
CHAPTER 7 OF 12
7. Synchronise stage and velocity before multiplying them
The continuous computation is Q(t) = A[h(t)]V(t). The two factors must describe a compatible time and location. Logger clock drift, averaging windows or transport between the instrument and standard section can create errors even when both individual time series appear reasonable.
Suppose at time one area is 20 m² and mean velocity is 0.20 m/s, giving 4 m³/s. At time two area is 30 m² and mean velocity is 0.60 m/s, giving 18 m³/s. Pairing the first area with the second velocity produces 12 m³/s, a state that was never observed in this example.
Even perfectly synchronised averaging needs care. The mean of a product is not generally the product of means. Across the two equally weighted observations, mean discharge is (4 + 18)/2 = 11 m³/s. Mean area times mean velocity is 25 × 0.40 = 10 m³/s. The difference arises because area and velocity covary.
Compute at an appropriate validated time resolution before aggregation. If only coarse averages are available, acknowledge the lost covariance information. Do not claim that a smooth daily series preserves short reversals or peaks that were removed by long input averaging windows.
CHAPTER 8 OF 12
8. Use geometry and hydraulic events to detect rating changes
Two separate ratings create two distinct maintenance questions. Did the stage-area relation change, and did the index-to-mean-velocity relation change? A single discharge discrepancy can reflect either, both, or uncertainty in the independent reference measurement.
Imagine the old area is 40 m² but a new survey supports 36 m² at the same stage after deposition. If mean velocity remains 0.50 m/s, old geometry gives 20 m³/s while updated geometry gives 18 m³/s. Adjusting the velocity-rating intercept to absorb that difference would conceal a geometric change.
Now hold area at 40 m² but suppose altered vegetation diverts the fast core away from the index sampling volume. The old index relationship may no longer represent section mean velocity. A new area survey alone cannot solve that problem; independent discharge and velocity evidence are needed.
Preserve effective dates and the supporting observations for each rating version. A major flood or instrument remounting is a reason to review, not proof that the entire previous record must be replaced. Diagnose the mechanism and interval of change. Avoid applying a later rating backwards across conditions for which its validity is unknown.
CHAPTER 9 OF 12
9. Propagate uncertainty without assuming the two ratings are independent
For a small-error approximation with Q = AV, uncertainty in area and mean velocity both matter. If they were independent, standard deviations σA and σV would contribute approximately V²σA² + A²σV² to discharge variance. With dependence, a covariance term also enters. The approximation is a teaching tool, not a complete operational uncertainty procedure.
Take A = 40 m² with standard uncertainty 1 m² and V = 0.50 m/s with standard uncertainty 0.03 m/s. Under the stated independence assumption, σQ ≈ √[(0.50 × 1)² + (40 × 0.03)²] = 1.3 m³/s. Nominal discharge is 20 m³/s, so this corresponds to 6.5% standard relative uncertainty, not a 95% interval.
Independence is especially questionable when calibration mean velocity was calculated by dividing reference discharge by the same area estimate. Shared geometry errors can induce dependence. Sensor drift, extrapolation and domain changes may also create systematic components not captured by a simple regression residual.
Near zero discharge, relative uncertainty becomes unstable. Use absolute units and sign confidence instead. For a tidal exchange budget, a small bias persistent over many hours may matter more than large momentary scatter. Repetition reduces some random error; it does not automatically cancel a shared intercept or geometry bias.
CHAPTER 10 OF 12
10. Preserve gaps, qualifications and net-exchange meaning
If stage is valid but index velocity is missing, the full method is missing an input. Reverting to a stage-only rating is not automatically justified at a backwater site—the reason for installing the velocity measurement may be precisely that stage alone is ambiguous.
A gap estimate may use a documented alternative with suitable validation, but preserve its status and method. Keep observed, modelled, interpolated and revised intervals distinguishable. A complete-looking series with hidden substitution can be less useful than an honestly qualified record.
For signed flow, distinguish gross directional exchange from net volume. Suppose +20 m³/s persists for three hours and −15 m³/s for three hours. Downstream volume is 216,000 m³, upstream volume 162,000 m³ and net downstream volume 54,000 m³. The net is much smaller than the total water exchanged in both directions.
