Assessment of a Digital Coagulation Management Tool to Support Sustainable Drinking Water Treatment in Regional Operations

Abstract

Chemical coagulation is a highly important step of the conventional treatment processes, determination of the optimum coagulant dose to meet the demand of particulate materials and natural organic matters (NOMs) in raw water is crucial for good drinking water quality. WTC-Coag is a universal non-site-specific coagulant prediction model using three raw water quality parameters, UV₂₅₄, colour, and turbidity, as model inputs. The empirical model can determine the dose for maximum dissolved organic carbon (DOC) removal to achieve the conditions of enhanced coagulation; it also features an operator-selectable input—% setpoint (as % DOC removal)—to establish a dose for the desirable treated water quality. This hybrid modelling and control approach in practice is extremely useful for operators to be able to optimise the process by balancing between water quality and use of resources (chemical and sludge disposal costs) for sustainable operation. This paper discusses the practicality of this hybrid modelling approach via a long-term evaluation by comparing the plant dose against predicted dose using five years historical operations and water quality data. The assessment covered raw water quality change against treatment performance, predictability, usability and operator behaviour in response to the dose change situation. During the study period, five “black water” events were captured, and the performance of the predictability due to operational changes and operator’s response in these extreme events have been analysed. The comparison between the predicted enhanced dose and the plant dose indicated enhanced coagulation would not be always required. Furthermore, the selection of 50% setpoint from the targeted dose option matched well with the plant dose during which the lower-dose situation would be sufficient, with 90% of the predicted doses within ±10 mg/L of the plant dose and 95% of the predicted doses within ±15 mg/L of the plant dose during the normal period. The use of a correction factor to compensate for the particulate demand due to powdered activated carbon (PAC) dose during “black water” events has shown to be effective. The 50% setpoint matches with the plant alum dose over the entire period after accounting for the PAC dose, with 70% of the predicted doses within ±10 mg/L and 80% within ±15 mg/L of the plant dose. All the coagulation-related prediction functions have been evaluated and confirmed their non-site-specific nature. This study is unique in terms of using real operations data for an extended period to evaluate this novel hybrid modelling concept towards the sustainability goal.

1. Introduction

Water treatment plant (WTP) operators and production managers are continuously seeking better tools to improve treatment efficiency and plant operations to achieve drinking water quality targets and, more importantly, reduce unnecessary treatment costs to achieve a sustainable operation environment. Traditionally, jar-testing is the most common method to determine coagulant dose for process optimisation, but it is highly time-consuming, usually requiring over four hours of labour-intensive procedures [1,2]. During the rapid water quality change period, jar-testing may not be able to provide a speedy response to determine the required coagulant doses [3,4]. Determination of coagulant doses using modelling approaches can have the benefit of fast responses to the changes of raw water quality and allow more precise dosing control to achieve stable drinking water quality. A variety of advanced computing techniques, including multiple linear regression, adaptive neuro-fuzzy inference system, fuzzy weighting, partial least squares regression, artificial neural networks, decision tree regression, and random forest regressor, have been used to develop coagulation models and predict coagulant doses for drinking water treatment based on raw water quality parameters [5,6,7,8,9,10]. Some of these models were successfully employed by operators to assist in water treatment processes [11].

There are some commercial coagulant control systems, such as s::can com::pass and HACH’s RTC (Real-Time Control) solution, that use sophisticated sensors with advanced algorithms that continuously optimises coagulant dosing. These systems typically rely on sensors installed at the inlet of WTPs, and they often require additional instrumentation at multiple locations of the water treatment processes to improve dosing confidence. Generally, these systems require more resources, including expert advice, high initial capital cost for the instrument hardware (a typical plant instrument is about $20,000 AUD to $60,000 AUD + installation cost), and constant maintenance such as regular manual cleaning and periodic calibration to ensure reliable sensor performance [12]. These can therefore be expensive to manage, particularly when deployed across multiple sites. In recent years, several utilities have exploited collaborative arrangement with universities or engineering firms to develop a site-specific coagulant dose prediction system tailored to raw water characteristics and local treatment conditions [13].

