Two wells can be the same distance away and tell different stories.
A pumping test can lower water levels by different amounts at two observation points the same distance from the pump. If the points are open at different depths, that disagreement may be the most useful part of the test. Learn how to read it before fitting a single curve to the whole aquifer.
Start with a drinking straw in a sketch of a saturated layer. We will turn that sketch into a pair of readings at different depths, then use a second set of readings to ask whether a change comes from well geometry, a boundary, or something we have not measured. You need no equipment or access to a well. The worked data are fictional and are not instructions for field testing.
Hydrology subject guide · Learning that transfers into life · How X Works hub
Choose your reading route
Route 1 · Chapters 1–3: Picture the flow
Draw the openings before drawing the curve
Learn what the screen is, what drawdown measures and why depth can matter.
Route 2 · Chapters 4–6: Interpret the test
Worked example: compare equal distances at different depths
Compare a pair of observation screens, check a model and keep well loss separate.
Route 3 · Chapters 7–9: Plan the next evidence check
Check the depth contrast before interpreting it
Control the references, decide what to report and transfer the method to a different aquifer.
Full chapter index · Independent worked case · Questions and practical checks
Chapter index
Picture the flow
Draw the openings before drawing the curve
Distinguish head, water level and drawdown
Predict what changes near a short intake
CHAPTER 1 OF 9 · Picture the flow
1. Draw the openings before drawing the curve
An aquifer is a body of saturated material that can transmit groundwater. In a simplified cross-section, draw a 24-metre-thick layer. Make the top of the layer your starting line and mark its bottom. Now draw a well going nearly to the bottom. Erase a short section of the well wall in your picture between 15 and 21 metres below the top. The erased interval is the screen: the part where water can enter the well from the surrounding formation in this simple example.
The bottom of the borehole and the bottom of the screen are different things. A deep hole with a short open interval still draws water through that interval. The rest of the casing may be closed to flow, or the completion may have features that complicate the picture. Check the actual construction record, including gravel pack and seals where relevant. “A 25-metre well” does not tell you which parts of the aquifer communicate with it.
A fully penetrating pumping well is open through the full relevant thickness of the aquifer in the idealised comparison. A partially penetrating well is not. In the first sketch, the pump takes water through the lower part of the layer. Water arriving from above the screen must have some vertical component of flow near the well. It cannot all come straight in horizontally at its starting elevation. Draw two plausible paths, one starting alongside the screen and one higher up. That extra geometry is the point of the model.
Real aquifers are not always clean blocks of material. They can have layers, fractures, uncertain boundaries or a water table that moves appreciably during testing. The sketch still teaches the first question: where can water move into the pumping well, and where did we observe the response? If you cannot answer either part from the records, treat the choice of model as provisional.
CHAPTER 2 OF 9 · Picture the flow
2. Distinguish head, water level and drawdown
Hydraulic head is a measure of groundwater potential expressed as a length. In ordinary freshwater head comparisons it combines elevation and pressure head. A suitably installed short-screen observation well can indicate head near the interval it samples. The water surface inside that well does not mean there is a free water surface at the same depth in the formation outside the casing. For example, water may rise above a confined aquifer in a well that is open to it.
Drawdown is the decrease in head relative to a defined starting condition. We will use positive numbers for a fall in head. If a point starts at 12.60 metres of head in the common datum and reaches 12.25 metres during pumping, the drawdown is 0.35 metres. When you compare it with a second screen, calculate that screen’s drawdown from its own appropriate starting head. The starting heads need not be equal. Treat a drifting baseline as an evidence problem, not as a reason to force them equal.
At two nearby points, unequal head at the same time can indicate a hydraulic gradient, provided the elevations, density assumptions and measured intervals are comparable. Unequal drawdown tells you how their heads changed differently. Those are not interchangeable. If deeper head was higher at the start, the deeper point might drop more and still finish higher. To infer the final direction of vertical flow between connected points, compare final heads, not just their changes.
Likewise, a long screen is not automatically a tidy average of nearby heads. Connections along the open interval, conductivity differences and flow within the well can affect the measurement. A single level can hide opposing behaviour at different depths. If the decision depends on that variation, use observations that can resolve depth, not a more precise reading of the same long-screen level.
CHAPTER 3 OF 9 · Picture the flow
3. Predict what changes near a short intake
Suppose the screen sits low in the layer and groundwater can move through the formation horizontally and vertically. Near the well, water moves towards the intake from the sides and from higher parts of the layer. A deep observation point near the intake may respond sooner or more strongly than a shallow point at the same horizontal distance. That is a hypothesis to test, not a law for every construction and aquifer.
