Running a Field Intercomparison Between Paired Turbidity Sensors

Two turbidity probes operating in the same body of water for several days can resolve questions that no bench test ever answers: whether the optical windows stay clean, whether the readings track through real-world swings in suspended sediment, and whether a newly installed instrument matches the one already trusted by a long-running monitoring program. The exercise is simple in principle but unforgiving in practice, and the Australian conditions of strong sun, warm shallow reservoirs, estuarine stratification, and episodic flood plumes will test every weakness in sensor pairing, data handling, and field rigour.

Before the waders go on and the cables get spooled out, the purpose of the intercomparison has to be sharpened. Sometimes the goal is a formal cross-calibration that lets a new probe be slotted into a network without breaking the historical record. Sometimes the question is narrower: does a backscatter unit agree with a nephelometric one within a stated band, and at what point does the agreement fall apart? Pinning down the objective first shapes the choice of deployment window, the spacing of the two probes, and the level of statistical scrutiny applied once the data come off the logger. Readers who want a refresher on how turbidity is defined in monitoring practice can consult the frequently asked questions compiled by D & A Instruments.

Planning the site, sensors, and timing

A useful intercomparison starts with a site that genuinely exercises both instruments. A protected cove with steady, low-turbidity water will not stress a backscatter head designed for high-concentration dredge plumes, just as a violently stirred boat ramp will overwhelm a sensitive nephelometric probe designed for clear drinking-water reservoirs. For an Australian operator weighing options, the choice often comes down to whether the test is meant to bracket conditions in a Tasmanian hydro storage, a Brisbane River tributary after summer storms, or a Port Phillip Bay channel where wind-driven resuspension regularly lifts readings past several hundred FNU. Each of those environments stresses the optical path in a different way, and the period chosen for the intercomparison should overlap with the most demanding condition the sensors are expected to face in service.

Matching the two units before they ever touch water is just as important. Optical geometry, measuring wavelength, reported units (FNU versus NTU versus mg/L), and the recommended formazin range all need to sit on the same page. A common mistake is to pair a side-scatter sensor with a forward-scatter unit and assume the readings will simply differ by a constant offset; in reality, particle size and refractive index drive a non-linear relationship that only becomes visible across a wide concentration range. Confirming firmware revisions, wiper cycles, and depth ratings before mobilisation saves a wasted trip to a remote site such as Lake Gordon or one of the upper Murrumbidgee impoundments, where access is restricted to daylight hours and the nearest calibration fluid may be a full day's drive away.

Mounting both probes on a common frame

Mechanical separation between the two sensors is the most underappreciated variable in a field intercomparison. If one probe sits ten centimetres ahead of the other, it samples water that the upstream instrument has already disturbed, and any cross-flow around wiper arms or the wiper motor housing will bias the comparison. A common frame, with both optical heads facing the same undisturbed flow and a lateral spacing of at least three to five optical path lengths, removes the bulk of that self-induced noise. For a reservoir deployment from a pontoon, this often means a short aluminium bar bolted across the pontoon gunwale with the two sensors offset fore and aft; in a fast-flowing creek near a Murray-Darling tributary, it means a stainless bracket that keeps both heads a fixed distance from the streambed and well away from any bedload hop.

The wiring side deserves equal care. Two independent power supplies, ideally with separate solar regulators and battery banks, prevent a regulator failure from taking both probes offline at the same moment and silently corrupting the comparison. Surge protection on each power line, separate lightning isolation for the data cables, and clearly labelled connectors all reduce the chance that a single field event, like the dry electrical storms that roll across the Western Australian wheatbelt in spring, will destroy the entire dataset. Cable management also matters for data integrity: keep the data cables away from mains or inverter runs, and route them so that the weight of the suspended cable does not rotate the sensor head as water levels change.

