A water quality assessment model for the Crocodile River and Inyaka Dam

Abstract

“Water, water everywhere, nor any drop to drink”. These are the famous words from Samuel Taylor Coleridge’s poem, "The Rime of the Ancient Mariner," penned in the late 18th century, which highlight the lived realities of many South Africans. There is widespread consensus that South Africa is a water-stressed country with an average annual rainfall of only 464 mm, compared to the global average of 860 mm/year. Access to clean and safe water has become a luxury, especially in the rural areas of South Africa, where municipal services are either inadequate or non-existent. As such, these rural households rely on unsafe sources such as rivers and lakes. This study was motivated by the need to monitor the quality of surface water used by rural communities with limited municipal services. However, since the water quality monitoring process has financial implications, a more proactive and financially viable solution needed to be sought. Here, we present a desktop-based system that employs historical and ongoing data to predict water quality and provide decision-making support to officials on how and when to act during potential pollution events. This study was conducted within the context of the Mpumalanga province in the Ehlanzeni District, where there are notable potable water shortages, especially in the rural local municipalities. The study aimed to develop a water quality assessment model for the Crocodile River and Inyaka Dam. This was achieved by addressing the following objectives: (1) To investigate the microbial and physicochemical water quality characteristics of the Crocodile River and Inyaka Dam across all four seasons; (2) To investigate the bacterial community structure and the dominant bacterial taxa in the Crocodile River and Inyaka Dam across two seasons, and (3) To develop a water quality assessment prototype for the Crocodile River and the Inyaka Dam. To achieve these objectives, a quantitative method approach was employed. Data obtained from water samples were analysed as per the objectives using parametric and multivariate statistical analyses, such as Pearson correlation, bar charts, and principal component analysis, using OriginPro and SPSS. To achieve the second objective, 16S metagenomic sequencing was conducted using the PacBio platform. All bioinformatic analyses were conducted on the web-based QIIME 2 platform, while R was used to generate alpha and beta diversity graphs. Lastly, the water quality assessment model was coded and programmed on the Python platform and launched on Microsoft Windows-based computers. The key findings of this study were as follows: Literature exploration revealed that existing models of water quality assessment were vulnerable to distortions, such as eclipsing and ambiguity, which increased their uncertainty at various stages of model development and during water quality assessments. In this regard, this study recommended the incorporation of statistical methods and machine learning techniques into the development of these models to reduce uncertainty. The investigation of microbial and physicochemical characteristics of water quality in the Crocodile River and Inyaka Dam revealed that water pollution in these aquatic systems stems from both natural processes, such as stratification, and industrial activities like wastewater discharge and agriculture. The water quality in the Inyaka Dam is highly influenced by temperature stratification, which causes degradation in winter when surface temperature decreases. In contrast, the Crocodile River is largely influenced by changes in the flow regime between the observed seasons. A further exploration of the influence of physicochemical properties on the composition of bacterial communities revealed that certain microbial communities remain relatively stable despite seasonal fluctuations. Key physicochemical drivers in the Crocodile River, such as flow rate, nutrient concentrations of PO43-, NO3-N, Mn, Fe, and EC, influenced microbial distribution and diversity, irrespective of the season, favouring a more diverse and spatially variable microbial assemblage. On the other hand, the Inyaka Dam supported a more stable community structure across all seasons, largely due to water column stability and nutrient stratification. Although there were dissimilarities between the Crocodile River and Inyaka Dam, the study revealed that the phylum Proteobacteria dominated both aquatic systems across all observed seasons. Moreover, Gammaproteobacteria were more abundant across all seasons in both ecosystems, suggesting Proteobacteria's ubiquity and adaptation capabilities. These findings increase our understanding of the microbial biogeography, especially in these ecological habitats, and they also provide grounds for the development of an ecosystem-specific microbial metrics system for water quality management. This study successfully developed a predictive tool using an Artificial Neural Networks (ANNs) to forecast water quality parameters, demonstrating the potential of machine learning in improving water resource management. Despite the promising results, its effectiveness is constrained by the availability and diversity of data. The tool represents a significant step forward in the prediction of water quality, offering a more efficient and potentially more accurate alternative to traditional methods in decision support. However, to fully realise its potential, more research is needed to address its limitations, particularly to expand the data set and refine the model to improve its accuracy and applicability in various scenarios. Overall, the principal contribution of this study lies in the integration of four dimensions of water quality assessment often considered separately. These include assessment methodology, environmental variability, microbial ecology, and predictive modelling. The contribution is therefore not the production of an ANN program but the development of an integrated evidence base for context-sensitive and predictive water quality assessment.