A rain gauge can report a changing total because the weather changed, because the observation system changed, or because both changed. A double-mass curve helps inspect consistency by comparing accumulated observations at a target gauge with accumulated observations in a reference series. A change in slope identifies a change in their relationship. It does not identify its own cause.
This guide develops an original eight-year example, derives the adjustment factor carefully and shows why correcting a record requires more evidence than a bend in a graph. It then changes the reference network and the time scale to test what the method can still establish. The aim is a transparent observation record, with measured and adjusted values kept distinct.
The Hydrology subject guide explains rainfall depth, catchment boundaries and water accounting. Here the narrower question is comparability through time: when a long rainfall series is used to examine change, are we comparing like with like? The supplied calculations are invented. They do not describe the pictured waterway or certify any actual gauge record.
eduKateSG · Hydrology
Choose the question closest to yours
Define the reference, build accumulated totals, diagnose the slope change and derive a justified reporting-basis adjustment.
Define the comparison · Chapters 1–2.
Build and diagnose the curve · Chapters 3–5.
Adjust and inspect the evidence · Chapters 6–8.
Transfer across networks and time scales · Chapters 9–10.
Practise and check · Chapters 11–12.
Full chapter index · Subject foundation · Learning that transfers into life
Chapter contents
Define the comparison · Chapters 1–2
Build and diagnose the curve · Chapters 3–5
Adjust and inspect the evidence · Chapters 6–8
Transfer across networks and time scales · Chapters 9–10
Practise and check · Chapters 11–12
CHAPTER 1 OF 12 · Define the comparison
1. Define consistency before looking for a trend
A consistent rainfall record maintains an appropriately stable relationship with the reference information used for the question. It does not require identical annual rainfall. A wet year can raise both target and regional totals while leaving their relationship much the same. A dry year can lower both without making either record inconsistent.
Separate consistency from accuracy. Two instruments can share a bias and agree with each other. Conversely, a genuine local change in rainfall may make one station depart from the wider region without an instrumental fault. Agreement is evidence about a comparison, not an automatic certificate of truth.
Define the quantity first. This article uses annual precipitation depth in millimetres for matched complete years. We are not comparing instantaneous intensity, the largest hourly burst or a storm’s spatial pattern. An annual total can be useful while concealing changes within the year that matter to a different decision.
State the reason for the analysis. A classroom exercise might ask whether a gauge relocation created a break in comparability. A research project might need a defensible long record before assessing a climatic trend. Those aims are related, but an investigation of data consistency should not silently become a conclusion about climate.
The USGS report on double-mass curves describes the method and recognises both observation changes and physical changes as possible sources of a changed relationship. That distinction should remain visible throughout the analysis, especially when a neat correction factor tempts us to move too quickly.
The reference should represent relevant regional variation while providing information independent of the target record. For the first teaching example, suppose it is the arithmetic mean of three nearby gauges with complete matched annual records, suitable geographic representation and no known breaks during the comparison period.
Those assumptions are doing real work. Three nearby gauges are not automatically representative if a mountain barrier or persistent local storm pattern makes the target behave differently. A network selected only because it produces the straightest graph can create false confidence by building the desired answer into the comparison.
Keep the target out of its own reference where the aim is an independent consistency check. If it contributes to the reference average, a target error also moves the comparator, reducing the visible departure. The strength of this masking depends on its weight, but the conceptual problem exists before any calculation.
Record which gauges were used, their units, coverage and averaging rule. If one reference gauge disappears halfway through the record, an average of two gauges may not be directly equivalent to the earlier average of three. A change in network composition can create a reference break rather than a target break.
For students, begin with a supplied reference series rather than inventing geographic weights. Ask what information would be needed to defend the reference in a real investigation. A useful answer identifies location, exposure, observation practice and completeness; it does not merely request “more data” without saying what those data would test.
CHAPTER 3 OF 12 · Build and diagnose the curve
3. Accumulate matched observations step by step
A double-mass comparison plots cumulative target precipitation against cumulative reference precipitation. Start from the same origin and add one matched period at a time. The horizontal axis here is cumulative reference depth; the vertical axis is cumulative target depth. Reversing the axes reverses the slope ratio, so the labels matter.