The Water Treatment Control for Coagulation (WTC-Coag) model, a generic, universal tool for predicting coagulant dose, has been used at more than 10 water treatment plants across Australia for over 20 years. It was developed using hundreds of jar-test datasets and supporting mathematical equations [14]. The model is based on the relationship between alum dose and DOC reduction. This modelling approach originated from a research project within the CRC for Water Quality and Treatment’s Treatment Programme. It requires no prior site-specific calibration and is broadly applicable. The model was later adopted by industry for routine operational use to set plant alum doses and subsequently expanded into a weekly performance assessment tool [15]. WTC-Coag model utilises relatively simple water quality measurements as input data to generate dose predictions. Turbidity, colour, UV absorbance at 254 nm (UV₂₅₄), and alkalinity (optional) of raw water are input data for the software to generate the dose predictions, and they are easily accessible measurements at the on-site water treatment laboratory [15]. WTC-Coag can be used to predict the coagulant dose required to achieve maximum removal of dissolved organic carbon (DOC), described as predicted enhanced coagulation dose, based on reaching a coagulant dose where no further removal occurred (~0.15 mg/L DOC removed per 10 mg/L alum).

The model can determine the required alum dose based on the selected % removal of DOC (between 50 and 100%), which is an additional selectable setting to achieve the targeted water quality level decided by the operator in a hybrid mode (model prediction with user input based on experience). Using this option for operators to determine a lower alum dose for decreased level of treatment may be more suitable for some WTPs to achieve specified water quality targets while reducing chemical use and cost to support sustainability. The appropriate % setpoint should be determined by considering and assessing individual water sources as well as their specific requirements [14,15].

This paper demonstrates the application of this hybrid modelling and process control concept, which requires the operator’s choice of selectable input together with the modelling algorithm to determine the required dose. In addition, during the study period, several “black water” events occurred to allow a full evaluation of the predictability in extreme cases over an extended period to conduct this long-term evaluation. The second goal of this paper is to share the experience on how to implement this research outcome into a fully operational industry tool, including the initial setup and evaluation procedure using historical water quality and operations data, and how to determine the setpoint to mimic operator’s decision on dose selection. The application of this coagulation prediction model (WTC-Coag) to determine the coagulant (alum) dose required to achieve the desired drinking water quality, using both empirical modelling and user (operator) knowledge of the system, is demonstrated.

2. Methodology

2.1. Selected Water Treatment Plant

A conventional treatment plant (Plant A) with aluminium sulphate (alum) as a coagulant was selected for this study. WTP alum, soda ash, and powdered activated carbon (PAC) doses were determined by operators under their normal operations, mainly based on jar-tests of raw water, water quality data, and operators’ experience. Coagulation pH was controlled between pH 6.0 and 6.8, with an average of pH 6.4 over the study period. This plant was selected because of the highly variable water source (a major branch of the River Murray in the Riverina region of south-western New South Wales, Australia), so that the model can be assessed in transient raw water quality conditions.

2.2. Monitoring Period and Data Collection

The data used for this study covered the period from 2014 to 2018. Raw and treated water quality parameters, including DOC, UV₂₅₄, turbidity, and true colour (456 nm) in Hazen Unit (HU), were determined using a s::can spectro::lyser (Bedger Meter, Milwaukee, WI, USA). This instrument was calibrated using grab samples analysed using Standard Methods [16]. A data cleaning procedure was applied by excluding any days with missing data (either missing input water quality data or plant alum dose) from the historical dataset over the five-year study period (1811 sets of data in total). After data cleaning, 1710 sets of data points were included in this evaluation.

2.3. Coagulant Dose Prediction Software Description

Raw water quality data for WTP inlets, UV₂₅₄, colour, turbidity, were required to enter manually as the inputs of the WTC-Coag software (Version 1.0) to generate a single predicted coagulant dose (similar to a calculator, Figure S1). In this study, a semi-automatic data entry function was created to assist data entry for large datasets. WTC-Coag can be used to perform four primary functions: Enhanced Dose Prediction, Targeted Dose Prediction, Coagulation pH Control, and Post-coagulation pH Correction (details in SI). The first two functions, Enhanced Dose Prediction and Targeted Dose Prediction, can be achieved by inputting three commonly used water quality parameters—UV₂₅₄, colour, and turbidity. The third function, Coagulation pH Control, can be used to check and ensure coagulation pH is within the optimum range and can provide the amount of acid or alkaline required to bring pH to the optimum/targeted level. This function requires raw water alkalinity as an additional input, which was determined using Standard Methods [16].

In this work, only the first three functions were evaluated. Alum is expressed as aluminium sulphate, usually as Al₂(SO₄)₃•18H₂O. Other commonly used units, such as Al, Al₂O₃, and Al₂(SO₄)₃•14H₂O, Al₂(SO₄)₃•24H₂O, can be selected with built-in conversion factors. For coagulation pH control/adjustment, acid dose using either alum, sulphuric, or hydrochloric acid can be predicted; for alkaline dosing, the concentration of either quick lime 85% (CaO), hydrated lime 95%, soda ash (Na₂CO₃), or sodium bicarbonate (NaHCO₃) can be predicted.

