The Gold Standard of Watersheds | Bounding Mean Tracer Arrival Time When the Late Tail Is Unobserved

Choose a fictional injection origin and one downstream accounting section. Arrival time means elapsed time from that origin until tracer mass passes the section under the supplied model.

This definition differs from the age of every water molecule in a catchment, time spent specifically underground, or the duration of one storage process. A downstream tracer record can integrate several pathways without identifying them separately.

For the first exercises, tracer is conserved, all injected tracer eventually passes this section once, and its mass can be tracked exactly. Those are deliberate assumptions. They let the complete arrival-time distribution have total mass equal to the injected amount.

Later we will remove the eventual-passage assumption and see which conclusions survive. That order matters: the useful bound on a missing tail follows from a conservation and destination statement, not merely from the shape of a graph.

Write the origin, section and destination assumption above the first calculation. A result without that boundary can be numerically correct yet refer to a different population from the one the reader thinks is being timed.

Contents · Next chapter

Concentration describes tracer mass per water volume. Transported tracer mass also depends on how much water passes and for how long.

For an appropriately representative, background-corrected section concentration C(t) and water discharge Q(t), tracer mass flux is Q(t)C(t). Multiplying by a time increment gives a mass increment. Integrating produces recovered tracer mass.

If discharge is constant, it cancels from a ratio of concentration-weighted moments. If discharge varies, concentration alone generally does not supply the same mass weights. A tall concentration value during low flow can represent less mass than a lower value during high flow.

These statements assume that the measurement represents the relevant section and tracer. A point sensor in incompletely mixed water is not automatically a cross-section average. The mathematical multiplication cannot establish representativeness for us.

The first examples supply masses directly to isolate the observation-window issue. We will return to a concentration-and-discharge record after the main tail calculation, so that the meaning of the weights is secure before another measurement layer is added.

Contents · Previous chapter · Next chapter

A fictional pulse contains 100 grams. By the observation cutoff at four hours, exactly forty grams have arrived at one hour and forty at three hours. These are idealised point-arrival parcels, chosen to make the first moments easy to inspect.

The remaining twenty grams have not yet passed the section at or before four hours. Under the initial assumptions, they will arrive later. The ledger does not state when.

Recovered mass is eighty grams and the observed first moment is 160 gram-hours. The missing row is not a zero contribution. Its mass is known while its time-weighted contribution is unresolved.

The artificial point arrivals are not a claim that a real breakthrough curve has two instantaneous spikes. They provide a transparent distribution against which more complicated measurements can be checked.

Contents · Previous chapter · Next chapter

The mean among the eighty recovered grams is 160 divided by eighty, or two hours. Equivalently, each observed parcel has half the recovered mass, so its conditional mean is 0.5 × 1 + 0.5 × 3.

That calculation is valid. The mistake would be to remove the word recovered and announce that all injected tracer has a mean arrival time of two hours.

Relative to the injected 100 grams, the observed parcels each represent forty percent. The twenty-percent missing tail remains part of the complete population under the eventual-passage assumption. Renormalising the observed eighty to one has changed the denominator, not the physical fate of those twenty grams.

Preserve two separate outputs: recovered fraction 0.8 and recovered-only mean two hours. Together they describe more than a normalised graph alone. A graph with unit area can look complete even when the absolute mass ledger shows otherwise.

For a first learner check, ask which population the denominator eighty names. The answer should be a sentence about observed mass, not simply the instruction to divide by the total.

Contents · Previous chapter · Next chapter

Let μtail be the mean arrival time of the twenty grams still missing at four hours. The complete mean is μ = (160 + 20μtail)/100 = 1.6 + 0.2μtail.

Another useful form is μ = 0.8 × 2 + 0.2μtail. It separates the known observed fraction and conditional mean from the missing fraction and its unknown conditional mean.

This is a weighted-average identity, not a fitted transport model. It does not assume exponential decay, one storage zone or a particular distribution of pathways. It only needs the stated population to divide into the observed and later groups.

