Naturalised flow is a reconstructed scenario, not a hidden second gauge record
A downstream gauge records the river that actually passed it. Upstream reservoirs, diversions and return flows may have changed that water’s amount and timing. Streamflow naturalisation estimates what a specified record might have been without selected influences. The word “specified” matters: removing reservoir operations is not the same as reconstructing an untouched landscape.
The basic arithmetic is a water account. Its difficulty lies in deciding what belongs in each term, matching the time and spatial boundaries, and recognising effects that cannot be reversed by adding a monthly volume. This guide builds the account from a small fictional example, then tests delayed returns, groundwater pumping, storage drawdown and uncertain operating records.
Use the method to understand a reconstruction, not to make a water-allocation or engineering decision from a classroom calculation. An actual project needs current records, a defensible model, appropriate professional review and the applicable rules. The worked examples below are original teaching cases, not observations or an official naturalised series.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1: Define the counterfactual · Chapters 1–2
Route 2: Build the water account · Chapters 3–5
Route 3: Preserve timing and pathways · Chapters 6–8
Route 4: Test uncertainty · Chapters 9–10
Route 5: Interpret and transfer · Chapters 11–12
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Define the counterfactual · Chapters 1–2
Say which influences you are removing
Draw the control volume before choosing signs
Build the water account · Chapters 3–5
Convert discharge rates into comparable volumes
Read storage change with its direction intact
Distinguish withdrawals, returns and consumption
Preserve timing and pathways · Chapters 6–8
Handle imports, exports and repeated water
Keep delayed return flow on the correct dates
Do not add groundwater pumping directly to river flow
Test uncertainty · Chapters 9–10
Correct reservoir losses only against the chosen reference
Carry missing operations and correlated errors
Interpret and transfer · Chapters 11–12
Validate without using the answer twice
Report a scenario with a revision trail
CHAPTER 1 OF 12
1. Say which influences you are removing
Begin with a sentence such as: “Estimate daily discharge at the downstream station without the two upstream reservoirs and recorded surface-water diversions, holding the observed weather and the stated land-cover scenario fixed.” This is a counterfactual definition: an explicit alternative to what occurred. It identifies the effects being removed and those being retained. Without that sentence, two competent analysts can calculate different quantities while both label them naturalised flow.
A reconstruction may retain contemporary urban runoff, drainage, forest cover or river geometry. Removing dams alone does not undo those influences. Nor does it reconstruct a climate that existed before development. If the intended application requires a historical landscape, that is an additional modelling problem with its own evidence. Avoid the reassuring but inaccurate shorthand that a corrected discharge record is necessarily the river’s original natural state.
The definition should also identify the output location and time step. A monthly naturalised volume can support some seasonal accounting questions, but it cannot recover the peak hour of a flood. A daily sequence at a reservoir inflow is not automatically the daily sequence at a distant downstream gauge. Travel time and intermediate exchanges connect the locations. State which transformation is part of the reconstruction.
For a transfer check, change the question from seasonal water availability to ecological low-flow duration. The seasonal account may remain useful, but the new question needs an appropriately timed sequence and a clear treatment of releases and returns. A correct annual volume does not ensure correct durations below a threshold. The method transfers through conservation, while the required temporal resolution changes with the decision.
CHAPTER 2 OF 12
2. Draw the control volume before choosing signs
Place the gauge at the outlet of the reconstructed system and draw every relevant reservoir, withdrawal, return, import and export upstream. Identify which transfers cross the outer boundary and which merely move water within it. This prevents the same water being counted once as a transfer between reservoirs and again as an abstraction from the basin. A physical map often exposes accounting errors faster than a more elaborate formula.
Use integrated water volumes over a common interval for the first calculation. In a deliberately simplified surface-water system, naturalised outlet volume equals observed outlet volume plus upstream withdrawals minus upstream returns, plus the increase in reservoir storage. These signs follow from the counterfactual: undoing a withdrawal adds water back; undoing a return removes its contribution; undoing storage accumulation restores water withheld during that interval.