Let X be the running sum of reference totals and Y the running sum of target totals. If target precipitation is consistently 1.1 times the reference in the simplified example, Y increases by 1.1 units for each unit added to X. The cumulative points follow a line with slope 1.1.
The slope is a comparison of depths, so it is dimensionless when both axes use the same units. A slope greater than one means that the target accumulated more depth than the reference over that segment. It does not mean that rainfall was unusually intense or that runoff exceeded a safe level.
Use chronological order. The double-mass curve connects cumulative points from successive periods; it is not a ranking of years from wettest to driest. Sorting the years would change the interpretation of where an apparent break occurred, even though the grand totals remained the same.
Keep the annual table next to the cumulative curve. Cumulative values are strongly dependent on earlier values, and accumulation can make a relationship look impressively smooth. Inspecting the original increments and their ratios helps prevent visual smoothness from being mistaken for independent evidence at every plotted point.
CHAPTER 4 OF 12 · Build and diagnose the curve
4. Work through the eight-year record
Here is an invented complete record. Reference totals vary between years. The target initially records exactly 1.1 times the reference and later exactly 0.9 times it. The clean arithmetic isolates the method; real observations would contain variation around these relationships.
| Year | Reference, mm | Target, mm | Cumulative reference, mm | Cumulative target, mm |
|---|---|---|---|---|
| 1 | 800 | 880 | 800 | 880 |
| 2 | 1,000 | 1,100 | 1,800 | 1,980 |
| 3 | 1,200 | 1,320 | 3,000 | 3,300 |
| 4 | 1,000 | 1,100 | 4,000 | 4,400 |
| 5 | 900 | 810 | 4,900 | 5,210 |
| 6 | 1,100 | 990 | 6,000 | 6,200 |
| 7 | 1,000 | 900 | 7,000 | 7,100 |
| 8 | 1,000 | 900 | 8,000 | 8,000 |
For years 1–4, the segment slope is 4,400/4,000 = 1.1. For years 5–8, calculate the increments after the break: the target adds 3,600 mm while the reference adds 4,000 mm. The later slope is therefore 3,600/4,000 = 0.9.
Do not calculate the later slope as 8,000/8,000. That is the whole-record slope from the origin. It blends the two regimes and conceals their difference. A segment slope uses changes between the segment’s endpoints, not necessarily the final coordinates divided by one another.
The change is substantial in this deliberately exact example. Still, the numerical result establishes only a changed target–reference relationship. We have not yet learned whether the target gauge moved, the reference changed, a local physical process shifted or the records contain an error.
CHAPTER 5 OF 12 · Build and diagnose the curve
5. Diagnose the bend before adjusting values
Suppose a maintenance record now shows that the target gauge was relocated immediately before year 5. This timing makes an observational explanation worth investigating. It does not prove the move caused the entire difference. Check the new exposure, instrument configuration, observation practices and any overlapping comparison measurements.
Ask whether the reference network remained stable. If its gauges also changed at the same time, the apparent agreement between a target break and a relocation date may be incomplete evidence. An independent comparator is valuable precisely because it helps separate competing explanations.
Return to the original annual values. Look for transcription errors, unit changes and mismatched year boundaries. A millimetre–centimetre error has a different signature from a small exposure change, but both can become a bend in an accumulated plot. A plausible hydrological story should not prevent basic data checks.
Physical differences also deserve attention. A target site may respond differently from regional gauges during particular weather regimes. A change in that relationship can be real rather than a defect to remove. Correcting every departure until the target matches its neighbours would erase potentially meaningful local information.
Write a diagnosis record before proposing an adjustment: the break location, candidate causes, supporting observations, contradictory observations and remaining uncertainty. If the cause remains unresolved, report the candidate inconsistency and inspect alternative analyses. A transparent unresolved finding is preferable to an unsupported “corrected” series presented as unquestionably measured rainfall.