Configuring data loggers and time synchronisation

The single most common reason a field intercomparison is later declared inconclusive is poor time keeping between the two loggers. If the clocks drift by even a few seconds, fouling events and short-term spikes cannot be matched accurately, and the regression loses its strongest points. Both data loggers should be synchronised to a single reference, ideally a GPS-disciplined time source or a network time protocol feed taken from a mobile hotspot on the bank. Once the clocks agree at the start of the deployment, the loggers should be configured to log on a common interval, such as one minute or five minutes, and that interval should be short enough to capture the fastest expected change in suspended sediment without burying the storage media in redundant data.

Logging configuration also defines the units and statistics that will later be used for the comparison. Median or mean values over each interval are appropriate for steady flows, while minimum and maximum values become useful where resuspension pulses are expected. Storing raw counts or the underlying 4–20 mA signal alongside the engineering units gives a fallback if a sensor is later found to have a configuration error. For longer deployments, an internal event log of wiper activations, diagnostic flags, and self-test results should be enabled on each instrument so that any anomaly during the intercomparison can be traced to a specific time, a specific cause, and a specific sensor, rather than being blamed on the wrong unit after the fact.

Managing the in-situ period

Once the rig is in the water, the real work shifts from setup to supervision. A pre-planned site visit schedule, with visits at 24 hours, one week, and again near the end of the planned window, lets the operator check for biofouling, inspect the wipers, photograph the optical windows, and record any external changes such as algal scum, a new inflow of stormwater, or the arrival of wind-driven weed mats. In warm, sunlit Australian reservoirs, biofouling can begin to affect a forward-scatter sensor within seventy-two hours, and a backscatter head in a eutrophic coastal lagoon may need attention even sooner. Recording the condition of the optical windows on each visit turns an otherwise subjective judgement into a defensible data point when the comparison is later reviewed.

Field conditions during the deployment will sometimes force a pragmatic call. A flood event that pushes turbidity past the upper measuring range of the more sensitive instrument is a gift for the intercomparison, because it populates the high end of the regression, but it also means dragging a boat out in conditions that may be unsafe. Heavy rain following a long dry spell in a catchment like the one feeding Warragamba Dam can deliver a sediment pulse that is over before the next visit, so a turbidity-triggered data flag or an SMS alert from the logger is worth setting up in advance. Operators running longer campaigns in stratified systems, where temperature and oxygen layers shape where the sensors sit, will also benefit from reviewing reservoir stratification and sediment trapping before the deployment, so the probe pair is positioned in a layer where real-world change is most likely to occur.

Analysing the paired record and setting acceptance criteria

When the sensors are recovered, the data should be reviewed before the field gear is cleaned and packed away, because memory of what happened on site is freshest at that moment. Plot the two time series on the same axis, align them by timestamp, and look first for the obvious: sustained offsets, step changes that align with a wiper cycle, divergent behaviour during a known plume event, and stretches where one sensor flat-lines while the other tracks real change. These qualitative checks usually explain most of the unusual features that a purely statistical analysis will later flag, and they allow a sensible decision about which intervals to retain, which to discard as fouled, and which to annotate rather than discard.

The statistical core of the analysis is a paired regression across the retained data, ideally expressed both as a slope and intercept with confidence intervals and as a Bland–Altman plot that shows the mean bias and the limits of agreement as a function of concentration. For an intercomparison meant to slot a new probe into an existing network, the goal is usually a slope close to one and an intercept close to zero over the full range; for a comparison between two different optical technologies, a non-zero intercept and a slope that drifts at the extremes are often acceptable, provided the residuals stay inside a band defined in advance. That band should be set in the planning stage, written down, and not revised after the data are in front of the analyst, because the value of the intercomparison rests on its independence from the result.

Document everything at the end of the campaign. The site, the depth, the spacing, the cable layout, the cleaning schedule, the weather, the visits, the photos, the regression, and the final pass-or-fail judgement all belong in a single report that any future operator can read and reproduce. A well-kept intercomparison record is the difference between a network that quietly drifts out of agreement over a decade and one that remains traceable from its first probe to its latest deployment, and that traceability is what underpins defensible monitoring data in any Australian water-quality program.