Other corrections depend on the defined boundary. An imported volume that increased observed outlet flow is subtracted when its influence is removed. An exported volume absent from observed outlet flow is added back. If a reported withdrawal already includes an export, do not add the export a second time. Define mutually exclusive terms rather than copying every available database column into one sum.
Signs should survive a plain-language test. Suppose a reservoir supplies a dry-month release entirely by drawing down previously stored water. Observed downstream flow includes water that did not arrive naturally during that month. Naturalisation should reduce that month’s volume, not increase it. A negative storage change therefore enters as a negative correction. This check catches the common mistake of adding the absolute value of every storage change.
CHAPTER 3 OF 12
3. Convert discharge rates into comparable volumes
Discharge is a rate, commonly cubic metres per second. Storage and transfer records may be cubic metres or millions of cubic metres. For a constant teaching rate, multiply discharge by elapsed seconds to obtain volume. A mean 2 m³/s over a 30-day interval gives 2 × 30 × 86,400 = 5,184,000 m³, or 5.184 million m³. Use the actual interval length for real records.
A month with 31 days has 2,678,400 seconds, not the 2,592,000 seconds in the 30-day example. Mixing those denominators creates a false seasonal pattern. Leap days, partial records and time-zone conventions can matter when operational records are reconciled. Store interval start and end times with the data instead of relying on an ambiguous label such as “March total”.
If observed flow varies, integrate the time series or use an appropriate interval mean with its coverage stated. A mean calculated from only a few available days is not automatically the full-month mean. A missing high-flow day can dominate the volume error. Keep measured, estimated and missing intervals distinguishable so that the reconstructed record does not acquire false completeness through averaging.
To check units, divide a corrected volume back by the interval’s seconds. Suppose the corrected 30-day volume is 7.776 million m³. The mean discharge is 3 m³/s. This reversal is simple enough to perform independently and reveals thousand-fold or million-fold mistakes quickly. Numerical precision in the final discharge should reflect the quality of the transfers and storage records, not just the calculator’s display.
CHAPTER 4 OF 12
4. Read storage change with its direction intact
Storage change is final storage minus initial storage. If a reservoir rises from 12 to 15 million m³, the interval change is +3 million m³. If it falls from 15 to 12 million m³, the change is −3 million m³. Under the simple naturalisation account, the first correction adds withheld water and the second removes water supplied from earlier storage. Keep the sign attached to the value.
Water level is not storage volume. Converting level to volume requires an appropriate elevation–storage relation and a consistent vertical reference. Sedimentation, changed survey data or revised datum can alter that relation. A jump caused by a revised curve is not necessarily a physical addition of water. Preserve the curve version used at both ends of the interval, especially where a small difference between large stocks drives the correction.
Reservoir storage may include more than the visible surface pool if bank storage or other exchanges matter to the question. Whether and how those effects enter requires a model consistent with the counterfactual. Do not add a generic “seepage loss” if the water returns upstream of the outlet and is already represented elsewhere. Follow the pathway and interval before deciding whether a loss is external, internal or delayed.
A useful changed-context test is a full reservoir at both month ends. Its net storage correction is zero, but operations may have strongly reshaped flows within the month. Zero monthly change does not mean zero regulation. If flood timing or daily low flows matter, analyse the within-month sequence. Endpoints can close a monthly volume account while concealing the very behaviour the new question asks about.
CHAPTER 5 OF 12
5. Distinguish withdrawals, returns and consumption
Withdrawal is water taken from a source. Return flow is water that reaches surface water or groundwater after use. Consumptive use is not interchangeable with either term. A withdrawal can be large while net depletion is smaller because part returns. However, the relevant return must be placed at the right location and time. Water returning below the chosen gauge cannot restore the observed volume at that gauge.
In a simplified same-interval example, 4 million m³ is withdrawn upstream and 1.5 million m³ returns upstream before reaching the outlet. The net correction is +2.5 million m³. Adding the full withdrawal while ignoring the measured return would overestimate naturalised outlet volume by 1.5 million m³. Subtracting the return twice, perhaps once explicitly and once inside a reported net-depletion term, creates the opposite error.
