Pumping Test Data Screening Worksheet

Pumping Test DataScreening Worksheet

Check the record before interpreting the aquifer, selecting the pump or issuing an operating instruction.

1. Engineering answer: a readable test sheet is not necessarily an interpretable test

A pumping test data screening worksheet checks whether water-level observations, discharge measurements, timestamps, pumping events and recovery records have a clear, consistent meaning before they are used in an analysis. It identifies the usable records, the corrections that have evidence, the unresolved intervals and the conclusions that must wait. It does not calculate an aquifer property merely because a file contains columns called time, level and flow.

The central question is specific: “Are these observations suitable for this stated calculation over this stated interval?” A dataset can support a discharge-volume check while remaining unsuitable for an early-time drawdown analysis. A valid low water level is not bad data simply because it is undesirable. A smooth curve is not good evidence if missing readings were silently replaced.

Its result is a screened dataset and a visible exception trail, not an automatic declaration of sustainable yield, groundwater availability or permission to operate.

2. What this page owns—and what it leaves to the existing pages

A suitable application accepts field sheets, logger exports, flow records and event logs, then returns records with retained identities, normalized quantities, qualification codes and explicit analysis eligibility. It should not conceal the original observations behind an apparently improved graph.

The workflow is relevant to pumped borewell tests. Drilling returns, airlift development and compressor-assisted cleaning must retain their actual event classification. Their observations should not be relabelled as a controlled pumping test merely to populate missing fields. A hydrogeologist or other appropriately qualified reviewer decides whether a particular investigation supports the intended interpretation.

3. Establish the well, test and measurement geometry

Start with a stable well ID and a separate test ID. Record whether each water-level series belongs to the pumping well, an observation well or an independent reference location. A pump serial number cannot stand in for well identity, particularly when rental equipment moves between installations. Capture the pump configuration, intake depth, measuring point, discharge location and applicable construction record.

A coordinate-derived separation needs a defined coordinate reference and appropriate accuracy. Screened or open intervals should be taken from construction evidence, not inferred from the depth of a sensor. Mark unavailable information as a specific limitation on the intended method.

USGS aquifer-test documentation guidance identifies construction information, site context, complete time-discharge measurements, pre-test water levels, pumping and recovery records, and documented adjustments as important components of a reproducible test package. This is supporting technical context, not an adoption of USGS administrative procedures for Reddy customers. [R1]

If two files describe different wells or configurations, keep them separate until an evidence-based relationship is established. A plausible overlay of their curves does not prove they belong together.

4. Choose the intended use before applying a pass label

Record the intended output before screening begins. Otherwise, an easy-to-pass arithmetic check can be mistaken for approval of a much more demanding interpretation. A discharge summary needs a valid flow-time boundary. Drawdown needs a compatible water-level reference. Recovery interpretation also needs a trustworthy pumping history and stop event.

Keep the analysis name and interval beside every disposition. “Usable for volume between 10:00 and 11:00” is informative. “Test passed” is not. Dataset suitability can change when the user changes the analytical method even though the underlying observations remain unchanged.

5. Preserve a compact but sufficient data dictionary

Keep identity, quantity, unit and record meaning explicit. A wide table containing both levels and flows must state whether they were measured simultaneously or joined later. Blank fields should retain a reason code instead of being forced into numerical zero.

A comment such as “normal readings” cannot replace these definitions. The dictionary should be short enough to use at import and precise enough that a second analyst can reproduce a calculation without asking the original technician what a heading meant.

6. Reconcile calendar time, elapsed time and clock offsets

Preserve calendar timestamps with dates and an explicit timezone. Retain the recorded elapsed time separately, including its unit and origin. For a chosen event origin t0, calculate elapsed minutes from complete timestamps; do not subtract clock-face strings that lose the date at midnight.

Elapsed minutes = (observation timestamp − event-origin timestamp) in seconds / 60.

