Recovery begins when pumping actually stops
A pumping-test recovery record begins at shutdown, when the imposed extraction rate falls to zero and water levels start moving toward their pretest condition. The measured rise is evidence of the aquifer and well responding to a history of pumping. It is not proof that the aquifer has fully replenished its regional groundwater reserve. The pressure response follows hydraulic diffusion through the connected formation, modified by well storage, boundaries, and other stresses.
The practical question is how to read Theis recovery plots, residual drawdown, the time ratio t/t′, and the transmissivity estimated from a semilog slope. Those terms describe an analysis carried out after a known pumping episode, not another constant-rate drawdown test. The shutdown time, discharge history, baseline level, and observation geometry must all be established before interpreting a curve. A smooth rising water level alone is not a complete aquifer measurement.
This guide teaches first principles, field-record checks, a dimensionally consistent synthetic calculation, common mistaken interpretations, and independent transfer exercises with answers. The examples are invented solely as mathematical teaching data, never presented as observed groundwater measurements. A reliable resource or pumping decision requires actual data, hydrogeological context, suitable analytical methods, and qualified professional assessment.
Separate the water-level rise from residual drawdown
Residual drawdown s′ is the difference between the appropriate baseline hydraulic head before the pumping test and the hydraulic head still observed during recovery. If water levels rise after pumping ceases, the residual drawdown normally decreases toward zero. The amount recovered is a different quantity: it is the drawdown present at shutdown minus the remaining drawdown at a later moment, after accounting for the intended baseline and corrections.
Imagine a well whose depth to water, measured downward from a fixed casing reference, was 5.00 metres before pumping. At shutdown, it is 7.00 metres, giving 2.00 metres drawdown. Ten minutes later, depth to water is 5.80 metres, leaving 0.80 metre residual drawdown. The amount recovered is 7.00 minus 5.80, or 1.20 metres. Thus 60 percent of the shutdown drawdown has recovered, while 40 percent remains.
A report that describes 1.20 metres as residual drawdown has reversed these concepts. The correction is to write the baseline, shutdown level, present level, and defined difference explicitly, preferably as a small calculation alongside the time series. A depth measurement increases downward, whereas the actual water surface rises during recovery. Keeping a sign convention visible stops this familiar but consequential mistake.
Establish both clocks
A Theis recovery calculation uses two elapsed times. Let t_p denote the duration of pumping from the true start to the true stop. Let t′ denote elapsed time since the pump stopped. At any recovery observation, t = t_p + t′ is the elapsed time since pumping first began. The ratio for the classical recovery plot is R = t/t′ = (t_p + t′)/t′, dimensionless.
Suppose the pump runs for 3,600 seconds and is then stopped. Five minutes after shutdown, t′ = 300 seconds, while t = 3,900 seconds, not 3,600 seconds. Therefore R = 3,900/300 = 13. At twenty minutes after shutdown, t′ = 1,200 seconds and t = 4,800 seconds, so R = 4. The ratio becomes smaller as recovery continues.
A widespread arithmetic error uses t_p/t′ in place of the total-time ratio. That shortcut is not equal to R, particularly when the recovery time is comparable to the pumping duration. Another error measures t′ from the last observation before shutdown rather than the actual cessation of discharge. Both can change the horizontal positioning of data and distort the slope or interpretation.
Why a pump stopping is a hydraulic event
In the ideal linear groundwater-flow model, a pump running at constant discharge Q is a stress that begins at a known time. When it stops, the mathematical effect can be represented as continuing the original withdrawal and simultaneously adding an equal and opposite discharge at the same location. The superposed negative stress cancels the future extraction while preserving the influence of the preceding pumping period.
This virtual-well construction does not mean anyone physically injects water when the pump is switched off. It is a way to calculate the net hydraulic head response from two mathematical histories. The continuing withdrawal component has acted for t seconds, and the cancelling component has acted for only t′ seconds. The difference produces the recovery signal. It explains why the time ratio contains both clocks.
Superposition is most defensible for linear flow assumptions. Strong changes in saturated thickness, complex boundary conditions, variable-rate histories, and nonlinear well losses may demand other approaches. The recovery method becomes powerful when its simplifying conditions approximately describe the measured interval, not because the algebra magically removes all geology from a pumping-test dataset.
Start with the raw recovery trace
Before opening a logarithmic graph, plot measured depth or head against actual clock time across the end of pumping and the complete recovery period. Mark the exact shutdown, any gradual reduction in flow, changes in instrument location, and external pumping nearby. Confirm whether the plotted water level genuinely rises, stalls, oscillates, or continues drifting in a direction inconsistent with the recorded operations.