Reported wastewater release is not always identical to river return at the accounting point. It may be routed elsewhere, reused, stored or infiltrated before entering the chosen system. A conveyance loss may also follow a groundwater pathway and return later. Build a transfer ledger linking source, destination and timing. The ledger should distinguish an observed transfer from a coefficient used to estimate a missing transfer.
For transfer, move the fictional return point from above the gauge to below it. Keep withdrawal and treatment volumes unchanged. The upstream gauge reconstruction now adds back the full relevant withdrawal, with no subtraction for that downstream return. A basin-wide consumptive-use statistic may not change, but the local reconstruction does. This is why a geographically correct ledger is more useful than a single catchment-wide percentage.
CHAPTER 6 OF 12
6. Handle imports, exports and repeated water
Imported water can become a return flow after use. If both an import record and an effluent record describe the same water at different stages, their roles need careful separation. The import entering the system is an external input; the effluent is often an internal transfer within the defined basin. Whether it contributes to outlet flow depends on use, storage, consumption and where it returns.
A simple teaching system imports 3 million m³, consumes 1 million m³ and returns 2 million m³ to the river above the gauge. If the chosen correction removes the import’s effect on outlet flow, the relevant contribution is 2 million m³ under those assumptions. Subtracting the full 3 and also subtracting the return would remove too much. This example requires its pathway assumptions; it is not a universal import correction.
Choose one coherent accounting approach. You can construct a full basin water balance, or use documented net alterations at the gauge with a consistent routing model. Mixing gross transfers from the first approach with already netted depletion estimates from the second invites double counting. Before combining datasets, ask whether each number is gross, net, source-based, destination-based or already adjusted for returns.
Repeated use makes this especially important. The same water may be withdrawn, returned and withdrawn again. Gross withdrawals can exceed unique water input without violating conservation. Summing withdrawals alone cannot reveal net river depletion. Trace the successive transfers or use a validated net accounting method at the required scale. A large withdrawal total is a record of activity, not automatically an equal volume missing from the outlet.
CHAPTER 7 OF 12
7. Keep delayed return flow on the correct dates
A withdrawal today need not produce a return today. Treatment, storage, irrigation drainage and subsurface movement can distribute returns over later intervals. Naturalising each day by a fixed same-day return fraction can distort both the low-flow sequence and event response. Document the lag assumptions and the evidence supporting them rather than treating timing as a cosmetic detail after the volume calculation.
Suppose 10 units are withdrawn on day one. Six units return on day three, while four are consumed under the teaching assumptions. The daily corrections are +10 on day one and −6 on day three. Applying a net +4 on day one preserves the three-day total but changes the daily hydrograph. Which representation is acceptable depends on whether the intended question is total volume or day-specific flow.
Routing from the alteration point to the outlet introduces another delay and possibly attenuation or exchanges. A recorded reservoir release at midnight should not automatically be matched to the downstream gauge at midnight. The appropriate routing may vary with flow conditions. A fixed lag can be a useful approximation only if its adequacy is checked for the range and time step used.
For the independent check, shift the return by another two days while keeping its volume fixed. The five-day total correction remains the same if the whole return lies within the window, but individual daily corrections move. If the window ends before the return arrives, its total also changes. Reporting window boundaries and lag treatment prevents a later user from comparing incomplete and complete return sequences as equivalent.
CHAPTER 8 OF 12
8. Do not add groundwater pumping directly to river flow
Groundwater pumping can be supplied by changes in aquifer storage and changes in flows to or from streams and other boundaries. The streamflow effect depends on the hydrogeological connection and time since pumping began. It is not generally equal to the pumping rate at the same instant. Treating every cubic metre pumped from any upstream well as a same-day missing river cubic metre can misplace both magnitude and timing.