CHAPTER 6 OF 12 · Adjust and inspect the evidence
6. Derive the adjustment ratio in the right direction
For this next exercise only, assume independent evidence establishes an observation-system change, and the earlier 1.1 relationship is the chosen reporting basis. We want to express later values on that basis. The correction multiplier is desired slope divided by observed segment slope: 1.1/0.9 = 11/9, approximately 1.2222.
Apply it to the later annual target observations. Year 5 becomes 810 × 11/9 = 990 mm. Year 6 becomes 990 × 11/9 = 1,210 mm. Years 7 and 8 become 1,100 mm each. Their adjusted total is 4,400 mm, consistent with 1.1 times the 4,000 mm later reference total.
| Year | Original target, mm | Adjusted to early basis, mm |
|---|---|---|
| 5 | 810 | 990 |
| 6 | 990 | 1,210 |
| 7 | 900 | 1,100 |
| 8 | 900 | 1,100 |
Check the direction before multiplying. The later relationship is lower than the desired basis, so the multiplier should exceed one. Using 0.9/1.1 would reduce the values further and move them away from the stated aim. This directional check can catch a ratio inversion before the final total is calculated.
The adjustment is a model-based reconstruction. Keep the original observed values and label the new series clearly. The arithmetic cannot turn the reconstructed 990 mm into a direct observation made by the original gauge at its former location. Good recordkeeping preserves that distinction even when the adjustment is well justified.
CHAPTER 7 OF 12 · Adjust and inspect the evidence
7. Show why the chosen basis matters
There is another possible reporting basis. Suppose the later configuration is independently judged the more appropriate reference condition, and the aim is to express the early record on that basis. Then the early values would be multiplied by 0.9/1.1 = 9/11, approximately 0.8182.
The first four adjusted values become 720, 900, 1,080 and 900 mm. Their total is 3,600 mm. Together with the unadjusted later total of 3,600 mm, the full series totals 7,200 mm. By contrast, expressing the later record on the early basis gave a full total of 8,800 mm.
Both calculations can be arithmetically consistent with their chosen basis. They do not establish two equally valid statements of absolute true rainfall. Selecting the reporting basis needs an evidential and practical justification. A consistency graph alone cannot decide which observation configuration best represents the intended quantity.
This is a useful lesson about reference frames in data work. A number can change because the reporting basis changed even when the original observations remain preserved. The report should identify which periods were adjusted, why that basis was selected and what the adjustment is intended to support.
For a student check, ask which conclusion survives both versions. The record contains a relative relationship change at year 5 in the supplied model. Which conclusion does not survive? The reconstructed eight-year total. That total depends on the selected basis, so it should never be reported without the accompanying decision.
CHAPTER 8 OF 12 · Adjust and inspect the evidence
8. Inspect residuals and avoid misleading smoothness
A residual series can make the departure easier to see. Choose the early slope 1.1 as a provisional comparator and calculate annual residuals target minus 1.1 × reference. The first four annual residuals are zero in our exact example. The later residuals are −180, −220, −200 and −200 mm.
Accumulating those residuals gives zero through year 4, then −180, −400, −600 and −800 mm. The downward progression highlights the persistent later shortfall relative to the early relationship. It still does not tell us whether the shortfall came from observation changes or physical changes.
Real residuals fluctuate. A single unusual year can reflect weather variation rather than a durable break. A long run may deserve closer attention, but its interpretation depends on the reference, serial dependence, seasonality and the model used. A visually obvious bend is not automatically a formal statistical test.
Avoid fitting a straight line to cumulative points and treating a high correlation as proof of consistency. Both accumulated totals generally rise with time, and the shared accumulation structure can produce an impressive association. The question is whether the relative increments and segment relationships remain appropriate, not whether two growing sums correlate strongly.
The independent check is to reconstruct the curve from the annual table. If a suspicious bend vanishes after correcting a mismatched year label, the issue was in alignment rather than rainfall physics. Inspecting the intermediate record is often more informative than polishing the final graph.