A useful field record includes the pretest hydraulic head, pumping start and end times, discharge measurements, observation-well distances, and sensor specifications. When water levels are reported as depths from casing tops, record the reference benchmark and its surveyed elevation if different wells must be compared. A single reference error can make two observation wells appear to behave differently even if their true changes are similar.
An immediate steep rise in the pumped well after switch-off may reflect disappearance of dynamic well loss as well as aquifer recovery. That is why observation-well records can be valuable: they are not extracting water through the same screen at the instant of shutdown. Preserve raw and corrected traces separately so later analysts can see which portion of the apparent rise is a measurement and which part is an analytical adjustment.
Confirm that discharge truly reached zero
Pressing a switch is not always the same as hydraulic shutdown. A pump may coast down, an automatic valve may change flow gradually, or water may continue moving through pipework for a brief interval. The effective cessation of discharge should be documented from suitable instrumentation or operating observations. An uncertain stop time is especially damaging for very early recovery readings because t′ is then small.
Suppose the reported stop occurs at 14:00:00, but discharge actually declines over forty seconds. The observation labelled t′ = 20 seconds cannot honestly be treated as twenty seconds after an instantaneous full shutdown. In some systems the interval may be discarded for simple recovery analysis, or the flow history may be modelled more carefully. Inventing a precise zero-time from an operational note is not an appropriate repair.
This concern matters less at times far larger than the coast-down period, but that depends on the test. Always distinguish the timestamp of an electrical action, the point at which discharge became negligible, and the first valid sensor observation. The pump schedule is the forcing function of the groundwater experiment, and mistakes in the forcing function propagate into interpretation.
Build the residual-drawdown column
With a stable pretest water level, residual drawdown is calculated for each recovery observation by subtraction. Using depth measured downward, s′ equals current depth to water minus pretest depth to water. If the original depth was 5.00 metres and successive recovery depths are 6.40, 6.00, and 5.60 metres, the residuals are 1.40, 1.00, and 0.60 metre respectively.
If the reference water level was already moving before pumping started, simple subtraction can mix the imposed pumping response with a regional trend. A nearby control well, known external pumping schedule, or other defensible correction may be needed. State exactly what baseline was used and why it is appropriate. A negative residual late in the trace is a reason to inspect background changes, rather than automatically set it to zero.
Retain the measured level and calculated residual in separate columns. Readers should be able to reconstruct every difference. A spreadsheet that contains only processed residuals can hide whether a casing datum was changed, which missing observations were interpolated, or whether a slowly varying baseline was assumed. Traceability is often more valuable than a highly polished final graph.
Form the time-ratio series without units
For every observation, calculate R = (t_p+t′)/t′. Both numerator and denominator must use the same time unit, so seconds divided by seconds or minutes divided by minutes yields a dimensionless ratio. Do not put seconds in one column and minutes in the other unless they have been converted first. When the pumping duration is positive and t′ is positive, R is greater than one.
For a one-hour pumping period, at t′ = 60 seconds the ratio is 61; at t′ = 600 seconds it is 7; at t′ = 3,600 seconds it is 2. The ratio approaches one as recovery continues for much longer than the original pumping period. Consequently its base-ten logarithm declines toward zero with increasing time. These are useful automatic spreadsheet checks.
A ratio of less than one usually signals reversed division, misinterpreted clock columns, or another error for the simple setup. Such a ratio would also create a negative log10(R), contrary to the expected positive residual-drawdown relation in the idealized equation. Inspect the timing before attempting to explain the sign through unusual aquifer behaviour.
The Theis recovery equation
Under classical late-time assumptions for confined radial flow after constant-rate pumping, the residual drawdown can be written s′ = [2.303 Q/(4πT)] log10(t/t′). Here Q is the prior constant pumping discharge in cubic metres per second, T is transmissivity in square metres per second, and s′ is residual drawdown in metres. The numeric factor 2.303 converts the natural logarithm to base ten.
An equivalent natural-log expression is s′ = [Q/(4πT)] ln(t/t′). Do not mix the two forms. Using log10 without the 2.303 coefficient, or using ln while retaining it, creates a factor-of-2.303 numerical error. The exponent and log base are not mere display preferences: they define how a measured slope is translated into T.
This compact formula follows from a late-time approximation to the Theis well function and the superposed stop event. It is not universally exact at the earliest moments or in every aquifer. Wellbore storage, heterogeneity, leakage, delayed drainage, partial penetration, and finite boundaries can disturb the relation. An analyst must establish a suitable interpretation interval before using its slope.
What a one-decade slope means
Plot residual drawdown s′ on an arithmetic vertical axis against log10(R) on the horizontal axis. Within an appropriate straight-line region, the vertical change Δs′ corresponding to a tenfold difference in R is the semilog slope magnitude. The equation implies T = 2.303 Q/(4πΔs′). This form can be evaluated without needing a numerical intercept for storativity.
