Browsing by Author "van Zyl, Jakobus"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Open Access
    Evaluation of water quality in urban drinking water distribution networks: a case study in Johannesburg, South Africa
    (2024) Luyaba, Lubabalo; Okedi, John; van Zyl, Jakobus
    In the Republic of South Africa (RSA) access to drinking water is a constitutionally guaranteed human right. The supply of safe drinking water to consumers is a legal requirement and the numerical limits for drinking water quality are described in SANS 241:2015. However, in 2023 over 30% of RSAs Water Services Providers (WSPs) supplied water that was of poor quality, with breakouts of cholera becoming more frequent. Poor drinking water quality is a result of a complex combination of factors requiring different interventions, one of the complex factors is drinking water quality management in a distribution network. What makes drinking water quality management in a distribution network complex is inter alia the fact that the water quality deteriorates within the distribution network. This presents a need for tools that can assist WSPs to better understand and manage drinking water quality in a distribution network. The evaluation of widely used, credible and freely available modelling tools, that can assist WSPs with limited resources (financial and human capital) for drinking water quality management in their distribution networks, was considered worth exploring. The main objective of the study was therefore to evaluate water quality in an urban drinking water distribution network, considering a case study in RSA, utilising a widely used and freely available modelling tool. The study calibrated (hydraulic and water quality) and validated (water quality) the distribution network following international best-practice, prior to commencing with the evaluation. To ensure practical relevance for WSPs, the study focused on the key sources of uncertainty in drinking water quality modelling in water distribution networks. Drinking water quality (considering various specific determinants) was the dependent variable, with the independent variables being: hydraulic definition, level of calibration, pipe age, pipe material, water demand pattern, load shedding and tank (reservoir) mixing model. The study also considered the practical usefulness of a free and widely used tool (EPANET 2.0) in optimising a network drinking water quality sampling programme. To optimally evaluate the sources of uncertainty two independent models were developed through skeletonisation and reduction. These were the Medium-Level Detail Model – MLDM (reduced all pipes model) and the Low-Level Detail System – LLDS (significantly reduced and skeletonised). It is reasonable to assume that utilities will not always have all the distribution network hydraulic data, but they may still need to model the water quality of these networks with incomplete data (in the interest of protecting public health). A consideration of various water quality determinants(physical, chemical and biochemical), modelled on the MLDM and the LLDS showed that distribution network simplification (through reduction and skeletonisation) does not compromise water quality modelling accuracy. However, the MLDM proved accurate for more determinants (dissolved oxygen, total chlorine, chloramine and dissolved organic carbon) than those of the LLDS (free chlorine, biodegradable organic carbon). It was therefore concluded that there is some benefit in investing in additional hydraulic detail (hydraulic definition) as the returns are higher levels of water quality modelling accuracy, for certain determinants. The study then considered the impacts of the level of calibration on water quality modelling accuracy for both the LLDS and the MLDM. As expected, the level of calibration was shown to correlate directly with water quality modelling accuracy. However, the investigation showed that hydraulic definition was more important than the level of calibration, in extended period simulation. This was because the second best calibrated MLDM (known control status), produced more accurate results than the most calibrated LLDS (variable control status) when considering total chlorine. This further highlighted the need to prioritise hydraulic definition as far as practically possible. Distribution network pipes are generally underground as such their material, age and or condition is often not reliably known, resulting in a need to estimate the roughness coefficient (C-value). For both models (LLDS and MLDM) it was clear that the pipe material and age were significant parameters as there was notable variance in total chlorine concentrations (up to 34% for the MLDM and up to 95% for the LLDS) with changes in the roughness coefficient (C-value).When considering the same C-value, the MLDM was generally more accurate (range was between 10% and 20%) than the LLDS, as observed from the lower errors (difference between model outputs and field measurements). Both models followed the same pattern of water quality deterioration, meaning they were both useful for modelling purposes. For both models, the most significant adverse impacts (on water quality) were for badly corroded steel, iron and clay pipes. This meant that special care to determine pipe age (condition) must be taken when modelling networks with these materials, otherwise the model will produce errors that render the model useless. Pipe age, condition and material were found to have the most significant influence on water quality modelling accuracy. Therefore, the most effort should be expanded on this source of uncertainty, as incorrect C-values can render the water quality model outputs useless, thus undermining the entire exercise. Water demand patterns are another source of uncertainty in practise, as they are not always exhaustively known. In this study demand patterns were found to have a very minimal impact on water quality, with an absolute maximum difference of 4% when considering total chlorine. However, demand patterns are critical for hydraulic and operational considerations. South Africa has been struggling with rolling electricity blackouts (load-shedding) for well over a decade. Load-shedding has a direct impact on water supply as the study area relied on pumps (additionally, the absence of electricity influences water demand patterns). Load-shedding was shown to have the most pronounced adverse water quality impact during stages 5 (five) to 8 (eight), with water quality precautions (additional dosing or boil notices) needed during stages 7 (seven) and 8 (eight); this was because water ages increased to a point (well above 70 days) where free chlorine was well below 0.2mg/l. Tank (reservoir) mixing models were shown to influence water quality most significantly under conditions of short-circuiting (First-In-First-Out), presenting a need to understand under what conditions tank short-circuiting was likely and how it could be prevented. The identification of optimal critical sampling points is a key facet of drinking water quality management (adhering to a preventive risk management philosophy), and the selected modelling software was shown to be one of the options that can be considered for this exercise. The importance of both hydraulic and water quality parameters was observed in all scenarios, underscoring the need to understand both independently and jointly (the impact one has on the other); as the end objective is ensuring safe drinking water is delivered to consumers, to preserve public health. It was concluded that EPANET 2.0 was a useful software, that could add significant practical value to WSPs in managing drinking water quality in a distribution network to protect and preserve public health.