When the Data Can't Be Trusted: Continuous Monitoring's Two Failure Modes in Australian Water

Emily Cox

8 hours ago

7 min read

Continuous monitoring at critical control points (CCPs) is expected to catch treatment failures before they reach the consumer, and to provide the evidence that it did. Two failure modes are quietly undermining that, one at each end of the Australian water sector.

The operator's daily bench turbidity check on the filtered water reads 0.42 NTU. The online turbidimeter, on the same sample line at the same moment, reads 0.08. Nobody can say how long it has been drifting: the last calibration check wasn't recorded, and there is no written rule for what happens when the two disagree. Every reading in the SCADA historian since then says the critical control point was in control, and none of it can now be relied on to prove it. At the larger end of the Australian water sector, this is the problem: plenty of data, but not the governance to trust it. At the smaller end, the analyser is often not running at all. Either way, the monitoring exists on paper but can't do its job.

The Monitoring Frequency Increasingly Expected of Water Utilities

Ensuring the quality of drinking water is essential for protecting public health and maintaining confidence in water supplies. Traditionally, drinking water quality has been monitored through grab sampling, in which samples are collected manually at specific points in time. However, water quality, like water itself, is fluid, and subject to change due to a variety of factors that intermittent sampling cannot wholly capture. Much like having surveillance cameras rather than a single daily photograph, monitoring is most effective when it is constant.

As a result, continuous monitoring is seen as the gold standard for water quality monitoring. The Australian Drinking Water Guidelines (ADWG) state explicitly that 'online and continuous monitoring should be used wherever possible, particularly at critical control points.' Additionally, 'wherever possible', a supervisory control and data acquisition (SCADA) system should be used, which is useful for real-time continuous monitoring and for maintaining records of historical data.

CCPs lie at the foundation of the ADWG's Framework for Management of Drinking Water Quality. A CCP is defined as "an activity, procedure, or process at which control can be applied, and that is essential to prevent a hazard or reduce it to an acceptable level". The Guidelines build monitoring frequency into this definition, specifying that a CCP requires monitoring "frequently enough to reveal any failures in a timely manner", with online and continuous monitoring explicitly stated as preferable. In September 2022, the introduction of microbial health-based targets (HBTs) in Chapter 5 of the ADWG sharpened what CCP control actually means in practice.

HBTs establish a quantitative health outcome target of no more than 1 × 10⁻⁶ disability adjusted life years (DALYs) per person per year, and operationalise that target through log reduction values (LRVs): the pathogen reductions each treatment barrier must achieve for a utility's source water risk category. Critically, the log reduction a treatment barrier contributes toward the health outcome target cannot be assumed; it must be continuously demonstrated by way of CCPs. A utility that cannot produce a trustworthy, complete and reviewed operational monitoring record for its CCPs therefore cannot defensibly claim the LRV credits that underpin its HBT compliance position. In this way, the HBT framework has introduced the concept of a claim that requires continuous operational evidence, and in doing so has made the monitoring governance gap visible in a way it wasn't before.

Notably, although the ADWG acknowledges that incorporating HBTs into drinking water management will take time, in 2026 there are increasing expectations to do so. In Victoria, LRV credit must be demonstrated, and NSW Health has stated that water utilities should 'assess all water supply systems against HBT guidance'. Other jurisdictions are likely to follow suit.

Plenty of Data, Not Enough Governance to Trust It

Large utilities typically have the infrastructure: years of sensor data sitting in SCADA historians covering every CCP, verification and operational monitoring data, and continuous monitoring well documented in their DWQMPs. The predominant problem, therefore, is not that there is no data, but that the data was never governed to the standard HBTs now require.

Calibration records can be incomplete, sensors are not regularly calibrated or cleaned, manual interventions to operating conditions go unrecorded, and invalid data may remain in the record despite being operationally useless, with no reconciliation procedure for resolving disagreements between online readings and grab samples.

The ADWG states clearly what is required: "results must be reviewed frequently to confirm that records are complete and accurate, and that there are no deviations from critical limits or target criteria." That review is often not happening.