A decade means multiplication or division of the ratio by ten, such as R = 100 to R = 10. It does not mean a ten-minute period, a ten-second difference, or a change from R = 10 to R = 20. The slope is a difference in metres for one unit of log10(R). If the chosen data cover less than a full decade, a fitted slope may be extrapolated to one log cycle with the fitting procedure documented.
Since R decreases during recovery, measured points move toward the left on an axis whose values increase to the right. Residual drawdown also declines, so the fitted relation has a positive slope when s′ is plotted against log10(R), even though s′ falls forward in real time. Confusing graph direction with physical movement can lead to an erroneous negative transmissivity.
A reproducible synthetic recovery experiment
Assume a well was pumped at Q = 0.008 cubic metre per second, equivalent to eight litres per second, for exactly 3,600 seconds. Imagine an idealized aquifer response with T = 0.003 square metre per second. These chosen numbers define a teaching model; they are not an observation from a real site and are not a recommended pumping rate.
The coefficient multiplying log10(R) is 2.303 × 0.008 divided by 4π × 0.003. Its numerator is 0.018424 cubic metre per second; its denominator is approximately 0.037699 square metre per second. The result is about 0.4887 metre per unit of log10(R). Units cancel to metres because a discharge rate divided by transmissivity has dimensions of length.
Thus the ideal residual drawdown is approximately 0.4887 × log10(R) metres for this construction. The relationship is intentionally simple enough to check with ordinary calculator arithmetic. The interpretation still depends on whether the assumed aquifer and pumping history could plausibly produce such a response; equations can be internally consistent while being physically inappropriate for a particular well.
Calculate the time ratio at several instants
At 100 seconds after shutdown, the total time since pumping began is 3,700 seconds. R = 3,700/100 = 37, and log10(37) is about 1.5682. The ideal model predicts residual drawdown 0.4887 × 1.5682 ≈ 0.766 metre. This is the water-level deficit remaining relative to the baseline, not the amount of recovery since the pump stopped.
At 400 seconds after shutdown, R = 4,000/400 = 10, so log10(R) = 1 and s′ ≈ 0.4887 metre. At 900 seconds after shutdown, R = 4,500/900 = 5, log10(R) ≈ 0.6990, and s′ ≈ 0.3416 metre. At 3,600 seconds after shutdown, R = 7,200/3,600 = 2 and the predicted residual is about 0.1471 metre.
Notice how the water level approaches its reference condition as time increases. The formula does not predict that all recovery completes after one pumping duration; at that moment the ratio is still two, not one. Late recovery can become slow and difficult to distinguish from environmental variability. A small residual is a physical observation to measure carefully, not a promise that the aquifer has experienced no lasting change.
Infer transmissivity by reversing the slope
Imagine a synthetic semilog plot with suitable straight-line points at R = 100 and R = 10. Suppose corresponding residual drawdowns differ by 0.4887 metre. The change is over exactly one decade, so Δs′ = 0.4887 metre. Substitute Q = 0.008 cubic metre per second into T = 2.303Q/(4πΔs′). The estimate is approximately 0.00300 square metre per second.
Converting to square metres per day, multiply by 86,400 seconds per day to obtain about 259.2 square metres per day. Reversing the formula returns the same 0.4887-metre semilog slope to rounding precision. Those two directions of the calculation are a useful arithmetic audit, especially when the regression result was produced by unfamiliar software.
If an author reports T in metres per second rather than square metres per second, the calculation has lost a dimension. Hydraulic conductivity uses length per time, while transmissivity uses area per time. Converting between them requires a relevant saturated thickness and a supporting hydrogeological assumption. Recovery analysis by itself does not automatically identify which thickness should be divided into the estimated T.
Why the initial recovery may be unsuitable
Immediately after shutdown, the pumped well can still be affected by water stored in the casing, pump coast-down, local well loss disappearing, and turbulence settling. The first measured rise may therefore be much faster than the characteristic regional aquifer response. Treating those points as the classical Theis recovery straight line can produce a misleading transmissivity estimate.
A nearby observation well may reduce some active pumping-well losses, but it does not make the measurement immune to all dynamic effects. Instrument resolution, incomplete stop timing, vertical hydraulic gradients, partial penetration, and hydrogeological boundaries can still matter. A small t′ also makes the ratio highly sensitive to a few seconds of timing error, further reducing confidence in very early points.
Inspect the raw trace, rate history, and log-ratio plot. Choose an interval supported by the model’s applicability and residual behaviour, not merely by its correlation coefficient. If the data never enter a defensible interval, do not force the result. Report that the recovery record is insufficient for this form of analysis and explain which next measurement would help.