Compare this result with the supplied elapsed-time column. A mismatch may reflect transcription, a different origin, a unit error or clock drift. It is a question requiring evidence—not permission to select whichever value makes the curve look better. An observed time of 00:05 after a 23:55 start on the previous date represents ten minutes, not a negative interval.

A constant clock offset shifts a series consistently; drift changes its alignment progressively. Store the correction model and evidence separately. One documented synchronization check can support an offset at that time, but does not establish a drift rate for the entire test. Where pump events and logger readings have different clock uncertainties, assess their relative timing before using early response data.

Maintain original row order for traceability and a separate chronological analysis view. Out-of-order exports are not necessarily out-of-order measurements. Sorting must not remove duplicate conflicts or erase an instrument restart.

7. Segment the test by actual operating events

Identify pre-test monitoring, each pumping stage, pauses, shutdown, recovery and any restart. Use start and end event records rather than filling the entire interval with the label printed in the report title. A test planned as constant-rate can contain actual interruptions; the plan does not override the field record.

The proposed interval convention is start-inclusive and end-exclusive: an interval owns its starting instant but not the start of the next interval. A boundary water-level observation may retain an explicit before-stop or after-stop qualifier. This avoids assigning one discharge interval to both pumping and recovery while preserving instantaneous observations at the boundary.

For step tests, preserve step number, elapsed time since initial pumping, and elapsed time since the current step began. Restarting a local clock does not remove the hydraulic history of earlier stages. For recovery after interrupted or variable-rate pumping, retain the complete sequence for method selection. The worksheet should not collapse that history into a convenient single pumping duration.

8. Normalize units without changing the underlying quantity

Convert values only after confirming what they represent. Litres per second, litres per minute and cubic metres per hour are different rate units. A totaliser reading is a volume, not a flow rate. A water-level depth is a length, not a pressure reading that happens to carry a similar number.

Useful exact unit relationships are: 1 L/s = 3.6 m³/h; 1 L/min = 0.06 m³/h; 1 ft = 0.3048 m; 1 h = 60 min. These conversions do not resolve an unknown unit. A field marked “gpm” also needs the gallon convention before conversion.

Preserve original_value and original_unit beside the normalized field or a stable source link. Record the conversion-rule version and affected records. Do not infer a missing flow unit from motor power, hose diameter or the apparent smoothness of the result. Numerical formatting must also distinguish decimal separators from thousands separators.

Treat sentinel values such as −9999 dictionary, not as a universal rule. A negative depth or a negative drawdown can have physical meaning under a declared convention; a blanket rule deleting every negative value can discard valid observations.

9. Reconcile measuring points before calculating drawdown

For depth below a common fixed measuring point, increasing depth means that the water surface has fallen. Let d0 be the reference depth and d(t) the observed depth on that same basis. Then s(t) = d(t) − d0, where positive s denotes drawdown. For water-level elevations h0 and h(t) on a common datum, the equivalent expression is s(t) = h0 − h(t).

Where the measuring point changes, derive the transformation from geometry. If z is measuring-point elevation and d is depth below that point, water-level elevation is h = z − d. Therefore:

d_common = d_observed + z_common − z_observed.

All terms must use compatible length units and the same vertical datum. This equation is a coordinate conversion, not an adjustment for a real groundwater trend. A constant reference shift applied consistently can leave differences unchanged; inconsistent treatment of pre-test and pumping readings can create false drawdown.

Record which reference is used and its effective dates. Do not attach a precise absolute elevation to an unverified local datum. When reference geometry is unavailable, retain the raw depths but hold the affected comparison. The worked example in Section 20 shows why a modest casing-reference change can produce a material percentage error in a small drawdown.

10. Check instruments without treating instrument corrections as aquifer corrections

Review range, installation depth, scaling, verification observations and configuration changes. A flat line at a sensor limit is not evidence that water level stabilized. A sensor moved during the test needs a dated installation record. Compare logger readings with suitable independent measurements at matched times and references where available.