For most utilities, this was never a significant problem, because the historian was not being asked to do much. The online analyser's job was real-time control: alarms and shutdowns. The stored record served as evidence that monitoring was taking place and as a resource for investigating incidents after the fact. For those purposes, a data gap, a fouled probe or an unflagged spike was an inconvenience rather than a failure. HBTs change the job the record is doing. Under the LRV framework, the historian becomes the basis for deciding whether a barrier earned its credit, and a single excursion above a critical limit can decide whether a filtration or disinfection step is credited at all. Data governed as a record of activity is now being asked to serve as evidence of performance, and it was never built for that.

Representation of a Water Treatment Plant (AI generated)

Digital Transformation in the Brochure, Broken Sensor in the Plant Room

At the other end of the market, the problem is more fundamental. Small and regional utilities across Australia are under genuine pressure to implement online continuous monitoring systems. These systems often require specialised sensors and equipment, can be logistically complicated to implement and integrate, and carry initial expenses for calibration, among other things. So while such systems are encouraged, their significant up-front cost can be a barrier for smaller utilities with small budgets. Moreover, even after installation, continuous monitoring systems can become a white elephant.

Online analysers carry ongoing costs for probe maintenance and calibration, cleaning, and the other consumables required for accurate water quality results. Remote and regional settings may also lack the supporting infrastructure the technology depends on, such as reliable power and stable internet connections. And the skilled personnel who can fix problems when they arise may be a flight away and weeks out in scheduling.

The ADWG's own guidance for small supplies is clear: "Ideally, disinfection should be monitored continuously using automatic systems with alarms." But a sensor offline for three months, or an alarm system never properly commissioned, does not constitute continuous monitoring in any meaningful sense, and it does not constitute the ongoing proof of performance LRV credits require. The white elephant problem is a structural mismatch between what the technology assumes and what the operating environment can provide, and it carries exactly the same regulatory consequence as the governance gap at larger utilities: the LRV credit for the affected CCP has little usable operational evidence to support it.

Getting the Approach Right at Both Ends

Neither failure mode requires a new technology solution. Both require a governance and capability solution applied at the right point in the process.

For large utilities, the starting point is accepting that data collection alone does not constitute ongoing monitoring in the sense the ADWG intends. A defensible CCP record requires documented calibration schedules traceable to a reference standard; written data-flagging rules that distinguish valid process data from sensor faults, maintenance periods and anomalies; a formal grab-sample reconciliation procedure triggered consistently when readings diverge; and regular trend review against target criteria, not just alarm thresholds. Alarm rationalisation – reviewing which alarms are meaningful, resetting thresholds through a documented and auditable process, and formally verifying that control systems respond to exceedances as intended – is a practical starting point that directly restores the integrity of the validated monitoring arrangement.

For small utilities, the most important intervention comes earlier: before procurement. Honest answers are needed on available communications infrastructure, realistic staff hours for system management, the proximity and cost of servicing, whether the proposed sensor suite matches the utility's actual calibration capability and, less visible but just as critical, whether the ongoing costs are realistically affordable. Where the honest answer is that a sophisticated platform cannot be sustained, a simpler and more serviceable instrument set will deliver more defensible evidence than an enterprise-level platform nobody can keep operational.

The ADWG is not asking utilities to claim the largest log reduction their treatment train could theoretically deliver; it is asking them to manage risk appropriately. In practice, that means claiming only the LRVs that can be reliably validated from the operational record, and having an honest conversation about how any shortfall against the target is met. Whether the gap is a data quality problem or an infrastructure problem, the outcome on paper is the same, but the solution is very different.

The Questions to Ask Now

Continuous monitoring was installed to control the process in real time, and the record it left behind was governed for that purpose. Health-based targets now ask that same record to prove every LRV a utility claims. For larger utilities, the test is whether CCP historian data would hold up if an auditor asked which readings were genuine process data, which reflected maintenance, and whether water was being supplied at the time. For smaller utilities it is simpler: how many days last year was each CCP analyser actually online, alarming and calibrated? If either answer is uncomfortable, the barrier is carrying a weaker LRV claim than the DWQMP says it is.