The approximation has a domain of validity
The theoretical Theis response assumes a sufficiently extensive, homogeneous, isotropic confined aquifer with uniform thickness, radial flow and an idealized fully penetrating pumping stress. The familiar semilog recovery form is derived from an approximation that becomes useful under appropriate dimensionless time conditions. It is not guaranteed merely because several plotted points look nearly straight.
For a confined aquifer, the dimensionless argument u = r²S/(4Tt) characterizes the late-time form at an observation distance r, where S is storativity. A recovery superposition involves corresponding arguments for the elapsed pumping and recovery components. Evaluate whether the late-time approximation is plausible for both relevant terms where the selected formulation requires it, particularly at the early recovery times being considered.
A published USGS explanation of the Theis recovery method explicitly connects its validity to the assumptions of the Cooper–Jacob approximation and a small-u condition. The exact screening depends on model geometry and the observed time range. A field analyst should not replace that assessment with a universal claim such as ‘all readings after five minutes work’ because distance, transmissivity and storage vary widely.
Why simple recovery does not directly identify storage
The late-time residual-drawdown formula contains Q, T and the logarithmic time ratio, but not an independently identifiable storativity term. Storativity is involved in the broader diffusion physics and in the applicability of the approximation, yet it cancels in the simplified late-time difference under the ideal formulation. It is therefore misleading to take the recovery straight-line intercept and announce a uniquely measured S.
Estimation of S may require pumping-phase time or distance data, a full response model, known geometry, multiple observation wells, or another defensible source of constraints. Even then, storage estimates are vulnerable to partial penetration, boundaries, unconfined drainage and baseline errors. The absence of S from the simplified recovery equation is a clue about what information that particular line can and cannot provide.
If software generates an S value from a more elaborate recovery fit, examine the model and inputs rather than rejecting it automatically. The point is that such an estimate is not justified by the simple one-decade slope formula alone. A concise report should specify whether recovery was used to constrain only T or whether additional information supported storage identification.
Compare recovery with pumping-phase drawdown
A constant-rate pumping analysis observes drawdown as pumping continues, often plotting it against log10 elapsed time. The recovery analysis observes the remaining head deficit after shutdown and plots it against log10(t/t′). Under compatible model assumptions and sufficiently high-quality records, both may yield comparable estimates of transmissivity.
For an instructional companion to this method, the Cooper–Jacob constant-rate drawdown guide shows a worked semilog calculation using the during-pumping signal. The present recovery approach makes shutdown and residual head its primary reader task. Do not interchange the two horizontal axes or copy a stopping-time ratio into a pumping-only expression.
If the two analyses disagree, first inspect discharge constancy, baseline corrections, fitting windows, early storage, and whether one analysis relied on the pumping well while the other used observation wells. Geological boundaries or flow nonlinearity may also matter. A discrepancy is diagnostic evidence, not automatically grounds to average the values or discard the lower estimate.
Distinguish recovery from step-drawdown efficiency
Step-drawdown testing deliberately changes pumping discharge through multiple rates and analyzes rate-dependent losses, including the extra resistance that can grow faster than linearly with Q. The Jacob–Rorabaugh well-loss guide examines coefficients and model-based hydraulic efficiency. Those parameters should not be confused with residual drawdown after shutdown.
Well losses can collapse quickly when pumping ends, producing a large immediate rise in the pumped-well level that is not simply far-field aquifer recovery. A test with severe rate-dependent losses may therefore show dramatic early rebound even if aquifer pressure continues restoring much more slowly. The practical repair is to separate the well’s local hydraulic behavior from the broader aquifer response where the data and model allow.
A word like efficiency is particularly susceptible to misuse across these reports. Hydraulic well efficiency is not the same as recovery percentage or electric pump efficiency. Writing what numerator and denominator mean is safer than relying on a shared everyday adjective. The relevant physical quantity for the classical Theis recovery slope remains residual drawdown per logarithmic time-ratio cycle.
A variable-rate history changes the mathematics
Real pumps sometimes fluctuate rather than remaining at one exact rate for the entire stress period. The classical single-rate recovery expression assumes a known constant rate that is fully stopped at the recorded shutdown. If Q changes substantially, the prior aquifer stress is not represented accurately by one rectangle of constant extraction.
A linear groundwater model can handle multiple rate changes through superposition: each incremental increase or decrease is a separate hydraulic input beginning at its own timestamp, with the associated response evaluated at its own elapsed time. Shutdown is then another negative rate increment. The net head response is the sum of all such contributions, which is not in general identical to the one-term simple recovery equation.
Keep a table of each measured rate segment, its start and end, and the corresponding observations. If rate measurements were never recorded, a sophisticated-looking Theis line cannot restore that missing information. Explain the limitation and consider whether the test supports a rough comparison or whether a more carefully controlled repeat would be necessary for a decision.