An absolute pressure transducer used for water-level measurement requires atmospheric-pressure compensation; a functioning vented gauge transducer already references atmospheric pressure. Applying the same subtraction indiscriminately can introduce an error. Removing atmospheric pressure from a sensor measurement is different from modelling the aquifer's physical response to barometric changes. These two operations must not share one unexplained “barometric correction” field. [R2]

This worksheet does not supply a universal sensor correction. Require the instrument's method, raw quantities, timing and review evidence. An unresolved instrument problem may limit only a particular interval or channel. Keep valid manual observations visible rather than excluding the entire investigation automatically, but do not claim they reconstruct the missing logger response between observations.

11. Reconcile discharge records with their measurement meaning

For a measured interval-average rate Q̄i in m³/h over Δti hours, interval volume is Vi = Q̄i × Δti. Add non-overlapping accepted intervals to obtain the volume represented by those records. A documented constant rate over an interval can use the same equation, but its status must distinguish a maintained or assumed rate from an interval-average measurement.

For instantaneous rates at successive timestamps, trapezoidal integration gives Vi = (Qi + Qi+1) × Δti / 2 only under the chosen assumption of linear variation between those observations. It is not appropriate across an undocumented pump stop or a rate change known to be abrupt. Show the integration method in the result, not just in hidden software settings.

For a totaliser, use V = ending total − starting total, after resolving units, meter reset, rollover and the measurement boundary. A negative difference is a reset/conflict flag, not a negative pumping volume to be averaged away. Meter bypasses or recirculation can also separate pumped discharge from water delivered to storage.

Compare independent volume estimates only over the same interval and boundary. A mismatch is a reconciliation issue until its cause is established. It is not automatically a leak, a faulty pump or an aquifer deficit.

12. Screen rate stability rather than assuming that an average is constant

A constant-rate interpretation requires attention to the actual rate history. Preserve all available readings and transitions. A mean of 10 m³/h does not make a sequence alternating between zero and 20 m³/h equivalent to continuous pumping at 10 m³/h.

For a supported set of intervals, calculate Q̄ = Σ(Q̄i × Δti) / ΣΔti. This is a time-weighted mean, not a row-count mean. A screening statistic can express maximum deviation from that mean: Dmax = 100 × max|Qi − Q̄| / Q̄, for Q̄ > 0. Identify whether Qi are interval means or instantaneous measurements; the latter may reveal excursions hidden by averaging.

Use a project-selected review threshold and record its basis. Passing the selected screen does not establish that every hydrogeological assumption is satisfied. Failing it does not necessarily invalidate the measurements: the appropriate response may be segmentation or a variable-rate analytical method.

A simple specific-capacity ratio, Q/s, is descriptive only at a documented rate, duration and level basis. It is undefined for s = 0 in this screen and does not become transmissivity or sustainable yield through relabelling. Retain the observation conditions wherever that ratio is transferred.

13. Preserve the pre-test reference and background behaviour

Identify the observations used to establish the reference water level and whether pumping or other disturbances had already occurred. The final pre-start reading is not automatically an undisturbed static level. Record nearby pumping, rainfall observations, site activity and known background trends where relevant to the test record.

Separate the original reference from an adjusted analytical reference. A reviewer may need to account for a background trend, but the method, interval and supporting observations must remain visible. Do not choose a different reference merely to increase apparent recovery or reduce apparent drawdown.

When the baseline itself is uncertain, the report should state which derived values depend on it. A flow-volume result may remain usable while drawdown relative to that baseline needs qualification. A reference-well record can support background assessment only when its relationship to the test is appropriate; nearby does not automatically mean unaffected.

The purpose of screening is to preserve these alternatives for a competent interpretation, not to force every dataset into one textbook curve shape.

14. Keep missing, censored, estimated and zero observations distinct

Missing data mean that an observation is unavailable; they do not mean the value was zero. A stopped pump may have a confirmed zero discharge, while a communications failure may merely report a stale zero. Water level below a measuring instrument's accessible range is a qualified limit, not an exact reading at that limit.