The rate at which differences between depths smooth out depends on the material and the geometry. If the aquifer conducts water more readily horizontally than vertically, pressure changes can spread sideways without becoming uniform through the thickness immediately. A tight layer can make the difference stronger or give the system a different structure from a single anisotropic layer. The word anisotropy describes a dependence on direction; it does not establish that the aquifer is made of one material.
Farther from the well, vertical differences caused by the local intake can become less important to some comparisons. Do not impose a universal distance or wait time. Use a model appropriate to the aquifer and check the part of the record you intend to use. For an unconfined test, free-surface effects and delayed drainage may also influence the curve. A change of slope alone does not tell you which process is responsible.
This gives a useful order of thought. First put screens and boundaries on a cross-section. Next predict what an intake effect would do near and far, and whether those observations are within reach of the instruments. Then compare alternative explanations. Starting with a curve shape can be useful, but only if you return to a physical picture before naming the cause.
CHAPTER 4 OF 9 · Interpret the test
4. Worked example: compare equal distances at different depths
In a fictional confined freshwater layer, the pump takes water through a short screen low in the formation. At 12 metres horizontal distance there are two separately completed observation points: D near the intake depth and S near the top of the same layer. Both start at 19.80 metres of head. At 20 minutes D reads 19.25 metres and S reads 19.65 metres. Their drawdowns are 0.55 and 0.15 metres. Equal radius did not produce equal response.
At 120 minutes D reads 18.88 metres and S reads 19.07 metres. The drawdowns are now 0.92 and 0.73 metres. The depth contrast fell from 0.40 to 0.19 metres while both points continued to decline. A second pair 48 metres away reports 0.26 and 0.24 metres of drawdown at 120 minutes. Suppose the uncertainty in each head change there is a few centimetres. That second depth contrast is not clearly resolved. Do not round the numbers to make all the pairs look equally informative.
These numbers are compatible with a stronger depth dependence near the short intake and a smaller contrast farther away. They do not establish that the formation is homogeneous or that distance alone caused the change. The far pair could be in different material, the shallow screen could be delayed by a faulty installation, or the pump rate could have changed. Look at observations and construction records before turning compatible into confirmed.
The initial heads in the example were deliberately equal. Change the starting deeper head to 20.40 metres while keeping its 0.55-metre drawdown. The deeper head at 20 minutes becomes 19.85 metres, still higher than the shallow 19.65. Even with stronger deep drawdown, the head ordering at that time is reversed compared with the original example. That is why a plot of drawdown alone cannot answer every question about vertical flow direction.
CHAPTER 5 OF 9 · Interpret the test
5. Choose a model that can represent the evidence
A simple horizontal radial model can be a sensible starting point if the conditions and observations support it. It treats the response at a given radius in a way that does not distinguish screen depth. In the fictional near pair, such a model cannot explain the two simultaneous readings at the same radius under those assumptions. Fitting a separate horizontal model to each point and naming the two fitted values as different aquifer properties skips a basic model check.
Instead, compare a model that represents the pumped and observation intervals, and examine sensitivity to vertical and horizontal hydraulic conductivity, thickness, and appropriate storage behaviour. Do not estimate everything merely because the software allows it. A smaller vertical conductivity, different screen geometry and a low-conductivity layer can each change the predicted depth contrast. If several explanations fit, report what combination of data and assumptions limits the result.
Test predictions the fitting did not use. For example, fit a subset of the near and intermediate observations, then compare the predicted far pair or a separately measured recovery period. Keep the pump-rate history in the calculation rather than assuming it was constant if the log says otherwise. A poor prediction at a new depth may matter more than a small improvement in residuals at the depth used for fitting.
The USGS Cape Cod study is a useful reminder that geometry can matter without explaining everything. It analysed a partially penetrating pumping well in an unconfined outwash aquifer and found that accounting for gradual drainage helped match drawdown, especially at observation points near the water table. The study’s fitted properties belong to that site. The transferable lesson is to compare predictions at multiple depths and consider whether the water-table response needs its own representation.
CHAPTER 6 OF 9 · Interpret the test
6. Do not assign all extra drawdown to the aquifer or all of it to the well
A level measured in the pumping well can include effects local to the well as well as drawdown in the surrounding formation. A constricted intake, disturbed material around the well, screen losses and wellbore storage can affect the response in different ways and at different times. Partial penetration can also increase drawdown near the pumping well because flow converges toward a shorter open interval. These effects are not the same explanation simply because they occur close to the pump.