Boundary effects can persist after stopping
A pumping-induced pressure change can reach a hydraulic barrier or a recharge boundary during the pumping period or recovery. When the pump stops, those boundary influences do not disappear instantaneously. They continue shaping the evolving head response. A late-time recovery curve can therefore differ from a simple extensive-aquifer line even when the discharge history was well controlled.
An impermeable boundary may have increased drawdown during pumping, while a connected recharge source may have reduced it. The recovery signature depends on spatial configuration and timing; one downward bend is not enough to identify a river, fault or geological contact. Compare several observation wells and consider whether a candidate boundary explains both pumping and recovery records.
Before interpreting deviations as geology, rule out rate changes, independent nearby pumping, rising river stage, barometric effects and logger drift. A defensible report distinguishes what the residual series visibly does from the physical explanation proposed for it. Ideally, the explanation predicts observations at other times or locations, making it testable rather than merely plausible.
Unconfined drainage is not a simple confined response
A water-table aquifer can release water through drainage as the saturated thickness changes. Its response may include delayed yield and vertical redistribution that a confined Theis idealization does not capture adequately. Recovery can exhibit characteristic time-dependent differences when refilling pores and adjusting gradients. A semilog line may still be drawn, but the inferred parameter depends on whether the model applies.
If the original pumping drawdown was large relative to saturated thickness, the aquifer geometry changed during extraction. A straightforward confined recovery approximation may then misrepresent the process. Check water-table depth, screened interval, degree of penetration and evidence for delayed drainage. Some cases require models designed for unconfined conditions rather than a simple textbook expression.
This does not make every shallow aquifer recovery curve useless. It means the analyst must state what simplified quantities can be obtained and whether they should be treated as approximate. Geological honesty is more valuable than insisting on a perfect mathematical pattern from a system whose physical boundaries violate the derivation.
Leakage, layers and partial penetration
Aquifers may receive water across semiconfining layers, contain high-conductivity channels, or exhibit strong vertical anisotropy. A pumping well screened through only part of an aquifer can create three-dimensional flow near the well. The resulting drawdown and recovery vary with observation depth as well as distance. A one-dimensional semilog presentation cannot encode all of that structure.
Imagine one observation well screened high in a formation and another screened low. Their recovery traces differ even at similar horizontal distances. That difference may reflect vertical hydraulic resistance, leakance, well construction or differing connections to fractures. A simple average of fitted transmissivities can conceal the very heterogeneity a groundwater investigation needs to understand.
Document well completion details and the conceptual aquifer section alongside the graphical analysis. If leakage or partial penetration is material, use a model that acknowledges it or present the simple result as conditional. A claim that a recovery slope proves uniform regional permeability exceeds what a single monitored depth can establish.
Observe the end of recovery long enough
An attractive early recovery curve is not a complete record. Monitor until the response approaches a baseline to the degree practical and informative for the hydrogeological setting, while also noting when external stresses make further interpretation unreliable. The remaining head deficit may become small compared with transducer resolution and natural groundwater variations, so a long numerical series need not contain equally useful information throughout.
Suppose a sensor resolves 0.01 metre and late residual drawdown fluctuates between zero and 0.02 metre. Fitting that tail to numerous decimal places is unsupported; quantization, drift and background events can dominate. State the resolution and the range in which the response was distinguishable. Some late-time data may be valuable for identifying a different process rather than extending the original straight-line fit.
There is no universal endpoint defined as ‘one hour after shutdown’ or ‘when ninety percent has recovered’. The duration depends on pump history, aquifer diffusivity, radius, boundaries, instrument performance and intended questions. A reader needs the recorded end condition and its limitations, not an arbitrary completion badge.
Compare observation wells at different radii
The pressure disturbance propagates outward during pumping and the effects of shutdown evolve spatially as well. Wells at different distances may therefore show different onset and recovery histories. In an ideal model, a suitable late-time slope can be compared across distances to test consistency of transmissivity. In a heterogeneous formation, some discrepancies may be real reflections of spatial structure.
Survey the wells and record the depth interval actually being monitored. A distant observation point may have a small or delayed response that is hard to separate from natural fluctuations. A close pumped well may have substantial local storage effects. Neither proximity nor distance alone guarantees superior data; each changes the trade-off between signal strength and model contamination.
USGS aquifer-test reports have applied recovery methods to observation wells alongside pumping-phase analyses, illustrating why multiple records are more informative than one neat line. When different observation wells yield distinct estimates, inspect their completion, model applicability and boundary context before treating the values as interchangeable samples of a single material property.
Rate and baseline errors create numerical biases
The transmissivity expression is proportional to Q and inversely proportional to the semilog recovery slope Δs′. If the flow meter overstates Q by ten percent while the slope is correctly measured, the computed T will also be overstated by about ten percent. If the slope is overstated by ten percent while Q is correct, the inferred T becomes 1/1.1, roughly 90.9 percent of its previous value.