Record separately: not collected; lost record; sensor out of range; suspect reading; corrected measurement; and estimated value. Avoid replacing these meanings with one blank or one colour. An observation can be displayed for context while excluded from a specific calculation.

For scheduled observations, slot completeness is 100 × accepted matched slots / expected slots under an explicit matching tolerance. For interval-average measurements or an approved time-support model, report supported time divided by requested time separately. Row completeness and time coverage answer different questions and must not share one unqualified “coverage” label.

Do not interpolate across a pump start, stop, restart or known step change by default. Long gaps remain visible. Where an estimate is justified for a limited purpose, retain it in a separate view and identify its effect on the result. A complete-looking estimated series must not be exported as the original measured test.

15. Screen recovery with the correct reference and two time origins

Recovery records need the actual stop event and prior pumping history. Define τ as elapsed time since pumping stopped. For a single continuous pumping period of duration tp, total time since pumping began is t = tp + τ. A recovery coordinate used by some methods is t/τ = (tp + τ)/τ, with both times expressed in the same unit and τ > 0. Creating this coordinate does not establish that the method is suitable.

Using depth below a common point, residual drawdown is s′(τ) = d_recovery(τ) − d0. A descriptive recovery percentage can be 100 × (s_end − s′)/s_end, only when a valid positive s_end exists and the end-of-pumping observation and reference are compatible. If s_end is missing, return “not calculated”; do not substitute the largest recorded depth without review.

At τ = 0, a logarithm of the recovery coordinate is undefined. Keep that observation in the linear-time record rather than changing zero to an arbitrary small time. A restart ends the simple uninterrupted-recovery segment; later points retain the new phase and full history.

A recovery percentage above 100% or a negative residual is not automatically deleted. Investigate background changes, datum, timing and instrument behaviour. Recovery approaching the reference over one observation period does not alone prove a repeatable daily abstraction schedule.

16. Use diagnostic views to reveal questions, not manufacture an ideal response

Review water level and discharge against chronological time with event markers. Add elapsed-time and recovery views only where their prerequisites exist. Mark excluded and suspect observations distinctly and retain access to them. A smooth line joining widely separated points can suggest evidence that was never collected; visible gaps are more honest.

Screen abrupt changes, frozen values, impossible unit transitions, conflicting duplicates and observations beyond recorded instrument limits. Compare each flag against the event log. An abrupt water-level change may be a real restart response, not a numerical outlier. A different manual reading may reflect a different measuring point rather than sensor drift.

Inspect each observation well separately before aggregating. Do not average pumping-well drawdown with an observation-well series. Early-time data, late-time data and recovery can have different usefulness for a proposed method.

No protection setting should change as a consequence of a screen. This is an evidence-review resource, not a controller. Operational concerns and unsafe conditions remain matters for the site's authorized personnel.

17. Separate review status, data origin and analysis eligibility

A corrected value can be accepted for one use and excluded from another. An estimated value can be useful for planning without becoming a measured observation. The proposed model therefore stores three separate attributes: data origin, review status and eligibility for a named analysis. A single “good/bad” column is insufficient.

Flag severity should reflect the consequence for the requested output. An unresolved test start blocks elapsed-time interpretation but may not block a totaliser difference over separately timestamped endpoints. An unknown level datum can block cross-well head comparison while allowing an internally consistent within-series difference under a stable local reference.

18. Make every correction reproducible and reversible

Each correction needs an ID, affected records, before-and-after values or a reproducible transformation, units, reason, evidence, author and review status. Preserve both the source dataset revision and the corrected dataset revision. Avoid a manually edited “clean final” file with no explanation of what changed.

Correct the demonstrated problem only. A documented unit conversion is different from smoothing noisy observations; a verified measuring-point offset is different from removing a real baseline trend. Record uncertainties that remain after the transformation. If the correction changes time alignment, recalculate affected elapsed times and integrations rather than retaining old derived columns.