Because every tail arrival is later than four hours, a finite μtail exceeds four. Therefore μ exceeds 2.4 hours. The value 2.4 is an infimum: it can be approached by placing the tail arbitrarily close after four, but is not attained when all missing arrival times must be strictly greater than four.

That endpoint detail is small but meaningful. A bound should preserve whether a permitted construction actually reaches its endpoint.

Contents · Previous chapter · Next chapter

In Completion A, put all twenty missing grams at five hours. The complete first moment is 160 + 100 = 260 gram-hours, and the complete mean is 2.6 hours.

In Completion B, put the same twenty grams at twenty hours. The first moment becomes 160 + 400 = 560 gram-hours, giving a mean of 5.6 hours.

Both completions reproduce every observed entry through the four-hour cutoff. Both conserve the injected mass and satisfy eventual passage. Their complete means differ because the record has not resolved the later timing.

This is constructive evidence of non-identification. A method using only the observed prefix cannot distinguish these two histories unless it introduces additional assumptions. A smooth extrapolated curve might choose one history, but that choice comes from the extrapolation model rather than from the observed prefix alone.

Ask a learner to explain which number changed between completions and which recorded facts stayed fixed. They should point to the missing time-weighted contribution, rather than assume that the observed two-hour mean was calculated incorrectly.

Contents · Previous chapter · Next chapter

Place all missing mass at a finite time L greater than four. The complete mean is 1.6 + 0.2L. By choosing L as large as desired, that mean can exceed any proposed finite upper bound.

Each individual constructed distribution has a finite mean. We do not need an infinite-mean distribution to demonstrate that the family has no finite upper bound. The distinction separates a property of one distribution from a property of all distributions compatible with the record.

The conclusion is conditional on allowing arbitrarily late passage. If independent physical evidence establishes a maximum arrival time, it changes the feasible family and can supply an upper bound. The observed cutoff itself is not such evidence.

Do not write that the material must arrive soon because the main peak is over. That would substitute a visual impression for a tail constraint. A low-amplitude or small-mass tail can still contribute substantially to a first moment because late time multiplies its mass. The next chapter quantifies this without invoking any particular field process.

Contents · Previous chapter · Next chapter

Consider a different original pulse with 100 grams. Ninety-nine arrive at one hour, and one gram remains unobserved. Its recovered-only mean is one hour and recovery is ninety-nine percent.

If the final gram arrives at 1,000 hours, the complete mean is (99 × 1 + 1 × 1,000)/100 = 10.99 hours. High recovery has not guaranteed a complete mean close to the observed one.

This does not claim that every real tracer has such a late tail. It identifies what recovery alone cannot guarantee without information about timing. If the final gram were instead known to arrive by two hours, the complete mean would lie between one and 1.01 hours under the corresponding lower-time assumption.

The missing fraction and the possible missing times belong together. Reporting one without the other may conceal the most important uncertainty in the mean.

For a changed-input check, halve the missing mass but double its possible delay. Explain why a smaller missing fraction does not necessarily produce a smaller missing first-moment contribution. Work with mass multiplied by time before relying on a percentage label.

Contents · Previous chapter · Next chapter

Return to the original eighty-gram prefix. Now supply a new condition: all twenty missing grams arrive within the closed interval from five to twenty hours.

Then 5 ≤ μtail ≤ 20, and 2.6 ≤ μ ≤ 5.6 hours. Both endpoints are attained by the two completions already constructed. Every intermediate mean can also be attained, for example by placing all tail mass at the corresponding intermediate time.

These are sharp deterministic bounds under the supplied support condition. They are not a confidence interval or a probability statement that the true mean is more likely near the centre.

The interval narrows the answer because it adds information absent from the first record. Its source matters. A plausible-looking outer time limit cannot be quietly borrowed from the observation cutoff or another catchment.

If a reader asks why the interval is still wide, point to the twenty-percent mass whose timing remains unconstrained inside a fifteen-hour span. This connects the numerical width of the answer directly to a specific information gap rather than treating uncertainty as an unexplained disclaimer.