Those simple arithmetic sensitivities are not a complete uncertainty analysis. Model selection, screen geometry, geological heterogeneity, other wells’ pumping, incomplete shutdown and pressure-sensor baseline drift can cause errors that do not behave like small independent measurement noise. A narrow regression confidence interval cannot eliminate uncertainty about whether the recovery equation describes the physical test.
Report the pumping-rate measurement method, baseline treatment, selected recovery interval, and alternative interpretations when they matter. Indicate which assumptions were observed or measured and which were adopted as approximations. The quality of the final transmissivity is constrained by the least secure important part of that chain.
Check every conversion of units
Discharge commonly appears in litres per second, gallons per minute or cubic metres per day. Transmissivity appears in square metres per second or square metres per day. The coefficient in a formula may be different if it has been rearranged for a particular imperial unit system. Before substituting any number, select a coherent unit set and write the dimensions next to Q, s′, T and time.
In the synthetic calculation, 8 litres per second equals 0.008 cubic metre per second. A reported T of 0.003 square metre per second corresponds to 0.003 × 86,400 = 259.2 square metres per day. It does not mean 259.2 cubic metres can be pumped daily from a well without other information; volumetric extraction depends on gradients, geometry, boundaries and operating conditions.
The ratio R is dimensionless only when its time values use the same unit. If one elapsed time is copied as hours and the other as seconds, the ratio can be wrong by a factor of 3,600. An automatic check is that R > 1 and approaches one as t′ increases. A compact unit-and-range audit catches errors before a fit produces an apparently credible transmissivity.
A clear spreadsheet workflow
Organize the worksheet into a raw record and a calculated record. The raw columns contain timestamp, pumping start, shutdown, discharge and measured head or depth to water. The calculated columns contain elapsed pumping time t_p, recovery time t′, total elapsed time t_p+t′, ratio R, log10(R), corrected head and residual drawdown s′. Each derived column should have a visible formula and unit description.
Do not substitute pretest head for a corrected baseline without documenting the difference. Do not round intermediate time ratios to integers when fitting a log relation. Keep raw timestamps at their actual precision and propagate measurement uncertainty appropriately. A plot should show the analysis window with excluded observations clearly distinguished rather than silently dropped from the data table.
After fitting, calculate the slope in metres per log cycle and recover T from Q. Reconstruct predicted residual drawdown at several observed ratios and compare it with the actual trace. Save residuals, model assumptions and a simple parameter sensitivity check. Such a worksheet is reproducible; a picture of a straight line without its source columns is not.
Choose fitting points for physical reasons
It is tempting to mark the two points that most closely match the expected slope and ignore the rest. A more defensible procedure begins by screening early shutdown effects, choosing a time window compatible with the theory, and checking whether the remaining residuals vary systematically with time or ratio. A formal fit can then quantify the slope and its sensitivity to reasonable window choices.
If the slope derived from the first few minutes differs sharply from a slope measured later, that is a diagnostic finding. It might reflect well storage, rate history, leakage, a boundary or an inaccurate baseline. It does not justify declaring whichever slope produces the desired transmissivity to be correct. Investigate the physical source of the discrepancy.
When the valid window spans less than one decade in R, regression can still estimate a per-decade slope if the data are suitable, but greater extrapolation magnifies vulnerability to noise. State the observed range of R and the number of points used. A high coefficient of determination over a tiny interval is not proof of an appropriate groundwater-flow model.
Use pumping recovery to investigate well losses
In a pumping well, active extraction creates local dynamic losses that may disappear quickly when flow ceases. Separating their decay from aquifer recovery can help explain why pumping-phase drawdown and recovery-based estimates differ. Yet early recovery is not a direct measurement of one isolated loss term. Residual water-column motion, casing storage and the timing of shutdown influence the signal.
Combine recovery records with a well-loss study when the purpose is well performance. The step-drawdown interpretation guide describes how rate-dependent additional losses are modelled across distinct discharge stages. That is a different analytical problem from a Theis slope obtained after stopping a constant-rate test.
If the pumped well returns rapidly while an observation well recovers much more gradually, consider the difference as potential evidence of local well effects, not as proof. Compare well construction, discharge history, screen arrangement and observation positions before drawing conclusions. The most reliable interpretation separates observations from hypotheses and uses independent tests to discriminate among them.
Why an equal slope does not prove equal geology
Two observations may yield similar recovery slopes because the sampled responses are compatible with a shared transmissivity model over one interval. That agreement is useful, but geological differences can remain hidden. Fracture connections, anisotropic pathways, variable thickness and local boundaries may produce compensating effects or emerge outside the chosen time window.