A constant +1 m³/s bias over those six hours adds 21,600 m³ to the net, or 40% of the true net in this example. That is why near-zero bias and reversal timing deserve careful validation even when the larger instantaneous flows appear well represented. State whether the use concerns net exchange, gross movement, peak flow or ecological exposure.
CHAPTER 11 OF 12
11. Practise a backwater comparison with a geometry change
A fictional stage-area relation gives A = 24 m² at stage 1.5 m. A validated teaching velocity relation is V = 0.01 + 0.80vᵢ. On day one vᵢ = 0.70 m/s, so V = 0.57 m/s and Q = 13.68 m³/s. On day two the same stage accompanies vᵢ = 0.20, giving V = 0.17 and Q = 4.08 m³/s.
The stage is identical but discharge differs by 9.60 m³/s. Explain the missing information a stage-only method would ignore. The index method distinguishes the states because it observes a velocity proxy, assuming its calibrated relationship remains valid under both conditions.
Now a survey shows that deposition reduced the day-two area to 22 m². With the same inferred velocity 0.17 m/s, Q becomes 3.74 m³/s. The geometry correction changes discharge by 0.34 m³/s. Do not confuse this with the much larger backwater-related velocity difference between days.
For transfer, reverse the index to −0.10 m/s. The teaching relation predicts V = −0.07 m/s and Q = −1.54 m³/s using 22 m². Before treating that as reliable, ask whether upstream flow was included in calibration and whether near-zero uncertainty spans the sign. Correct arithmetic is necessary but cannot extend a rating beyond its evidence.
CHAPTER 12 OF 12
12. Document a continuous record as a maintained inference
A reproducible record includes the standard section survey, stage datum, index sampling configuration, independent discharge measurements, paired averaging windows, fitted rating forms, residual checks, validation results and effective dates. The final discharge series should identify provisional and revised intervals.
Keep the raw index velocity and stage alongside computed mean velocity, area and discharge. When a later review changes one relationship, users can identify which part of the inference changed. A single final column of discharge conceals too much of the measurement chain.
Review the calibration domain in several dimensions: positive and negative index velocity, stage, overbank conditions, season, downstream controls and equipment configuration. A point may lie within the numerical velocity range but outside the hydraulic conditions that supported the original relation.
The Gold Standard is not a more elaborate equation for its own sake. It is a transparent answer to a physical ambiguity. Measure the missing state, calibrate the proxy, maintain the geometry, align time, and test the result under changed conditions. Then report what the record supports without turning extrapolation or gap filling into an unlabelled observation.
Independent worked case and evidence checks
A fictional low-gradient channel uses a stage-area relation A(h) = 16(h − 0.40) m² within a stated in-bank range. At h = 2.40 m, area is 32 m². Its independently validated teaching velocity relation is V = 0.015 + 0.85vᵢ. With index velocity +0.50 m/s, mean velocity is +0.440 m/s and discharge is +14.08 m³/s. The raw index is not substituted for the mean.
Later, backwater holds stage at 2.40 m but the index falls to +0.10 m/s. Mean velocity becomes +0.100 m/s and discharge +3.20 m³/s. Stage-only inference would fail to distinguish these states. A third interval has index −0.20 m/s, giving mean velocity −0.155 m/s and discharge −4.96 m³/s, provided the upstream-flow calibration is valid.
Let each interval last two hours. Their volumes are +101,376, +23,040 and −35,712 m³. Net downstream volume is +88,704 m³. Gross downstream movement is +124,416 m³ and gross upstream movement is 35,712 m³. Summing absolute movement gives 160,128 m³; that is not the net export. State the quantity before reporting the total.
An independent validation measurement during a later negative-index interval gives Qₘ = −5.60 m³/s at area 32 m², implying mean velocity −0.175 m/s. At the same index −0.20, the rating predicts −0.155 m/s. Residual is −0.020 m/s, corresponding to −0.64 m³/s. One residual does not prove a rating shift. Review reference uncertainty, timing and whether several upstream measurements show a coherent pattern.
Now a cross-section survey supports area 30 m², not 32 m², at that stage. The unchanged mean-velocity predictions give +13.20, +3.00 and −4.65 m³/s. Over the same intervals net volume becomes +83,160 m³. Geometry alone changes the net by 5,544 m³. Re-fitting the velocity relation without recognising the new area would blur two different sources of change.