Contents · Previous chapter · Next chapter

If the recovered fraction is f, the observed conditional mean is μobs and the missing conditional mean lies in [a,b], then the complete mean lies in [fμobs + (1 − f)a, fμobs + (1 − f)b].

The interval width is (1 − f)(b − a). This identity makes two routes to improvement visible: reduce the unresolved mass fraction, or narrow its possible mean-time range. The routes answer different evidence questions.

In the original example, width is 0.2 × 15 = 3 hours. If independent evidence reduces the tail support to [5,10], width becomes one hour. If the support remains [5,20] but only five percent of mass remains unresolved, the width contribution becomes 0.75 hours.

The prefix mean and recovered fraction would need updating consistently in a real revised ledger; these are separate algebraic scenarios showing the sensitivity factors. Do not change f in one formula while retaining an incompatible observed mass total elsewhere.

The practical learning point is to ask which missing quantity controls the answer. More detail around the already well-observed peak may do less for this particular uncertainty than evidence about late mass.

Contents · Previous chapter · Next chapter

By one hour, forty grams have arrived. That is forty percent of the injected population but fifty percent of the recovered eighty grams. By three hours, the corresponding fractions are eighty percent and one hundred percent.

The normalised observed cumulative curve therefore reaches one at the last observed parcel. The complete cumulative curve reaches only 0.8 there and remains unresolved later.

Using the lower-quantile convention, the recovered-only median is one hour: its cumulative fraction first reaches one-half at the first parcel. The complete median is three hours: the injected fraction is only 0.4 at one and reaches 0.8 at three.

The different medians arise from different populations, not from an error in a quantile algorithm. The complete ninety-percent arrival time lies in the unobserved tail and is not identified by the prefix.

State the quantile convention because discrete point masses can create ambiguity in informal uses of median. The original parcel example makes that convention visible; a continuous curve can conceal the same denominator issue behind a smoother picture.

Contents · Previous chapter · Next chapter

Suppose additional evidence establishes μtail = 10 hours, so the complete mean is 3.6 hours. That does not determine the tail’s shape.

One compatible tail puts all twenty grams at ten hours. Another puts ten grams at five and ten at fifteen. Both tail means are ten, and both complete means are 3.6. Yet their late cumulative fractions and spread differ.

For the first completion, no tail mass arrives by six hours. For the second, half of the tail has arrived by then, adding ten percent of the injected mass to the prefix. A known mean cannot reconstruct every time-specific arrival fraction.

This distinction matters when a question asks about a particular time rather than an average. Use the summary suited to the question, and preserve the underlying distribution when it is available.

The existing Water Treatment lesson on bounds from mean residence time and time limits addresses a related but different task: controlling a model reaction output when distributional information is partial. Here the unresolved quantity is initially the complete arrival-time mean itself.

Contents · Previous chapter · Next chapter

The initial example assumed all 100 grams eventually pass the section. Remove that assumption. Suppose the missing twenty grams could instead leave by another route and never pass this particular section.

Then the observed eighty grams might be the entire eventual passage population. Its complete passage-time mean would be two hours. The same observed prefix can therefore support a different completion when the destination assumption changes.

It would be incorrect to assign an ordinary finite arrival time at this section to mass that never arrives. One can define other mathematical objects involving non-arrival, but they must be named rather than mixed silently into a finite passage-time mean.

A low recovery can have several possible explanations, including unresolved later passage, changed tracer behaviour, unobserved routes or measurement limitations. The arithmetic alone cannot select among them.

For a scientific account, record what evidence supports conservation and eventual passage through the chosen boundary. The observed mass deficit is a fact under the supplied measurement basis; calling all of it delayed storage is an interpretation requiring further support.

Contents · Previous chapter · Next chapter

Suppose the observed parcels remain forty grams at one hour and forty at three, but the injected amount is not established. The conditional recovered mean is still two hours because both its numerator and denominator come from the observed ledger.