CHAPTER 9 OF 12 · Transfer across networks and time scales
9. Transfer the method to a changing network
Now change the context. Suppose a reference average initially uses gauges R1, R2 and R3, but R3 stops reporting after year 4. Its location was consistently wetter than the other two. If later reference totals average only R1 and R2, the comparator may shift even if every remaining gauge is operating consistently.
The target–reference slope can therefore change without a target defect. This is why a reference series needs a documented construction history. Treating a changing mixture as one unchanged instrument can make a correct target look inconsistent.
One possible investigation compares a fixed subset of gauges available throughout the record. Another examines normalised station departures under an explicitly stated method. Neither choice is universally best; the appropriate construction depends on geography, coverage and the question. The important move is to test whether the suspected target break survives a defensible alternative reference.
For a numerical illustration, suppose a target is 1,000 mm each year. The original three-gauge mean is 1,000 mm, but the remaining two-gauge mean is 800 mm. The target-to-reference ratio rises from 1 to 1.25 although the target record has not changed. This is a comparator change, not evidence of a 25% rainfall increase at the target.
Ask the learner to state the missing fact before correcting anything: did the reference change because weather changed or because its composition changed? That question transfers the method from manipulating slopes to inspecting the evidence-producing system.
CHAPTER 10 OF 12 · Transfer across networks and time scales
10. Keep annual consistency separate from storm detail
An annual correction factor cannot automatically reconstruct the pattern within each year. Two years with the same annual rainfall can have different numbers of wet days, different storm intensities and different dry-spell lengths. The annual total has discarded those distinctions.
Consider two fictional years, each with 1,000 mm total rainfall. In one, rainfall is distributed across many moderate events. In the other, a few large events supply much of the total. Their positions on an annual double-mass curve could be the same, while their drainage implications differ substantially.
Multiplying every hourly observation by an annual adjustment factor adds a new assumption: the inconsistency acted proportionally at the hourly scale. That may be inappropriate for effects related to wind, precipitation type, event intensity or changing recording thresholds. The annual comparison does not establish that assumption merely because it gives a tidy annual total.
Missing intervals require particular care. A missing storm is not a proportionally low reading throughout a year. Scaling the available readings may reproduce a desired total while inventing a false time distribution. Preserve missingness and choose a reconstruction suited to the temporal question where reconstruction is justified.
The flow-duration guide offers a neighbouring lesson: a useful summary can remove information needed for a different question. Here the annual double-mass curve supports a consistency comparison, while storm behaviour needs data and methods at the relevant time scale.
CHAPTER 11 OF 12 · Practise and check
11. Practise an independent case and audit trail
Try a fresh invented case. Before a documented observation change, a target gauge accumulates 2,700 mm while the reference accumulates 3,000 mm. Afterwards, the target adds 1,800 mm while the reference adds 2,000 mm. Should the comparison alone suggest a proportional break?
Both segment slopes are 0.9. Under these supplied exact values, the double-mass comparison does not show a slope break. The existence of a maintenance event is a reason to inspect the record, but it does not compel an adjustment when the relevant relationship has remained stable in the evidence provided.
Now change the later target increment to 1,500 mm. Its slope becomes 0.75. If independent evidence supports adjustment to the earlier 0.9 basis, the multiplier is 0.9/0.75 = 1.2. A later annual value of 600 mm would become 720 mm. The ratio direction and reconstructed segment total provide independent checks.
A useful audit trail preserves original values, reference membership, aligned periods, segment definitions, the candidate explanation, the chosen basis, the multiplier and the adjusted values. Add the method version and the reason an adjustment was accepted or withheld. A reader should be able to reproduce both the arithmetic and the decision pathway.
For learning, remove the formula on a later attempt. Ask the student what relationship they want the adjusted record to have, then derive the multiplier from that aim. The resulting understanding is portable to other calibration problems where the labels and units change.
There is another useful check before closing the record: compare corrections with the same headline factor but different missingness. Suppose one later year records 900 mm because every event was measured proportionally low, while another records 900 mm because a major storm was missing entirely. Both annual totals could be multiplied by 1.2 to give 1,080 mm. Their reconstructed storm histories would still have different credibility.