2.4. Evaluation of Coagulant Dose Prediction Software

The coagulant dose predictability of the WTC-Coag software was evaluated by comparing between the actual plant doses versus the model predicted doses, using historical water quality parameters as model inputs. The first step involved the use of the Enhanced Dose Prediction option to produce a reference level of the maximum DOC removal and best treatment against the actual plant dose. The next step was to use the Targeted Dose Prediction function with different % removal settings, generally between 50% and 80% with 10% steps, for comparison against plant doses to establish the desired level of treatment by matching the % setpoint. This procedure was used to establish the % setpoint that mimics the operator’s decision of the desirable level of treatment using historical records. The degree of variability in the inlet water quality was utilised to evaluate the predictability and model performance under various water quality and treatment conditions. The WTP actual and predicted coagulant doses, as well as other parameters, were superimposed on the time-series chart for various comparisons. The pH control predication was evaluated in a similar way by comparing the plant soda ash dose against predicted soda ash dose using raw water alkalinity, plant alum dose, and measured coagulation pH as model inputs. All predictive functions were assessed and Microsoft Excel was selected to perform the statistical analysis, including R² determination.

3. Results and Discussion

3.1. Raw Water Quality Overview

Natural organic matter (NOM) imparts colour in water and is a precursor for the formation of disinfection by-products, including trihalomethanes. It also contributes to chlorine demand and causes reduction in chlorine residuals. There are several water-quality parameters that serve as indicators of NOM in water, including DOC or total organic carbon (TOC), UV₂₅₄, UV transmissivity at 254 nm (UVT), and true colour [17,18]. The raw (WTP inlet) water, as indicated by four key water quality parameters—DOC, UV₂₅₄, colour, and turbidity—follow a similar slight upward trend over the 5-year study period from January 2014 to December 2018 (Figure 1). The DOC concentration was in the range of 0.9–12.6 mg/L with an average of 3.6 mg/L, UV₂₅₄ was between 0.04 and 0.59 cm⁻¹ (average 0.14 cm⁻¹), colour was between 5 and 145 HU (average 31 HU), and turbidity was between 11 and 223 NTU (average 33 NTU). All water quality data were within the WTC-Coag design limits (UV254 < 0.7 cm⁻¹, colour at 456 nm < 200 HU, and turbidity < 1000 NTU). WTC-Coag can provide alum dose prediction in the water quality range.

During the study period, the WTP was challenged by several extreme water quality incidents (high DOC, UV₂₅₄, turbidity, and colour) each year (i. July 2014–September 2014, ii. August 2015–December 2015, iii. June 2016–February 2017, iv. September 2017–December 2017, v. September 2018 to December 2018). Generally, these incidents are described as “black water” events, and one of the major “black water” events was in 2016 (Figure 1a). Visual inspection shows that DOC, UV₂₅₄, and colour follow a similar diurnal trend. Generally, “black water” events occur after heavy rains caused flooding and NOM is washed into the river. Turbidity increased with DOC in 2017 and 2018 events, but during the 2014, 2015, and 2016 events, it showed no increase or a slightly decrease compared with the other three water quality parameters. The regression lines show a slight upward trend for DOC, UV₂₅₄, and colour, except for turbidity, which shows a flat/slight downward trend over the five-year study period (Figure 1a). This shows that the sources of NOM and turbidity may not be related and change differently.

It is well understood that the removal mechanism of particulate matters (turbidity) by coagulation is not the same as the NOM (as DOC). Usually, removal of turbidity occurs more readily in a lower alum dose range than DOC. This may have an impact on the removal performance, particularly during “black water” events when these two parameters are changing differently. It is then further complicated by NOM character playing an important role in the coagulation efficiency and is strongly related to DOC removal [17,18]. Using simple parameters such as specific UV absorbance (SUVA) and Colour-to-DOC ratio is an easy and simple way to assess whether there was a NOM character change during “black water” events. Figure 1b shows that “black water” events not only exhibit an increase in DOC concentration (Figure 1a), but also an increase in both SUVA and Colour-to-DOC ratio, which are expected to affect the efficiency of the coagulation process as NOM character change can be linked with DOC removal performance. The change to a higher SUVA value and higher Colour-to-DOC ratio potentially improve the coagulation efficiency, resulting in a higher DOC removal at the same dose [19]. Considering higher SUVA means that the DOC is more amenable to being removed by the process during the event, meaning even the raw-water DOC concentration is higher compared to the normal or “black water” event in unreported periods, alum dose may not need to be increased in proportion to the increase in the DOC concentration. Thus, using UV₂₅₄ as one of the input parameters for the WTC-Coag model to reflect the NOM character in relation to coagulant dose is practically important.

