Groundwater Profiler Data Quality: Manual And Automated Spike Detection

Groundwater profiler data can reveal sharp changes in salinity, temperature, turbidity, dissolved constituents and suspended material with depth. Those changes may represent genuine hydrogeological boundaries, a contaminated lens, sediment mobilisation or an inflow from a different aquifer. They may also be caused by a fouled optical window, unstable cable movement, an air bubble or an isolated communications error. Groundwater profiler data quality assurance therefore depends on distinguishing a meaningful feature from an implausible spike.

Manual review remains valuable because experienced operators understand the site, the instrument and the sampling conditions. Automated spike detection adds consistency, speed and a defensible audit trail, especially when a deployment produces thousands of readings. The strongest approach combines both methods: software identifies suspicious observations, while a trained reviewer confirms whether they should be retained, flagged or removed.

Why Groundwater Profiles Produce Spikes

A groundwater profiler collects measurements while a sensor package moves through a bore, well, riverbed or water column. Sudden changes can occur when the instrument passes between strata or encounters a plume with a different optical or chemical signature. A real boundary often appears across several related variables, such as a concurrent shift in temperature, conductivity and turbidity. Its position may also repeat during upward and downward traverses.

Instrument behaviour can create a similar pattern. A brief interruption in telemetry may produce a zero, a repeated value or a value outside the sensor’s operating range. Cable vibration can disturb settled particles near the intake, creating a short turbidity peak. Optical sensors are particularly sensitive to bubbles, biological growth and deposits on the measurement window. Practical terminology for parameters, sensors and sediment observations can be checked in the water-quality glossary when teams are aligning field and database records.

Groundwater conditions add another source of complexity. In the Great Artesian Basin, a profiler may cross long-established aquifer layers with different chemistry, while a shallow urban bore near Perth or Brisbane may encounter rapidly changing recharge water after rainfall. Treating every sharp transition as an error would remove valuable evidence about groundwater movement.

Manual Review And Its Strengths

Manual spike detection begins with plotting the profile against depth and time. The reviewer examines the raw signal, the rate of change between adjacent readings, instrument status fields, deployment speed and any field notes. A suspicious observation is then compared with nearby values, duplicate passes and other channels. This process is especially useful when a profile contains a small number of complex transitions rather than a large, repetitive data set.

Experienced reviewers can recognise patterns that a fixed rule may misclassify. A narrow turbidity maximum at the interface between an alluvial layer and a clay-rich unit may be physically credible. A conductivity change that coincides with temperature and redox movement deserves a different treatment from a single-channel peak with no supporting evidence. Manual interpretation can also account for known local events, such as a storm-driven sediment pulse in the Murray-Darling Basin or disturbance caused by purging a monitoring bore.

The limitation is variability between people. Two analysts may apply different thresholds or make different decisions about whether a value is “too high”. Fatigue makes long records harder to inspect, and undocumented edits can weaken traceability. Manual review is therefore strongest as a contextual verification step, supported by preserved raw files and explicit decision notes.

Automated Detection Methods

Automated screening converts a profile into a set of measurable tests. A simple adjacent-point rule calculates the change between consecutive observations and flags a result when the difference exceeds a defined threshold. A rolling median or moving-average filter can identify departures from local behaviour without assuming that the entire bore has one normal value. Median absolute deviation is often more robust than standard deviation when a profile contains a few extreme observations.

Thresholds should reflect the sensor, sampling interval and environmental setting. A fixed turbidity limit may be appropriate for identifying instrument saturation, but it is less useful for identifying a genuine geological boundary. Rate-of-change limits should be expressed in relation to depth or travel distance, since a profiler moving at different speeds will produce different point spacing. Automated checks should also test for impossible values, duplicated timestamps, flat-lined periods, missing depth references and abrupt reversals in profiler direction.

A useful system assigns categories rather than making an irreversible delete-or-keep decision. For example, a record may be marked as valid, suspect, invalid, saturated, missing or requiring review. The original measurement remains available, while a quality flag and reason code are added to the processed data. This approach supports later reanalysis when a better baseline, calibration result or field note becomes available.

Comparing The Two Approaches

Manual review is more adaptable, while automation is more repeatable. A field scientist can recognise that a spike aligns with a known fracture zone, a bore construction feature or a change in pumping status. An algorithm can apply the same initial screen to every deployment, reducing the chance that a busy project team overlooks a short anomaly in a large archive.