In the first hypothetical mechanism, a proportional adjustment might be appropriate if supported by instrument evidence. In the second, multiplying existing events does not restore the omitted event’s timing or magnitude. The same final annual number can therefore conceal different evidential problems. A report needs the reason for the adjustment, not only its arithmetic.
For an independent sensitivity exercise, try desired slopes of 1.05 and 1.15 around the teaching basis of 1.10, while holding the observed later slope at 0.90. The later total of 3,600 mm would become 4,200 and 4,600 mm respectively. These are alternative assumed bases, not a statistically derived uncertainty interval. They show how a reporting decision propagates to the total.
Ask which additional observation could resolve that decision. An overlap comparison before and after relocation may be more informative than simply adding more years under the new arrangement. This does not guarantee a particular calibration method; it identifies why overlap evidence addresses the disputed relationship. A good next measurement targets the assumption carrying the conclusion.
Does a bend prove that a gauge is faulty?
No. It identifies a changed relationship with the selected reference. Observation changes, reference changes, physical differences and errors in alignment can all require investigation.
Does a straight line prove that the rainfall is accurate?
No. Shared biases or an unsuitable reference can hide errors. The graph is a consistency diagnostic, with a defined comparison and limited evidential reach.
Why use cumulative totals?
Accumulation helps reveal persistent differences in relative behaviour. Keep the original increments available too, because cumulative smoothness can obscure individual years and create misleading visual confidence.
Which slope belongs in the numerator of a correction?
The desired reporting-basis slope divided by the observed segment slope. Check the direction: a segment being raised to a higher relative basis needs a factor greater than one.
Should all unusual years be corrected?
No. Unusual weather is not automatically inconsistent observation. An adjustment needs a defensible account of the observation-system problem and a method suited to the intended use.
Can the same multiplier be used for hourly extremes?
An annual comparison does not establish that hourly errors were proportional. Examine temporal behaviour and the mechanism of the inconsistency before extending a factor across scales.
What is the most useful habit to carry forward?
Keep observed and reconstructed records separate, state the reference and reporting basis, and investigate the cause before correcting a pattern. A transparent record is more useful than a perfectly straight curve with its history removed.
Continue through the connected guides
Compare a connected investigation
Distinguish consistency in the observation record from the separate question of whether a catchment intervention changed an outcome. Continue with The Gold Standard of Hydrology | Paired Catchments and Attribution of Change.
Measurements, summaries and inferred components
The Hydrology root guide connects rainfall, storage, catchments and water accounting. The hydrograph-separation article distinguishes a measured total from a modelled partition. The flow-duration article distinguishes frequency from chronology. Read them together when deciding which features of a record a particular method preserves.
Evidence and practical water decisions
The environmental laboratory testing guide follows representative samples and clear reporting limits. The water-quality monitoring guide connects observations with the question they can answer. The water-resources engineering guide carries water estimates toward planning and community needs. None of these applications can be established by a tidy graph alone.
Mechanisms, restoration and transfer
The groundwater investigation guide examines subsurface pathways. The river-restoration guide connects interventions with functioning systems and evidence of recovery. Continue through the education parent for learning that transfers, or the How X Works hub to compare how evidence is used in other subjects.
Sources and scope
Searcy, Hardison and Langbein: Double-mass curves (USGS, 1960) is the foundational methodological reference for consistency comparisons, slope changes and the need to distinguish collection changes from physical changes.
The eight-year record, adjustment calculations and changing-network examples are original invented exercises. They are not adjusted observations from an actual station. The header photograph shows a canal and bridge at Punggol Waterway Park; it illustrates an urban water setting, not the fictional gauges. An annual consistency comparison alone does not establish a climate trend or an operational flood forecast.
The numerical datasets and classroom scenarios in the worked examples are invented for learning. Sources were checked on 21 October 2026. “Gold standard” describes the educational benchmark explained here; it is not an accreditation.
Hydrology root guide · The Gold Standard of Education · Explore How X Works