A small correction in a total volume may be decisive when the affected point determines a specific-capacity comparison or the start of recovery. A result that remains unchanged still needs a documented assessment when it relied on the corrected field.

Retain rejected correction proposals as review records where they explain an issued decision. No source file should be overwritten solely to make an exported chart appear cleaner. The analyst must be able to reproduce both the former result and the revised result from their respective recorded inputs.

19. Issue a purpose-specific disposition, not a universal quality score

The proposed outcomes are ready for the stated screening use, usable with stated limitations, hold pending evidence, and not suitable for the stated method. None means that the well is safe, licensed or capable of delivering a particular long-term yield.

Evaluate prerequisite failures before counting accepted rows. Ten thousand ordinary readings do not compensate for an unknown stop time in a recovery calculation that depends on it. Likewise, a partial early-time record may still support a different late-time interpretation if a qualified reviewer establishes that applicability. Keep the excluded interval and reason with the retained series.

A useful disposition names the dataset revision, well, test, method candidate, accepted intervals, unresolved issues and next action. It should say whether further evidence is being recovered, a correction is awaiting review or a different analytical method is needed. The review must remain reproducible after files are renamed or moved.

Example wording: “Discharge-volume arithmetic is supported for the listed intervals. The simple continuous-pumping recovery view is on hold because an interruption is recorded and the pumping history is incomplete.” That preserves usable information without granting unsupported analytical approval.

20. Worked case A: midnight rollover and a changed measuring point

Synthetic example; not a customer pumping test. Pumping starts at 2026-10-01 23:55:00 +05:30. A reading at 2026-10-02 00:05:00 +05:30 has ten elapsed minutes. The recorded elapsed value of 600 is consistent only if its unit is seconds. Treating it as minutes would misplace the observation by a factor of sixty.

The original measuring-point elevation is 100.20 m on a local reference. The pre-test depth is 20.00 m, giving water-level elevation 80.20 m. A later documented measuring point is 100.50 m. Its pumping depth is 32.30 m.

The disposition is not “subtract 0.30 m from every future record.” It is “apply the documented reference conversion to the affected interval, retain the originals and update the dependent results.” With no evidence for the new measuring-point elevation, the conversion would remain a proposal and the affected drawdown would be held.

21. Worked case B: unequal intervals, rate stability and a missing period

Synthetic example. A pumping record contains the following interval-average discharge measurements. The interval from minute 25 to 30 has no usable flow record. Its duration is known, but its discharge is not.

Accepted represented volume is 9.6667 m³ over 55 minutes. Supported-time coverage is 55/60 × 100 = 91.67%. The weighted mean over those accepted intervals is 9.6667/(55/60) = 10.5455 m³/h. The unweighted mean of the three rates, 10.6667 m³/h, answers a different question and must not replace the time-weighted value.

The maximum deviation of the accepted interval means from their weighted mean is 13.79%. With an explicitly illustrative review threshold of 5%, the simple constant-rate screen flags the candidate interval. This does not prove the measurements are invalid, nor does it assess unobserved within-interval fluctuations.

The full-hour volume remains unknown. For a separately labelled sensitivity only, a documented assumption of 0–12 m³/h during the gap would add 0–1.0000 m³, giving 9.6667–10.6667 m³. Without evidence supporting that assumption, even this range is not a measured or validated bound.

22. Worked case C: recovery clocks and an undefined logarithmic point

Synthetic example. A single uninterrupted pumping period lasts 120 minutes. The reference depth is 18.00 m, and a matched end-of-pumping observation is 30.00 m. End drawdown is therefore 12.00 m. The table uses the same measuring point throughout.

At τ = 10 minutes, total time since pumping began is 130 minutes and log10(t/τ) is approximately 1.113943. The zero-minute observation remains valid as a boundary observation, but cannot be placed on that logarithmic coordinate. Replacing τ = 0 with 0.001 minutes would invent a coordinate and must not be done automatically.