The recovered fraction is no longer known to be 0.8. It depends on the injected amount and on whether that amount represents the population intended for the passage question.

If the injected total is only supplied in [90,110] grams and all of it eventually passes, the missing fraction ranges from 10/90 to 30/110. Combining that uncertainty with tail-time limits requires retaining the common injected-mass denominator throughout.

Do not calculate the smallest observed fraction using one total and the smallest missing mass using another as though they could occur together. They are coupled by Mmissing = Minjected − 80.

This is a useful extension for learners who can already calculate a simple weighted mean. The algebra stays modest, but the evidence discipline becomes more demanding: each uncertain input must remain connected to the balance that defines it.

Contents · Previous chapter · Next chapter

Keep the injected total M in [90,110] grams, eventual passage, the observed first moment 160 gram-hours and a tail confined to [5,20] hours.

For any fixed M, the complete mean lies between [160 + 5(M − 80)]/M and [160 + 20(M − 80)]/M. These simplify to 5 − 240/M and 20 − 1,440/M.

Both expressions increase with M over the positive interval. The overall lower bound is therefore 5 − 240/90 = 7/3 hours, approximately 2.333. The upper bound is 20 − 1,440/110 = 76/11 hours, approximately 6.909.

The lower endpoint uses M = 90 and tail time five; the upper uses M = 110 and tail time twenty. These explicit constructions show the endpoints are feasible under the supplied family.

This wider interval does not contradict the earlier [2.6,5.6] answer. That earlier answer knew M exactly as 100. Removing information expands the compatible family. The direction of that change is a useful sanity check on an uncertainty calculation.

Contents · Previous chapter · Next chapter

Consider a separate fictional section record with background-corrected concentration and representative discharge constant within each supplied interval. From zero to one hour, Q = 1 cubic metre per hour and C = 2 grams per cubic metre. From one to three hours, Q = 3 and C = 1 in the same units.

The first interval transports two grams, and the second transports six. Their exact first moments use the interval midpoint times because flux is stipulated constant: two times 0.5 plus six times two equals thirteen gram-hours.

The recovered mass mean is 13/8 = 1.625 hours. If someone instead weights only by concentration integrated over time, the interval areas are two and two, while their first moments are one and four. That gives 5/4 = 1.25 hours.

The discrepancy comes from dropping varying discharge. It is not fixed by refining decimal places or normalising both curves to one. The appropriate weights depend on which quantity is being averaged. All rates and intervals here are supplied exactly for the teaching calculation, not inferred from sparse field readings.

Contents · Previous chapter · Next chapter

A table of isolated concentration readings does not establish the concentration throughout every interval between them. Numerical integration then needs an interpolation or quadrature rule, whose assumptions should remain visible.

The previous example deliberately supplied constant values throughout each interval. Its exact mass and first moment followed from that stronger evidence. Reusing its arithmetic on two isolated samples would change the information basis.

For a simple uncertainty exercise, suppose forty grams arrive somewhere within [0,2] hours and another forty somewhere within [2,4], but the exact times are unknown. Their combined first moment lies between eighty and 240 gram-hours under closed-bin bounds. The conditional mean therefore lies between one and three hours.

If an actual partition uses half-open bins to avoid double membership at endpoints, some endpoint bounds may be limiting values rather than attained values. Preserve the chosen convention.

Assigning every bin’s mass to its midpoint gives a convenient approximation. It does not convert coarse timing evidence into exact arrival times. A good report distinguishes the approximation from what the supplied bins alone establish.

Timing deserves its own check. If every supplied arrival time includes the same extra half-hour clock offset, every correctly weighted arrival mean also includes that half-hour. The recovered mass and its fraction can remain unchanged while the time interpretation shifts. Correcting only the observed parcel times while leaving the cutoff and tail limits on the old clock would create an inconsistent record. Put injection time, sample time and support limits on the same declared basis before comparing moments. This exercise distinguishes a shared origin error from an unresolved tail: an origin error shifts known times, whereas missing late evidence leaves their additional contribution undetermined. One correction cannot be used as a substitute for the other.