Figure 2 shows the correlation between UV₂₅₄, DOC, and colour, which is a simple method to understand how these parameters are inter-related, as well as providing additional information to understand the characteristics of raw water. UV₂₅₄, colour, and DOC are closely linked with good linearities (both R²s are > 0.9), which confirms the observation that UV₂₅₄, colour, and DOC follow similar trends in Figure 1a, while turbidity does not. In addition, the observed SUVA and Colour-to-DOC ratio increase (NOM character change) in Figure 1b may not be as significant during the “black water” events as the UV₂₅₄ and DOC, and colour and DOC are still following a linear relationship in Figure 2a,b.

3.2. Comparison of Predicted Doses with Actual WTP Doses

In this section, two WTC-Coag prediction functions, Enhanced Dose Prediction and Targeted Dose Prediction, are discussed. Enhanced Dose Prediction requires only the three raw water quality parameters, turbidity, colour, and UV₂₅₄, to predict the required alum dose for enhanced coagulation that can achieve maximum DOC removal. In Figure 3a, the predicted enhanced dose is shown to be higher than the plant dose, which shows that the level of treatment by enhanced coagulation would not be required (operator’s decision). Especially during the normal or “black water” event during unreported periods, the Enhanced Dose Prediction was almost double that of the plant alum dose. This indicated high-quality treated water from enhanced coagulation was not required by operations. Therefore, the Targeted Dose Prediction function of WTC-Coag software, which allows us to input a % setpoint to select the most suitable treatment level for the targeted treated water quality, would be a more appropriate option in matching the decision of the operator.

The term “coagulable DOC” is introduced and defined as the maximum amount of DOC that can be removed by coagulation (approximately between 50 and 80% of total DOC); the remainder is recalcitrant and cannot be removed. An initial implementation procedure was set up using historical data to determine the % setpoint which can mimic the operator’s decision. It is worth mentioning that the original model design was based on using the response curve of % removal of coagulable DOC against dose [14]. However, during the initial model implementation by operations, operators suggested to name it as % setpoint, which is easier for operators to understand and apply. The optimum % setpoint is determined by using a series of dose prediction outputs (timeseries plots) in a range of setpoints, such as 50, 60, 70, 80, and 90% coagulable DOC removal dose predictions, as well as by identifying the closest visual match between the plant and predicted alum doses from one of the % setpoints. In addition, in a threshold-based evaluation using the criteria of less than 10 mg/L difference between plant alum dose and predicted alum dose over the entire period, 50% setpoint was determined and selected for future evaluation (40% setpoint: 73.9% prediction within 10 mg/L, 50% setpoint: 74.5% prediction within 10 mg/L, 60% setpoint: 72.0% prediction within 10 mg/L, 70% setpoint: 55.8% prediction within 10 mg/L, 80% setpoint: 22.2% prediction within 10 mg/L).

From the first observation, during the normal periods, the predicted dose based on 50% setpoint matches well with the plant alum dose. However, during the “black water” events, the predicted dose based on 50% setpoint did not match with the plant alum dose. The plant alum dose during those periods was much higher than the targeted 50% predicted dose. This indicated that either the water quality and NOM character or the treatment process changed during the “black water” periods. It was also possible that both were changed. It would be logical to consider the “black water” period and normal period separately and use different % setpoints. An evaluation of the prediction during the normal period only, using 50% setpoint, was conducted, whereby 90% of the predicted doses were within ±10 mg/L of the plant dose and 95% of the predicted doses were within ±15 mg/L of the plant dose.