Conversely, different slopes need not mean one data set is wrong. Different observation intervals can sense different hydraulic domains or be influenced by other stresses. A report should compare fitting windows, instrument quality, baseline correction and pumping chronology before attributing differences to the formation. These checks matter particularly when wells are screened in different depths.
A qualified conceptual model should make predictions beyond the fitted number. Can the model reproduce time-dependent trends in several wells? Can it explain both drawdown and recovery? Is a proposed boundary supported by independent geology? A single transmissivity is one parameter in such an explanation, not a final map of the aquifer.
Four common misunderstandings
Misunderstanding one: recovery percentage is the same as residual drawdown. It is not; they measure complementary quantities relative to a chosen shutdown value and baseline. Repair: calculate both from raw depths and label them. Misunderstanding two: t/t′ means pumping duration divided by recovery duration. Repair: use total time since pumping began divided by time since stopping, with units matched.
Misunderstanding three: the late-time residual-drawdown line directly gives storativity. The simple Theis recovery slope yields T under its assumptions, while S cancels from the late-time difference. Repair: state what data and additional model would be needed to constrain storage. Misunderstanding four: a rising water level proves regional groundwater reserves were fully restored. Repair: distinguish hydraulic-head recovery from a regional water budget.
A fifth common error deserves equal attention: copying the same coefficient into a natural-log and a base-ten equation. Repair: check the 2.303 conversion. A sixth is fitting the pumped well’s earliest post-shutdown jump without considering loss disappearance or incomplete stop timing. Repair: examine the raw trace and select a physically justified interval. These repairs connect terminology directly to actual measurements and equations.
Transfer exercise: calculate a correct ratio
Exercise one: pumping ran for 2,400 seconds. A recovery measurement was taken 600 seconds after shutdown. Find R and its base-ten logarithm. Answer: t = 2,400 + 600 = 3,000 seconds, t′ = 600 seconds, R = 5, and log10(5) ≈ 0.6990. Using 2,400/600 = 4 would omit the recovery time from total elapsed time.
Exercise two: pumping ran for 30 minutes, and the instrument recorded water level 15 minutes after shutdown. Calculate R using minutes. Answer: total time since start is 45 minutes, so R = 45/15 = 3. Repeating the calculation in seconds gives 2,700/900 = 3. That agreement checks that the ratio is truly dimensionless.
Exercise three: after a long recovery time, a spreadsheet reports R = 0.25. What should be checked? Answer: whether t and t′ were reversed, whether the original pumping start was confused with shutdown, or whether the units disagree. For the stated positive pumping duration and positive recovery duration, (t_p+t′)/t′ must always exceed one.
Transfer exercise: check the slope and units
Exercise four: a valid semilog fit yields Δs′ = 0.30 metre per log decade, with prior constant discharge Q = 0.006 cubic metre per second. Compute T. Answer: T = 2.303 × 0.006/(4π × 0.30) ≈ 0.003665 square metre per second, or about 316.7 square metres per day. Write the units at each step; this is transmissivity, not hydraulic conductivity.
Exercise five: suppose the actual discharge was 0.003 cubic metre per second but the analyst mistakenly uses 0.006 with the same slope. What happens? Answer: computed transmissivity doubles. That is a rate-meter or unit-input error, not evidence that the aquifer changed. Exercise six: an analyst uses a slope of 0.30 in a natural-log formula as though it were 0.30 per log10 decade. Answer: the bases and slope definitions must be reconciled before calculation.
Exercise seven: original depth to water was 4.30 metres, depth at shutdown was 6.10 metres, and depth during recovery is now 5.20 metres. Answer: shutdown drawdown is 1.80 metres, residual drawdown is 0.90 metre, and the recovered portion is also 0.90 metre, or 50 percent. This numerical coincidence does not make residual and recovered amounts the same definition.
Independent check against the worked model
For the synthetic experiment with Q = 0.008 cubic metre per second and T = 0.003 square metre per second, the theoretical slope is about 0.4887 metre per log cycle. At R = 10, the residual is approximately 0.4887 metre. At R = 100, it is about 0.9774 metre. Their difference is about 0.4887 metre, exactly the intended one-decade slope to rounding precision.
Try the same at R = 5. log10(5) is approximately 0.6990, so the predicted residual is about 0.3416 metre. The pumping duration and elapsed recovery time must produce this ratio before the point belongs on the model graph. If the observation occurs at t′ = 900 seconds after a 3,600-second test, R is indeed five.
This independent substitution check can reveal a spreadsheet that used t_p/t′ or logged the wrong time column. It also detects a fitted slope reported with the wrong sign or log base. Arithmetic agreement is necessary but insufficient; an actual field record still requires hydrogeological model checks, equipment validation and an uncertainty assessment.