Contents · Previous chapter · Next chapter

A late tracer signal may become difficult to distinguish from background. A measured zero after correction and an unresolved signal below a reliable detection basis are different statements.

Drummond and colleagues’ 2012 primary study of breakthrough-curve truncation distinguishes ending observation too early from losing reliable sensitivity to the tail. That distinction motivates the separate information gaps in this guide.

The original parcel cases do not simulate detector noise or specify a detection protocol. They assume exact supplied masses so that the logical consequences of a missing tail can be isolated.

In a real evidence record, background treatment, calibration, section representativeness, timing and discharge would need appropriate methods. A small baseline offset integrated over a long period can alter recovered mass and moments. Calling every late low signal zero would introduce an unsupported selection.

For learners, label each input as supplied, measured under a stated method, inferred or unresolved. The labels make it easier to see which part of a mean is arithmetic and which part depends on measurement and interpretation.

Contents · Previous chapter · Next chapter

Return to the original site’s observed mean of two hours and eighty-percent recovery. Suppose a second fictional site has a fully known eventual-passage mean of four hours under a compatible definition.

Using Completion A for the first site gives 2.6, below four. Using Completion B gives 5.6, above four. Both reproduce the same first-site observed prefix.

Therefore the observed prefix does not identify the ordering of complete means. Announcing that the first site transports tracer faster because its recovered-only mean is two would compare a conditional mean with a complete mean.

A common observation cutoff can make prefixes more consistently described, but it does not guarantee equal missing fractions or equal unresolved tail behaviour. It repairs one reporting difference, not every physical or statistical difference.

The comparison should state which ranking is established, if any, and what additional information would settle the intended question. Sometimes the most useful answer is an interval crossing the other site’s value, because it shows exactly why a confident ranking is premature.

Contents · Previous chapter · Next chapter

A delayed arrival pattern can be compatible with several pathways or processes. The original constructed late parcel represents unresolved timing, not a labelled underground store.

Even if a long tail is established, separating surface storage, subsurface exchange, tracer interactions and instrumental effects can require other evidence and an appropriate transport model. A mean alone is still less informative than the full curve for that task.

The existing Soil Science breakthrough guide examines preferential pathways in a different setting. The hyporheic residence-time guide separates heat evidence from water-age interpretation.

Use these connections to ask a better mechanism question. Do not rename the missing twenty grams hyporheic storage merely because that term is available. The conservation ledger establishes a missing contribution under stated assumptions. Its physical explanation remains a separate layer of the account, to be tested with observations that distinguish alternatives.

Contents · Previous chapter · Next chapter

Try this changed case. A fictional pulse contains eighty grams. By six hours, twenty grams have arrived at two hours and forty at five. All remaining mass is independently stipulated to arrive within [8,14] hours.

Recovered mass is sixty grams, or seventy-five percent. Its first moment is twenty times two plus forty times five, giving 240 gram-hours. The recovered-only mean is four hours.

The missing twenty grams contribute between 160 and 280 gram-hours. Complete mean bounds are therefore (240 + 160)/80 = five hours and (240 + 280)/80 = 6.5 hours. Placing all missing mass at eight or fourteen attains the respective endpoints.

Notice that simply adding the average tail time to the observed four would be wrong. The two groups have unequal masses and must be combined with their population weights.

For an additional test, remove the maximum tail time fourteen while preserving the minimum eight. The lower bound remains five, while the finite upper bound disappears. Explain which supplied fact supported each side of the original interval.

Contents · Previous chapter · Next chapter

Once the information gap is explicit, ask what additional evidence would answer the intended question. If the target is the complete first moment, reliable late mass and timing are directly relevant. If the target is a pathway mechanism, a different observation may be needed.

Do not treat a larger peak, a smoother fitted curve or a longer observation period as automatic solutions. A longer period can still leave sensitivity limits unresolved. More points inside an already well-resolved early peak may not constrain the unobserved tail.