After obtaining additional dosing information from the WTP (Plant A), it was found that PAC was dosed at the head of the plant during the “black water” periods. Dosing PAC can add turbidity (particles) to the water, which increases alum demand. Therefore, using the same targeted 50% setpoint shows the predicted dose was lowered as compared with the plant alum dose used at that time, as the model did not consider the added PAC (particles) that would consume alum during their removal. Several equation types, including linear, exponential, logarithmic, polynomial, and power, were initially considered in establishing the relationship between turbidity (NTU) and PAC (mg/L) to obtain a conversion equation to convert PAC dose (as mg/L) to turbidity (NTU), with the combined turbidity used as the model input. A similar threshold-based evaluation using the same criteria of less than 10 mg/L difference between plant alum dose and predicted alum dose over the entire study period was employed to determine the best equation type and coefficients for the equation. An equation, Turbidity = 10^(a×PAC), a = 0.13, was determined to be used for further evaluation. Figure 3b shows PAC dose, plant alum dose, and targeted predicted dose with PAC corrected as turbidity increases. The 50% setpoint matches with the plant alum dose over the entire period after accounting for the PAC dose, 70% of the predicted dose were within ±10 mg/L, and 80% were within ±15 mg/L of the plant dose. This study only used a single % setpoint for evaluation purposes. In real operations, the operator will adjust the % setpoint to optimise the process to meet water quality goals.

3.3. Coagulation pH Control Prediction

Alum is acidic, and adding it to water for coagulation depresses pH. The final pH after coagulant (alum) is added depends on the alkalinity (buffering capacity) of the raw water. Coagulation pH is one of the key process control parameters for higher DOC removal; a pH between 6.2 and 6.5 would be the optimum pH range. In practice, coagulation pH would be between pH 6 and 7. For low-pH water, alkaline is needed to bring pH up to the optimum range (pH 6–7); for high-pH water, acid is needed to bring the pH down. The traditional method of pH control involves trial and error. The coagulation pH control prediction function provides a pH control adjustment function to determine the required acid or alkaline solution for the optimum pH range. The function can help water treatment operators maintain balanced pH levels of inlet water more easily and allow them to automatically adjust the pH by dosing the appropriate amount of chemicals, particularly during extreme water quality events [20].

The first two WTC-Coag prediction functions, Enhanced Dose Prediction and Targeted Dose Prediction, provide the predicted alum dose, and with the alkalinity of the water, the pH after alum addition can be predicted and the amount of acid or alkaline required to change to the targeted pH can be determined.

The standard WTC-Coag software obtained the predicted alum dose from the raw water quality. Then, using the predicted alum dose as the amount of alum (acid) added to the raw water and with the raw water alkalinity, the resultant pH can be predicted. Then, the amount of acid or alkaline needed to adjust the coagulation pH is determined by the difference between the predicted pH after alum is added and the targeted pH. In this study, a modified procedure was used with the WTC-Coag algorithms instead of using the standard WTC-Coag input procedure. The plant alum dose, alkalinity, and measured coagulation pH were input into the WTC-Coag algorithms in lieu of the standard procedure described. This modified input procedure uses the algorithm to predict the resultant pH under these conditions (plant dose and alkalinity). The difference between the predicted pH after alum addition and targeted pH was used to determine the acid or alkaline dose required. Figure 4 shows the predicted and actual doses at the treatment plant during the study period, and the close agreement between them demonstrates the predictive capability of the algorithm.

3.4. Treated Water Quality and Treatment Performance

Critical control point approach is used by the WTP to achieve good-quality treated water. Treatment plant operators target < 0.2 NTU and critical alert (shut down filter and contact regulatory agencies) if turbidity exceeds 0.5 NTU at the outlet of the individual filters. Corrective action is taken between target and critical alert to manage the incident. Generally, the dose selection is aimed at mainly managing both turbidity and DOC.

Both colour (0–5 HU) and turbidity (0.07–0.3 NTU) in treated water were well below the operation guideline levels (colour below 10 HU and turbidity below 0.5 NTU, with a small number of samples over the target level of 0.2 NTU) over the study period (Figure 5). DOC and UV₂₅₄ show higher levels during the 2016 “black water” event.

The treatment performance was assessed based on actual % removal of DOC, UV₂₅₄, colour, and turbidity (Figure 6a). Both turbidity and colour showed excellent removal and achieved 99.7% and 98.6% average removal, respectively. UV₂₅₄ had an average removal of 88.3% while DOC had an average removal of 78.0%. In general, UV₂₅₄ is expected to have a higher % removal compared with the DOC for the same alum dose; this is in-line with the earlier findings [14,15]. Upon observation, both UV₂₅₄ and DOC had lower average % removal compared to turbidity and colour; in some periods, DOC removal dropped to around 50% (the lowest DOC removal was 45.7%). The turbidity, colour, UV₂₅₄, and DOC removal were in the range of 82.2–100%, 70.6–100%, 48.5–100, and 45.7–98.9%, respectively. When comparing % removal (difference between raw water and treated water) before and after the “black water” events in the study period, even when the raw-water DOC and UV₂₅₄ levels increased during the events (Figure 6a), the % removal was higher than before the event. Higher DOC and UV₂₅₄ levels were observed in the treated water during the “black water” periods compared with the normal periods. This could be improved by optimised treatment to increase the removal of both parameters.