A conceptual teaching example pumps 100 units in an interval. A model attributes 60 to aquifer storage decline, 30 to reduced discharge to the river and 10 to other changed boundary flows during that interval. Only the modelled river effect enters a correction for that river at that time. Adding all 100 would confuse the groundwater withdrawal account with the surface-water response. The example illustrates bookkeeping, not a real partition.
Effects can continue after pumping changes or stops. A reconstruction that ends at the shutdown date may omit later stream depletion. Conversely, groundwater returns can arrive after the original water-use interval. Use an appropriate response model and retain uncertainty about aquifer properties, boundaries and operating history. An elegant daily series built on unverified connectivity is not more reliable because its time step is small.
Change the setting from a well near a connected stream to one separated by low-permeability material or a different flow system. The accounting principle remains conservation, but the response function may change substantially. Reusing the original correction by distance alone is not defensible. Obtain geological and hydraulic evidence, and report where the groundwater-to-stream pathway cannot be adequately constrained.
CHAPTER 9 OF 12
9. Correct reservoir losses only against the chosen reference
Reservoir evaporation can reduce water available downstream, but naturalisation needs the change relative to the defined no-reservoir scenario. The reservoir surface also receives precipitation, and the counterfactual landscape would have its own evapotranspiration and runoff. Adding gross open-water evaporation without considering the chosen reference may not represent the intended net alteration. State the convention and remain consistent across the record.
For a deliberately simplified teaching correction, suppose the approved model estimates that the reservoir creates an additional net loss of 0.8 million m³ during a month relative to the defined alternative. Add +0.8 million m³ to undo that loss. This is not a claim that every reservoir’s correction is positive or that gross pan evaporation supplies the number directly. It is the modelled incremental effect under a declared scenario.
Changing area matters because evaporation depth multiplied by one fixed area can misstate volume when the reservoir expands or contracts. Use appropriate area information and interval integration where required. If storage change and evaporative loss share the same uncertain elevation–area–volume relation, their errors may be related. Do not treat all terms as statistically independent merely because they occupy different columns.
The transfer test is to hold evaporation depth constant while halving the wetted area through much of the interval. The volume loss changes even though the depth does not. Then change the counterfactual from a dry surface to a wetland with appreciable water use. The incremental correction changes again. The physical water balance and the definition of the alternative both shape the answer.
CHAPTER 10 OF 12
10. Carry missing operations and correlated errors
Operational records can contain meter changes, unrecorded diversions, estimated returns and inconsistent storage curves. Start with a provenance table giving source, interval, units, coverage, revision date and uncertainty for each term. Preserve flags when creating the corrected series. A naturalised value built partly from estimates should not look indistinguishable from one supported by complete observations.
Suppose observed volume is 20 ± 1 million m³, and a net alteration correction is 5 ± 2 million m³ under stated bounds. The central reconstruction is 25 million m³. A simple outer-bound sensitivity is 22–28 million m³. This is not automatically a 95% interval. A probability statement requires a justified error model, including any dependence between observed flow and alteration estimates.
Common sources of error can move several terms together. The same meter may feed withdrawal and return estimates, or one reservoir curve may affect storage and surface area. Depending on the signs and relationship, errors can reinforce or partly cancel. Sensitivity analysis should expose the shared assumption. Adding many uncertain terms does not guarantee a narrow result, even if the final numerical series is smooth.
A negative reconstructed volume deserves investigation rather than automatic replacement by zero. It may indicate timing mismatch, double counting, an inappropriate model or uncertainty larger than a small true flow. Clipping negatives changes totals and drought metrics. Retain the original calculation, document the diagnosis and make any adjustment explicit. Physical plausibility checks should lead to revised evidence or assumptions, not invisible editing.
CHAPTER 11 OF 12
11. Validate without using the answer twice
Look for checks that do not reuse the same alteration estimates. Suitable comparisons might include upstream records, independently measured transfers, periods with little regulation or a separately assessed model. Each has its own spatial and temporal limits. Agreement is more persuasive when independent evidence tests a mechanism than when two calculations share the same uncertain operating ledger.