For a desk exercise, offer three hypothetical additions: an exact extra peak reading, a justified maximum tail time, or a confirmed alternative exit route. Ask which changes the conditional mean, which changes the complete-mean bounds and which changes the population definition.

The learner should explain the connection before calculating. Evidence is useful because of the constraint it adds to a particular question. A record containing more numbers can remain uninformative about the quantity that matters if all of its new numbers describe the wrong part of the process.

Contents · Previous chapter · Next chapter

Keep injected amount, section, time origin, observation cutoff, recovered mass, first moment and recovery fraction together. State whether mass is supplied directly or derived from concentration and discharge.

Preserve the unnormalised ledger as well as any normalised curve. Record assumptions about conservation, eventual passage, non-arrival and tail support. If a bound is strict or only approached, retain that endpoint meaning.

Separate numerical integration uncertainty from missing-data uncertainty and physical-model uncertainty. An exact spreadsheet sum cannot determine an unobserved tail. A smooth interpolation cannot prove that its assumed shape was present between samples.

A useful peer check asks another learner to reconstruct the complete-mean interval and one compatible history at each endpoint. Then ask for a different history with the same complete mean but a changed late fraction. Those tasks test whether the record preserves both its numerical result and the information it deliberately leaves unresolved.

The article’s examples require no field release or practical intervention. They train a way of reading and questioning evidence before a specialist method is chosen.

Contents · Previous chapter · Next chapter

For repair, use two observed parcels and ask what the denominator counts. Calculate the recovered-only mean, then add one explicitly supplied late parcel. Keep mass and time units visible throughout.

For stabilisation, hold the prefix fixed and vary the tail. Construct two different complete histories with the same observed record. Then supply time limits and show how the answer becomes an interval with attainable endpoints.

For extension, vary discharge, remove eventual passage, or make the injected amount uncertain. Require the learner to explain which earlier identity still applies before doing arithmetic. Different uncertainties should not be combined as though they were unrelated knobs.

An independent task should change the information, not merely the numbers. Replace exact arrival times with bins, or replace a known injection with an unknown total. A learner who notices that an exact answer has become a bound has transferred the underlying reasoning beyond the original exercise.

Contents · Previous chapter · Next chapter

Is the recovered-only mean incorrect?

No. It correctly describes the observed population under its measurement assumptions. The error is presenting it as the complete mean without justification.

Does high recovery guarantee a nearly correct complete mean?

Not without control of the missing times. The ninety-nine-percent example supplies a direct counterexample.

Can a fitted tail supply the missing answer?

It can supply a model-conditional estimate. Preserve its assumptions and distinguish it from information established by the observed prefix alone.

Does missing mass prove slow storage?

No. The initial bound assumed eventual passage. Removing that condition admits other explanations and can change the appropriate population.

Does a common cutoff make two complete means comparable?

It makes the observed windows alike in one respect. Different unresolved fractions and tails can still prevent a complete-mean ranking.

Contents · Previous chapter · Next chapter

For the accounting foundation, use Watersheds and the existing residence-time distributions and tracer curves lesson. That broader lesson introduces normalisation, moments and truncation. This article develops constructive completions, sharp support-based complete-mean bounds, quantile changes and coupled injection uncertainty.

For physical interpretation, compare preferential flow and solute breakthrough with groundwater–surface-water exchange and hyporheic residence time. For a different partial-information calculation, use remaining-fraction bounds from mean residence time and time limits.

The primary specialist source is Drummond and colleagues, Effects of solute breakthrough curve tail truncation on residence time estimates: A synthesis of solute tracer injection studies, published 11 August 2012 and checked 26 October 2026. It supports the distinction between observational and sensitivity truncation. The parcel distributions, bounds and changed-input exercises here are original teaching constructions, not reproduced study measurements.

Continue with the Education parent or How X Works when the next learning task asks what an incomplete record can genuinely determine.

Contents · Previous chapter · Return to Watersheds