In relation to the 2016 event, all four water quality parameters have shown good % removal prior to the event. In fact, both DOC and UV₂₅₄ showed even higher removal during the event, which indicated that the operators responded to the water quality change by using a different operation mode, such as a higher dose to ensure that they met the treated water quality target, or by introducing additional treatment, including PAC dose (to be further discussed later). Depending on how the targeted treated water quality was set, whether it was based on % DOC removal or treated water DOC concentration, it would be good to understand these water quality characteristics to ensure the lower DOC and UV₂₅₄ in treated water can be achieved during “black water” events.

During the 2016 event, DOC removal per alum was higher (0.12 mg/L of DOC removed by 1 mg/L of alum) and the average DOC removal per alum was 0.07 mg/L of DOC removed by 1 mg/L of alum) over the entire study period, which indicated a better removal was achieved with the same alum dose (Figure 6b). In other periods, DOC removal was between 0.04 and 0.06 mg/L of DOC removed by 1 mg/L of alum. This can be explained by using the SUVA value to reflect the NOM character: SUVA during “black water” events had a high SUVA range, where it was up to around 4.5 m⁻¹mg⁻¹L. Meanwhile, during the “non-black water” event period, the SUVA was between 3.0 and 3.5 m⁻¹mg⁻¹L. Organics with higher SUVA can be removed readily by coagulation.

3.5. Operator Response with and Without the Support of Online Water Quality Monitoring System

In this study, the WTP has an online water quality monitor installed and is able to display real-time water quality data to assist operators. In Figure 7, a 1st-order derivative is used to show the operator response (plant dose change) against water quality change; in this case, UV₂₅₄ was used. On the positive side of the y-axis (above zero) in Figure 7, it is referring to either raw-water UV₂₅₄ or plant alum dose increase; on the negative side of the y-axis (below zero), it refers to raw-water UV₂₅₄ decrease or plant alum dose decrease.

The fact that there was no lag between the two curves indicated that the operator adjusted the dose immediately when water quality changed (just only one or two occasions of time lag). This response was different from an earlier (Barossa WTP) experience [21] when operators were using jar-testing for optimisation, where a time lag was observed. This can be attributed to the use of an online water quality monitoring system, which provided immediate access to water quality change. This would be a different situation compared with that of the Barossa WTP case, in which their operations rely heavily on grab sample monitoring that delayed the response of the operators. The use of real-time water quality data is useful in terms of process control. WTP operations rely on the results of jar-tests to adjust coagulant doses when raw water quality changes, particularly for regional water utilities. However, when water quality events occur, operators often cannot adjust chemical doses quickly because jar-tests take hours, and repeated jar tests are often required. With online water quality monitoring instruments combined with coagulant dosing prediction software, operators can make dosing adjustments with greater confidence without worrying about under- or over-dosing, which may put additional constraints on the treatment process [4]. Findings of a previous study showed that operators in South Australia have great confidence in using the WTC-Coag software for coagulation control [21].

4. Conclusions

This long-term evaluation demonstrates that integrating the WTC-Coag model with operator oversight provides a robust and practical approach to optimising coagulation under highly variable water quality conditions, including multiple “black water” events. The hybrid modelling–control strategy successfully determined alum doses using the % setpoint option, with dose predictions closely matching plant practice even when PAC dosing required correction through turbidity inputs. The study also extended model functionality by using alum dose and raw water quality to predict treated water DOC through a response curve method, showing promising accuracy, while the coagulation pH control adjustment reliably matched plant soda ash dosing. By validating coagulation prediction function of the WTC-Coag against five years of historical data across several extreme events, the study confirms that the WTC-Coag software, originally designed with a non-site-specific structure, is applicable at a new WTP and consistent with its successful use elsewhere. The novelty of this work lies in demonstrating that a non-site-specific coagulation model can operate effectively in a real-world environment when paired with human judgement, overcoming limitations of fully automated systems that struggle with sensor drift, calibration issues, and unmodelled conditions. Future work should focus on integrating PAC dosing directly into model inputs and expanding validation across additional WTPs and climatic regions to further strengthen confidence in the model’s generalisability and operational value.