Imagine that the pumping well level falls 2.2 metres while the nearest external short-screen observation shows 0.7 metres of drawdown. It is tempting to subtract the readings and call the 1.5 metres “well loss.” The observation well is at a different position and may be open at a different depth. The difference contains spatial variation in the formation as well as whatever local well effects are present. You need an estimate of the formation head near the well and suitable evidence for the local loss before separating them.
A change in the relationship between discharge and pumping-well drawdown in a staged test can help investigate well behaviour, but it is not an automatic correction for partial penetration. Check which effects the test and model can distinguish. Compare independent observations outside the pumped well where they exist. A single water-level recorder in the pumped casing may be excellent for operational monitoring yet inadequate for estimating both aquifer properties and local losses.
For a practical check, write the label on each series in full: “Pumped well, lower screen, water level during pumping,” or “Observation D, short lower screen, 12 metres from pump, head relative to stated datum.” That is more than record-keeping. It prevents someone from subtracting two columns as though they were observations of the same location and process.
CHAPTER 7 OF 9 · Plan the next evidence check
7. Check the depth contrast before interpreting it
Verify that both instruments refer to known elevations in the same vertical datum, and check the conversion from observed pressure or depth to head. Compare clock alignment and pumping-rate changes. A time offset at the start of a test can look like a different arrival time at depth. A wrong measuring-point elevation can produce a persistent head offset, while a moved sensor or changed correction can introduce a step. Each has a different signature.
Next inspect the wells’ completion and condition. A poorly sealed annulus or an open borehole can connect zones that you intended to compare independently. A short screen may be slow to respond if the installation is impaired. A common borehole with several monitoring intervals can be useful if the intervals are properly isolated; a poorly isolated installation can disturb the very head contrast it was meant to observe.
Look for non-test signals too. A nearby pump, rainfall or changes in barometric pressure can affect readings, depending on the setting and instruments. Where available, use pre-test observations or a comparison point outside the response of interest. State the evidence used to distinguish a site-wide head change from the imposed pumping signal. A baseline correction should not remove an inconvenient part of the test without a defensible reason.
Before fitting, give a second reader the screen log, baseline and rate history. Ask them to tell you which pair can test a vertical contrast and which period can test a stable pumping assumption. If they cannot locate a screen relative to the aquifer, mark it unresolved on the cross-section. Guessing the interval is worse than carrying a gap forward because it can make an incompatible model appear to fit.
CHAPTER 8 OF 9 · Plan the next evidence check
8. Report what the test resolved and what it did not
A useful interpretation says how far the inference extends. For the fictional near pair, you can report a resolved early depth contrast under the stated head checks and a later reduction in that contrast. You can say the pattern is consistent with depth-dependent response near a partially penetrating well. You cannot report a unique vertical hydraulic conductivity from those snapshots or infer uniform flow at all radii because the far pair agreed within uncertainty.
If you use a simpler horizontal model for a subset, explain how you selected the subset and what conclusions you exclude. For example, “This estimate is conditional on the late-time observations in which no depth contrast was resolved at the far pair; it is not a prediction of near-screen head loss.” Say what would invalidate the simplification. A good report distinguishes an assumption that was tested from an assumption that is merely convenient.
If no suitable depth pair exists, it is often still possible to give a bounded or conditional result. Show how different reasonable completion assumptions affect the interpretation. If the decision is sensitive to those alternatives, ask for the information that actually separates them. A more complex fit may produce additional digits but cannot create a missing depth observation.
Attach the record a reviewer would need: screen top and bottom, aquifer top and bottom where known, well locations, measuring-point elevations, time basis, raw readings and corrections, discharge history, and the model and assumptions. Put the claimed precision in the context of measurement quality and model uncertainty. When passing the file to another team, include enough detail that they do not mistake a calculated head for an uncorrected depth below casing.
CHAPTER 9 OF 9 · Plan the next evidence check
9. Transfer: a shallow intake in an unconfined aquifer
Now move the pumping screen high in an unconfined aquifer instead of low in a confined one. At two equally distant observation points the shallow response is strong at first and the deeper response is smaller. Does the earlier method transfer? Yes: map the open intervals, compare heads and drawdowns separately, and test the response by depth and distance. The conclusion that the deeper point should respond more strongly does not transfer; that belonged to the original geometry.
The free water table changes the problem. Drawdown can change saturated thickness, and drainage near the water table can affect the time response. If the upper part of a screen becomes exposed during the test, the effective intake interval can change. These are reasons to check whether a fixed-thickness or fixed-geometry model still represents the relevant interval. The shallow case is not just the earlier drawing turned upside down.