Automated detection can produce false positives when a real profile boundary is sharp. Manual detection can produce false negatives when a slowly developing fouling problem is mistaken for a natural trend. The risk is higher in Australian projects where a single monitoring campaign may cover remote stations, long travel distances and changing weather conditions. Data collected near Darwin during a wet-season event may behave very differently from a dry-season profile in Adelaide, so site-independent thresholds require careful validation.

The practical comparison is therefore not a contest between methods. Automation should perform the first pass, prioritising observations for review and applying objective range and continuity checks. A qualified reviewer should then assess flagged points against the full profile, sensor diagnostics, duplicate runs and field records. Decisions should be recorded in a quality-control log that identifies the rule used, the person who reviewed it and the final treatment.

Building A Defensible Quality Workflow

Quality assurance starts before the instrument enters the water. Record the sensor serial number, calibration status, firmware, measurement units, profiler configuration, depth reference and planned traverse speed. A pre-deployment inspection should check connectors, cable strain relief, optical surfaces, battery condition and logger time. Where practical, collect a baseline in clean water and confirm that all channels respond within expected limits.

During acquisition, retain raw readings at their original resolution. Store depth, timestamp, direction of travel, instrument status and communications quality alongside turbidity, suspended-solids or other water-quality measurements. Notes about purging, pumping, rainfall, visible sediment, bubbles and pauses can explain anomalies that are impossible to interpret from a graph alone. Campbell Scientific support is relevant for projects using compatible logging and telemetry arrangements, particularly when data must move from a remote Australian site into a central quality-control system.

Post-processing should follow a controlled sequence: validate file structure, apply range checks, calculate depth-based changes, run automated spike rules, inspect plots and document decisions. Calibration checks and field duplicates provide an independent comparison. If the same feature appears during repeated passes at the same depth, confidence in its environmental significance increases. If it occurs once and coincides with a status alarm, the result should remain flagged until supporting evidence is available.

Handling Optical And Sediment Effects

Optical turbidity and suspended-solids sensors can respond to particle size, colour, shape and concentration. A calibration developed with one sediment type may not predict another accurately, so a sharp result can reflect a change in particle properties rather than a proportional change in mass concentration. This is especially important in estuaries, dredging areas and waterways where fine clay, organic material and coarse sand mix.

Fouling can create a gradual baseline drift, intermittent attenuation or an apparent increase in noise. Field teams working in productive waters should distinguish a single spike from a sequence of unstable readings. A practical sensor fouling guide can help analysts connect profile anomalies with algae, biofilm and deposits on optical surfaces. Cleaning records, inspection photographs and pre- and post-deployment checks should be linked to the data package.

Suspended material may also be disturbed by the profiler itself. Lowering a package too quickly, contacting the bore wall or moving close to a sediment bed can generate a real but deployment-induced plume. Record movement speed and direction, and compare descending and ascending traces. A spike that follows the instrument rather than remaining at a fixed depth is more likely to be operationally generated.

Practical Controls For Reliable Profiles

A robust programme uses thresholds as screening tools, preserves the original evidence and makes human review proportional to risk. A remote baseline survey may need conservative flags and a second field visit, while a well-characterised production monitoring site may support tighter automated limits. The quality plan should define who can alter data, which flags are reportable and how corrected outputs are versioned.

Useful controls for groundwater profiler data quality assurance include:

• Keep an untouched raw file and create a separate processed-data version for every deployment.
• Use depth-aware rate-of-change rules rather than relying on one absolute spike threshold.
• Combine range, continuity, flat-line, timestamp and instrument-status checks in the automated screen.
• Require manual review of flagged points that coincide with geological boundaries, pumping changes or duplicate-profile differences.
• Compare ascending and descending traverses to separate fixed aquifer features from movement-induced sediment disturbance.
• Record calibration, cleaning, fouling, weather, bore purging and maintenance information with the data package.
• Report the number and location of flagged observations, the decision applied and the reason for retaining or excluding each one.

This workflow supports transparent decisions across environmental consulting, groundwater remediation, research and OEM deployments. It also makes datasets easier to interpret when they are transferred between field contractors, laboratories, asset owners and regulators. In Australia’s varied monitoring environments, the combination of automated consistency checks and informed manual assessment gives profiler results a stronger technical basis without erasing genuine short-scale water-quality features.