Finally, suppose the negative interval loses index data. The valid stage of 2.40 m cannot reveal whether discharge was +14.08, +3.20 or −4.96 m³/s in the original setting. That is the decisive transfer check. A stage-only gap fill would reintroduce the original ambiguity. The defensible next step is a qualified alternative estimate supported by independent evidence, or a visible unresolved gap—not an invented continuous observation.
A separate validation exercise demonstrates why a random split can be misleading. Imagine twenty reference measurements from one quiet week and four from a later high-backwater week. If points from both weeks are mixed randomly between fitting and testing, the test may largely reproduce hydraulic states already represented in training. Holding out the later week instead asks the harder question: does the relationship transfer to a genuinely different episode? Neither split is universally correct; choose the test that matches the intended future use and preserve the hydraulic conditions in the validation record.
Suppose a relation predicts mean velocities of 0.10, 0.30, 0.60 and 0.90 m/s for four independent checks. Observed means are 0.14, 0.32, 0.59 and 0.88 m/s. Residuals are +0.04, +0.02, −0.01 and −0.02 m/s. Their average is only +0.0075 m/s, but the low-velocity positive residual is important. At 25 m² it represents 1.0 m³/s. A small average residual can therefore hide a condition-dependent error, especially if the use concerns small net flows.
Next change the geometry without changing those residual velocities. At 100 m², the same +0.04 m/s discrepancy represents 4.0 m³/s. This does not mean the velocity rating itself became less precise; its discharge consequence grew with area. Reviewing residuals in both velocity and discharge units helps connect the calibration to the water-management question. Near reversal, also report whether the observed and predicted signs agree, since a modest absolute error can reverse the inferred exchange direction.
Finally, test a constant intercept bias over a longer budget. An unrecognised +0.01 m/s mean-velocity bias at a constant 25 m² contributes +0.25 m³/s. Over one day it adds 21,600 m³ to the computed volume. Over thirty days at those unchanged teaching conditions it adds 648,000 m³. This is not a forecast for a real station; it shows why a tiny persistent bias can matter more to an accumulated net budget than visually dramatic short-lived scatter. Match validation priorities to the duration and quantity being reported.
The learner’s final decision is to specify the validation target before calculating a summary score. For flood peaks, examine high-stage geometry and velocity coverage. For tidal net exchange, examine near-zero bias, reversal timing and signed integration. For low-flow ecology, examine the low end and qualified gaps. One attractive overall fit cannot certify all three uses without the relevant evidence.
Worked comparison
| Same stage, same area | Index velocity | Inferred mean velocity | Discharge |
|---|---|---|---|
| h = 2.40 m; A = 32 m² | +0.50 m/s | +0.440 m/s | +14.08 m³/s |
| h = 2.40 m; A = 32 m² | +0.10 m/s | +0.100 m/s | +3.20 m³/s |
| h = 2.40 m; A = 32 m² | −0.20 m/s | −0.155 m/s | −4.96 m³/s |
Questions and practical checks
Is index velocity the whole-section mean?
No. It samples a selected volume or path and is related to whole-section mean velocity using independent discharge measurements.
Why are there two ratings?
One relates stage to wetted area. The other relates the observed velocity proxy to section mean velocity. Their product gives discharge. Either relation can change.
Can a downstream-flow fit simply be used with negative input?
The arithmetic accepts it, but physical validity requires calibration or validation under reversing conditions. Flow structure can change with direction.
Why compute discharge before taking daily means?
Area and velocity can covary. The mean of their product is not generally the product of their means, and long averages can conceal reversals and peaks.
What is the most important maintenance check?
There is no universal single check. Maintain geometry, datum, sensor configuration, timing and independent validation, then focus effort on the uncertainty that matters to the intended use.
Continue with the connected subjects
Understand why one stage can imply several flows
Rating-curve hysteresis and unsteady flow: the one-input limitation
Bank storage and flood-wave hysteresis: flow and storage need not change together
Streamflow mathematics: areas, weighted velocity and rating curves
Maintain the measurement chain
Gage datum shifts: reference changes can alter a continuous record
Hydroacoustics: acoustic observations and inference limits
Ice-affected discharge: do not transfer open-water relations without checking conditions
Validate discharge and review changing geometry
Moving-boat ADCP discharge: motion references, unmeasured regions and independent validation
Sources and scope
USGS streamgaging: stage, discharge and index-velocity observations
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.