Imagine the deeper point now shows a new late-time flattening. One explanation might involve delayed drainage; another might involve an outside hydraulic boundary or a change in pumping rate. Ask what each explanation predicts at the shallow point, in the far pair and during recovery. Then check the actual rate record. Do not label the curve by its shape before seeing whether the pump supplied the assumed stress.
The transfer test is practical: explain which observation you would add if the pumping decision depended on predicting a shallow water-table decline near a boundary. A good answer specifies a depth, location and time period, names the uncertainty it would reduce, and says which outcome would challenge the current interpretation. “Take more readings” does not yet tell the next team what to do.
Independent worked case and evidence checks
A different fictional test has two short-screen observations at equal radius. The shallower point has an initial head of 10.20 metres and the deeper point starts at 10.00 metres. After one hour they both show 0.30 metres of drawdown. Does this prove there is no head gradient between the screens? No. Their heads are now 9.90 and 9.70 metres. Their initial difference of 0.20 metres remains, even though the drawdowns match.
The deep screen is centred 8 metres below the shallow one. If the points are close enough laterally for this exercise and are in connected freshwater material, the magnitude of the vertical head difference divided by their separation is 0.20 ÷ 8 = 0.025. A higher head at the shallower point is consistent with a downward gradient between the points. It does not tell you the flow rate without information about the formation and the flow path.
Now add a correction: the deep measuring-point elevation was entered 0.20 metres too low. If that was a constant error across the test, correcting it increases both deep heads by 0.20 metres but leaves its drawdown unchanged. The corrected heads match in this toy comparison. This is why the datum belongs on the worksheet even when you think you are only looking at differences.
As an evidence check, ask a partner to calculate the same examples from the raw heads and then state which conclusion survives the datum correction. Matching drawdown survives. The inferred vertical gradient does not. You have found which claim depended on an extra assumption and which one did not.
Try a second model check. The near pair differs by 0.40 metres at 20 minutes, and a hypothetical prediction made before the test gives a 0.35-metre contrast at that time. The same prediction expects a 0.25-metre contrast at the far pair at 120 minutes, where the observed contrast is only 0.02 metres. If the difference is larger than the relevant measurement and prediction uncertainty, the near match is not enough. Check whether the model assumes the wrong conductivity contrast, screen position or aquifer geometry. The far pair has done useful work even though its two readings were nearly the same.
Worked comparison
| Observation | What it can support | Check before generalising |
|---|---|---|
| Near pair differs by depth | Depth-dependent response at those intervals | Screen construction, timing, datums and connectivity |
| Far pair agrees within uncertainty | No resolved depth contrast there during the comparison | Sensitivity and conditions at other radii or times |
| Pumped well falls more than a nearby monitor | The two measurements describe different places or processes | Model the formation response before attributing the remainder |
| Head difference remains while drawdowns match | A pre-existing head contrast persists | Common datum, density and hydraulic connection |
Questions and practical checks
Does equal distance imply equal drawdown?
Only under a model and conditions that make depth and other differences unimportant. Check that assumption against the observations.
Is every short-screen well a partially penetrating pumping well?
No. A short-screen observation point and a pumping well have different roles. Record the open interval of each relative to the aquifer thickness.
Does a long screen measure average head?
It may integrate responses from several intervals in a way influenced by hydraulic connections and flow. Do not assume a simple arithmetic average.
Can I ignore partial penetration at late time?
Sometimes a simpler model is defensible for selected observations. The useful distance and time depend on the setting, model and sensitivity needed.
Can a depth contrast establish vertical conductivity?
Not by itself. Screen geometry, layering, storage, pump history and other properties may also influence it. Use multiple observations and report alternatives.
Should students try pumping a real well?
No. Use the fictional data and cross-section. Real groundwater work requires suitable permissions, equipment and qualified oversight.
Continue with the connected subjects
Connect head changes to pumping and storage
Pumping-induced upconing: when the vertical geometry matters for water quality
Aquifer storage and recovery: distinguish stored water from what can be recovered
Water-table fluctuations and why recovery depends on the setting
Check measurement and alternative explanations
Pressure-based water levels: check density, elevation and the sensor record
Observation-well drawdown derivatives and boundary diagnosis
Step-drawdown tests and what well-loss estimates mean
Sources and scope
Shlomo Neuman, Water Resources Research: partial penetration and delayed gravity response
The numerical cases in this guide are original fictional teaching examples, not measurements from a real site. This is an educational explanation, not a field safety instruction, engineering approval or an official certification. For a real project, use qualified practitioners, current product documentation and site-specific evidence. Sources were checked on 5 November 2026.