A perfect water-account closure can be meaningless if one component was defined as the residual. Similarly, calibrating a model to a naturalised series and then declaring that same series independently confirmed by the model is circular. State what was used for construction, calibration and evaluation. Where independent periods or sites are available, keep their role clear rather than blending them into one impressive-looking fit statistic.
Volume bias alone is insufficient when the application depends on timing. A model can match annual volume and still misplace dry-season lows or flood peaks. Check the characteristics relevant to the stated purpose, such as seasonal volume, duration below a threshold or event timing. Do not claim performance for every use from one aggregate metric.
For transfer, imagine two reconstructed series with identical annual totals. One places a return pulse in a wet month and the other in a dry month. They may support similar annual accounting but different low-flow conclusions. Ask which observations constrain the placement. If none do, retain timing scenarios instead of selecting a single sequence with unsupported confidence.
CHAPTER 12 OF 12
12. Report a scenario with a revision trail
Present observed flow and reconstructed flow together, along with the corrections needed to understand their difference. Name the influences removed, retained landscape and climate, output station, time step and treatment of groundwater effects. This lets a reader decide whether the series fits a new purpose. A file labelled simply “natural flow” loses the distinctions that made the reconstruction defensible.
Include the equation, sign conventions, routing assumptions, data versions and unresolved intervals. If a return-flow estimate is revised later, update the relevant periods without overwriting the previous release silently. A changed series may reflect improved knowledge rather than a changed river. Version history prevents those two explanations being confused in a trend analysis.
A naturalised reconstruction is not itself an environmental-flow recommendation, a legal entitlement or a safe abstraction limit. Those decisions need additional ecological, social, engineering and regulatory evidence. The reconstruction supplies one description of water under an alternative scenario. It cannot decide how competing values should be weighed merely because it has removed selected operational influences.
Your final check is to explain one change that conserves total volume but alters timing, one that changes the boundary contribution, and one that changes the counterfactual rather than the observations. If you can identify which terms or models must change in each case, the learning has transferred. If you only know how to add a fixed percentage to gauge flow, the central accounting problem remains unresolved.
Independent worked case and evidence checks
Use a fictional 30-day basin account, with every volume expressed in million cubic metres. Observed outlet volume is 24. Surface-water withdrawals upstream total 6, and measured upstream returns within the same interval total 2. Reservoir storage rises from 18 to 21, giving ΔS = +3. For this first calculation, assume no other relevant alterations and no routing lag across the monthly boundary.
The reconstructed volume is 24 + 6 − 2 + 3 = 31 million m³. The net alteration correction is +7 million m³. The observed mean is 24,000,000/2,592,000 ≈ 9.26 m³/s; the reconstructed mean is 31,000,000/2,592,000 ≈ 11.96 m³/s. The calculation corrects a defined monthly volume. It does not determine the daily shape of the reconstructed hydrograph.
Now include a separately modelled incremental reservoir loss of 0.6 million m³ relative to the declared no-reservoir scenario. The result becomes 31.6 million m³, approximately 12.19 m³/s. The correction must not also be hidden inside an already netted storage or depletion series. Check the source definition before adding it. A new column is not necessarily a new physical contribution.
During the next 30-day interval, observed outlet volume is 22, withdrawal is 5, return is 2, and reservoir storage falls from 21 to 17. With no additional loss correction in this teaching step, naturalised volume is 22 + 5 − 2 − 4 = 21 million m³. It is lower than observed flow. That is plausible because drawdown releases water accumulated previously. Naturalisation need not always raise a record.
Combine the first simple month and the second: observed volume is 46, net surface-water withdrawal is 7, and total storage change is −1. Reconstructed volume is 46 + 7 − 1 = 52 million m³. This equals the sum of 31 and 21. This telescoping storage check is useful: the intermediate 21 million m³ stock cancels, leaving only the beginning and ending stocks of the combined period.