If the pump restarts at recovery minute 40, the 60-minute reading no longer belongs to the same uninterrupted-recovery segment. Retain it with the restart history and refer the interpretation for review. If the end-of-pumping depth had not been measured reliably, residual drawdown relative to the accepted baseline might still be calculated, but the descriptive recovery percentage would be withheld.

These computations screen definitions and arithmetic. They do not estimate transmissivity, storativity or a permitted pumping schedule.

23. Worked case D: a high acceptance percentage can conceal a critical gap

Synthetic example. An agreed schedule contains 12 water-level observation slots. An export has 12 rows because one of the 11 recorded observations appears twice. Ten distinct slots have accepted matched observations, one has a suspect reading and one is missing. The same-time independent manual check, if present, would remain a separate series rather than being treated as an export duplicate.

A calculation of 10/11 = 90.91% after removing the missing slot from the denominator would inflate the headline by changing the agreed population. Report the original expected population and the reasons for exclusions instead.

The dataset can still support discharge-volume arithmetic over a separate, well-defined metered interval. It cannot support a precise stop-time-dependent recovery coordinate merely because most level readings are accepted.

A further sensor difference should be tested on compatible references. For example, a matched manual depth of 24.00 m and a logger depth of 24.08 m differ by 0.08 m. Whether that warrants correction depends on the methods and uncertainty; it is not automatic proof of drift or permission to subtract 0.08 m from the whole test.

25. Frequently asked questions

Does a screened pumping-test dataset prove sustainable yield?

No. Screening checks record meaning and suitability for a stated analysis. Long-term operating recommendations require appropriate hydrogeological interpretation and consideration of the installation, operating conditions and applicable permissions. A software pass must not be presented as approval to increase discharge or runtime.

Can missing flow readings be entered as zero?

Only when independent evidence establishes actual zero flow for the interval. Missing measurements, communications faults and a confirmed stopped pump are different states. Where flow is unknown, retain the gap and report only the represented volume or a separately justified estimate.

Is the average discharge sufficient for a constant-rate test?

Not by itself. Examine actual rate changes, interruptions and time weighting. The same average can arise from very different pumping histories. A rate-stability flag may require a different analysis or segmentation rather than rejection of otherwise valid observations.

Can water levels from two measuring points be compared?

Only after their relationship and conventions are established. Convert through a documented common reference and preserve original readings. A casing-height change can alter recorded depth without moving the actual water surface. Unknown geometry is not resolved by making the plotted curves align.

Should every unusual observation be removed?

No. A sharp change may document a real pump event, interference or instrument action. Flag it, inspect the context and record its treatment for the intended use. Removing a point because it conflicts with the desired curve can destroy important evidence.

Why are there two clocks during recovery?

Time since pumping began and time since pumping stopped describe different histories. A simple recovery coordinate can require both. A restart or earlier interruption must remain visible; resetting the display clock does not erase the earlier pumping response.

What happens at zero recovery time on a logarithmic plot?

Keep the observation in the original and linear-time records. The simple t/τ coordinate is undefined when τ = 0, so that coordinate is withheld. Do not manufacture a positive elapsed time merely to make the point appear on a log axis.

Is a logger reading more reliable than a manual measurement?

Not automatically. Compare timing, reference, instrument condition and method uncertainty. A logger can provide dense observations but retain an offset, wrong scaling or a frozen signal. A manual observation can provide a useful check but does not reconstruct an unmeasured continuous record.

Can corrected and estimated records be used together?

Sometimes, for a stated purpose and under an explicit rule. They must retain separate origins and visible limitations. A corrected measurement still has an original observation; an estimate supplies information that was not directly measured. Do not export a combined series with every row labelled observed.

Which records should accompany an enquiry?

Provide the raw water-level and discharge files, original field sheets, well and test identities, instrument information, measuring-point definition and pump-event log. Include previous corrections and the intended calculation. Missing items should be identified directly so that evidence recovery is focused rather than replaced with assumptions.