Limits of the method in real decisions
A carefully measured recovery record can corroborate a transmissivity interpretation and reveal problems in pump-rate bookkeeping, baseline correction, or the assumptions of a drawdown model. It can be especially valuable when monitoring continues after extraction ends. Nevertheless, the method does not by itself determine sustainable pumping yield, hydraulic connectivity everywhere in the formation, recharge rate, or long-term contaminant transport.
Even when the pretest water level is reached, the regional groundwater system may have exchanged water with distant boundaries, other aquifers or streams during pumping. Hydraulic-head restoration is not the same as reversing all mass transfers through the landscape. Interpret pressure and water-budget claims separately. Where environmental consequences matter, longer monitoring and a fuller groundwater model may be required.
Report a recovery-derived T as conditional on the model and observed interval. Provide the pretest and stop times, discharge history, screened geometry, measured residuals, fitting window, slope and units, supporting wells, and relevant limitations. A reader should be able to reconstruct the calculation without guessing the analyst’s notation or silently assuming an ideal aquifer.
Frequently asked questions
Does a rising water level mean transmissivity is high? Not automatically. Rate, well storage, aquifer properties, geometry and boundaries shape the response. Does recovery analysis require the pump to have operated at a constant rate? The simple one-term Theis ratio assumes a controlled constant prior rate and complete shutdown; variable-rate histories can require superposition. Is t the pumping duration? No, t is elapsed time since pumping began, including the recovery period.
What does residual drawdown mean? It is the head deficit still present relative to a defined baseline during recovery. Why does R approach one? As recovery time grows, the original pumping duration becomes small relative to the total elapsed time. Why does the graph sometimes run backward in the order points were measured? Because R decreases in time, though its axis can be drawn increasing from left to right.
Can I derive storativity directly from the slope? Not with the simple late-time Theis recovery relation alone. Can the first points be trusted? They require careful checking for pump coast-down, dynamic well loss and casing storage. Is recovery more reliable than pumping drawdown? It can offer a useful complementary estimate, but neither phase is automatically free from model or measurement errors.
How should I compare results from different reports? Check unit systems, test geometry, whether discharge was constant, baseline treatment, fitting intervals and model assumptions. Do not compare an unconfined delayed-yield interpretation and a confined Theis slope as though they estimated an identical physical system without qualification.
Reading technical language precisely
Terms such as recovery, residual drawdown, elapsed time, transmissivity, storativity, and hydraulic conductivity each carry a specific measurable meaning. Translating a technical groundwater report requires preserving those boundaries. Substituting ‘water has come back’ for a measured residual head deficit can turn a pressure observation into an unsupported claim about resource replenishment.
The eduKate guide to meaning and precision stresses keeping near-neighbour terms distinct. Its retrieval and transfer resource encourages independent use of new vocabulary in unfamiliar contexts. Both habits make a recovery calculation easier to communicate accurately across disciplines and languages: define the symbol, show its measurement, and retain its units and assumptions.
A translated summary is accurate only when its reader can reconstruct which moment began each clock, how residual head was calculated and which logarithm the slope used. The best technical explanation is not always the shortest. It is the one that preserves enough of the measurement chain for another person to identify a mistake before making a consequential decision.
Final audit: what the curve really supports
Begin with the recorded facts: pumping occurred at a documented discharge, stopped at a particular time, and hydraulic head changed during recovery. Calculate residual drawdown from a defensible baseline. Calculate t_p, t′, t_p+t′ and R, then plot residual against log10(R). Check whether an interval is physically compatible with the Theis recovery approximation. Only then estimate a transmissivity from its semilog slope.
Use the synthetic one-decade result to test the arithmetic: Q = 0.008 cubic metre per second, Δs′ ≈ 0.4887 metre, and T ≈ 0.003 square metre per second, or 259.2 square metres per day. The time ratio is dimensionless, the slope is length per log cycle, and T is area per time. Check every label before discussing aquifer geology.
Finally ask what the model omits: variable pump rate, imperfect shutdown, local well loss, leakage, unconfined drainage, heterogeneity, boundaries and baseline drift. A scientifically useful recovery article carries its conclusions through these checks instead of equating a straight line with certainty. The transferable skill is to move from measured response to an explicit conditional interpretation, without letting terminology, attractive graphics or convenient assumptions replace evidence.
Direct primary sources and contextual links
USGS: Groundwater test analysis and aquifer-property estimates in a published research report.
USGS: Completion Summary TAN-2336, aquifer-test interpretation and recovery observation.
USGS: Aquifer-test methods and recovery interpretation, Scientific Investigations Report 2012–5082.
USGS Open-File Report 1985–200: Theis recovery slope equation and definition of residual drawdown.