Next change a return-flow location. Suppose 0.8 of the first month’s reported 2 million m³ return actually enters below the outlet gauge. Only 1.2 belongs in the gauge account. The first simple reconstruction becomes 24 + 6 − 1.2 + 3 = 31.8 million m³. Nothing changed physically when the mistake was discovered; the accounting location was corrected. Preserve the reason for the revision.
Consider a timing-only alteration on a three-day subperiod. A 90,000 m³ withdrawal occurs on day one and 54,000 m³ returns on day three. Correct daily outlet volumes by +90,000, zero and −54,000 respectively, under the simplified no-routing assumptions. The total correction is +36,000 m³. Applying +36,000 on day one alone would match the total but would not preserve the day-specific counterfactual.
Translate the daily corrections into mean rates: +90,000/86,400 ≈ +1.042 m³/s on day one and −54,000/86,400 = −0.625 m³/s on day three. These are interval-average adjustments, not instantaneous pulses. If a downstream travel-time model places the changes on different dates, use that routed sequence instead. Do not claim that arithmetic alone has supplied the missing travel-time information.
Add a groundwater example separately rather than hiding it inside surface withdrawals. A model estimates that pumping reduces river contribution by 0.4 million m³ in month one and 0.9 in month two, although pumped volume is larger. Undoing the modelled river effect adds those respective amounts. Do not substitute total pumping without justifying the stream response. The different timing is part of the reconstruction, not an accounting inconvenience.
An uncertainty check now asks whether first-month withdrawal might be 5.5–6.5 and upstream return 1.6–2.4, while observed volume and storage change are held fixed for this illustration. The simple reconstruction ranges from 24 + 5.5 − 2.4 + 3 = 30.1 to 24 + 6.5 − 1.6 + 3 = 31.9 million m³. This is a stated sensitivity envelope, not a confidence interval.
The next check changes the reservoir endpoints instead. If the storage-curve revision adds the same constant 2 million m³ to both endpoint stocks, their difference remains +3, so this correction is unchanged. If the revision changes the curve shape, endpoint differences can change. Distinguish a constant offset from a level-dependent revision before assuming that all datum changes cancel.
Finally identify the decision boundary. The reconstructed monthly means can illuminate the volume effect of the selected operations. They cannot establish an hour-specific flood peak, the ecological adequacy of a release or legal water availability. A defensible answer gives the 31 and 21 million m³ baseline results, explains the drawdown sign, and identifies which added evidence is required before changing the question.
Worked comparison
| Change to observed system | Simple correction direction | Essential qualification |
|---|---|---|
| Reservoir fills | Add positive ΔS | Same interval and storage definition |
| Reservoir draws down | Add negative ΔS | Do not use absolute change |
| Withdrawal above outlet | Add relevant effect | Avoid double counting net depletion |
| Return above outlet | Subtract relevant effect | Place it on the correct date |
| Return below outlet | No upstream-gauge return subtraction | Different boundary from basin-wide account |
Questions and practical checks
Is naturalised flow measured directly?
No. The observed gauge record is measured or estimated from observations. Naturalised flow reconstructs a stated alternative using additional data and assumptions. Keep the two records distinguishable.
Must naturalised flow exceed observed flow?
No. Reservoir drawdown or imported returns can increase observed flow during an interval. Removing those effects can reduce the reconstructed value.
Can monthly totals recover flood peaks?
No. Many daily or hourly sequences have the same monthly total. Peak reconstruction needs appropriately resolved observations, operations and routing.
Is withdrawal minus return always consumptive use?
Only with compatible boundaries, timing and definitions. Returns can occur later or elsewhere, and conveyance and storage pathways can complicate the comparison.
What is the most useful first check?
Draw the boundary and trace one parcel through withdrawal, use and return. Then make sure that no stage is counted twice and every correction has the intended sign.
Continue with the connected subjects
Build and test the reservoir account
Evaporation depth, changing area and reservoir water balance
Bank storage and delayed return flow
Correlated errors and the limits of budget closure
Sources and scope
USGS: withdrawal, return-flow and consumptive-use data for hydrological models
USGS: groundwater development, storage and changing system flows
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.