Exploring the drinking water supply in Rwanda with panel data analysis

The provision of water, electricity, and gas in Rwanda has undergone a transformative historical trajectory, spanning back to 1939 when the ‘Régie de Production et de Distribution d'Eau et d'Electricité’ (REGIDESO) initiated operations in Bujumbura, Burundi. Subsequently, in 1963, Rwanda established its own supplier, REGIDESO Rwanda, marking the inception of a journey marked by evolution and adaptation. As urban settlements expanded, the demand for water, electricity, and gas grew, prompting multiple restructures in the management of these vital resources (WASAC Group 2023).

In 2014, the establishment of the Water and Sanitation Corporation (WASAC) emerged as a pivotal response to challenges such as inadequate planning, inefficient operations, and management focus issues in Rwanda (Office of the Prime Minister 2014; WASAC Group 2023). WASAC's creation aimed to enhance the viability and autonomy of water utilities, with a focus on sustainable and customer-centric approaches. Entrusted with the task of establishing an effective monitoring and evaluation (M&E) system (Office of the Prime Minister 2014), WASAC is expected to play a critical role towards ensuring universal access to water supply, the sustainable functionality of infrastructure, the reliability of urban water supply services, and the strengthening of institutional frameworks (Ministry of Infrastructure 2016).

However, persistent challenges have threatened the attainment of safe, reliable, and affordable water supply services for all Rwandans. These challenges include water resource scarcity, human resource capacity gaps, low private sector investment, high non-revenue water (NRW) (water loss), and water quality compliance issues (World Bank 2012; Rwanda Utilities Regulatory Authority 2022). Aligned with the United Nations' clean water and sanitation sustainable development goal (United Nations Development Programme 2015), the Government of Rwanda, through its National Strategy for Transformation, aims for universal and equitable access to safe drinking water by 2024 (Ministry of Finance & Economic Planning 2017).

It is in that framework that this study endeavors to conduct exploratory data analytics to assess the effectiveness of the water supply system in Rwanda in meeting the strategic goal above. The specific objectives include assessing the reliability and quality of drinking water managed by WASAC, and identifying the water treatment plants (WTPs) that yield the most cost-effective water production. Additionally, the study aims to uncover the factors influencing the production cost, quality, and supply of drinking water. Through a comprehensive analysis, this research seeks to contribute valuable insights for enhancing the efficiency and sustainability of Rwanda's water supply infrastructure, addressing both the quality and accessibility aspects of water resources.

The previous literature made significant contributions to understanding the water challenges in Rwanda. While each study delved into specific aspects, the overarching message resonated with the imperative for proactive measures to ensure safe, reliable, and affordable water supply, aligning with global water sustainability goals. A common theme emerged from different studies, emphasizing substantial challenges related to the safety and reliability of drinking water. These challenges underscored the ‘pressing need for targeted interventions’ to address water quality and supply issues:

• – fecal contamination, seasonal variations in water sources, NRW, operational challenges, and the underutilization of water safety plans (Wyatt 2010; Huttinger et al. 2015; Karamage et al. 2016; Kirby et al. 2016; Mukanyandwi et al. 2019; Herschan et al. 2023).

• – The need for sustainable water management, considering climate change, population growth, and regional disparities (Barifashe 2014; Osei et al. 2015; Mukanyandwi et al. 2018; Ramirez Pasillas et al. 2019).

In the course of examining the literature above, it became apparent that their research methodologies predominantly revolved around cross-sectional data. The data collection methods used included laboratory tests, semi-structured interviews, systematic reviews, surveys, and case studies. The analytical techniques employed in these studies typically involved descriptive statistics, logistic regression, or linear regression modeling. However, this prevailing reliance on cross-sectional data emphasizes the need for more comprehensive analytical tools. This is particularly important for triangulating the results and addressing the intricate spatiotemporal dynamics inherent in water-related phenomena.

In seeking to enhance analytical depth concerning water supply, safety, and cost-effectiveness, this research incorporated panel data analysis. Recognized for its efficacy in investigating dynamic patterns across regions and time, this method emerged as a valuable approach to unravel the complexities inherent in water-related factors (Hsiao 2007). The potential adoption of panel data analysis aligned with the overarching goal of contributing to the discourse on sustainable water management, aiming to discern the multifaceted influences on water quality and supply.

Another notable contribution of this research lies in its departure from the narrow focus on specific aspects of water-related challenges, as characterized by the research discussed above. While previous studies tended to delve deeply into particular facets such as water quality, NRW, or water resource management, this research adopted a more holistic approach. By conducting a comprehensive examination, it aimed to encompass a broader spectrum of factors crucial to sustainable water management. This departure was particularly significant because it allowed for a more integrated understanding of the multifaceted dynamics of the water supply in Rwanda.

In summary, this study's relevance stems from addressing persistent challenges in Rwanda's water sector. By adopting a holistic approach and utilizing panel data analysis, it aims to provide nuanced insights, surpassing previous methodological limitations. The research aspires not only to fill existing gaps but also to offer actionable recommendations for policymakers and stakeholders. Ultimately, its significance lies in contributing to efficient and sustainable water management, aligning with Rwanda's goals for universal and safe water access.

During the research period, WASAC was undergoing a transformation in its approach to treating drinking water. Instead of employing the traditional sequence of rapid mix, flocculation, sedimentation, filtration, and disinfection, they have adopted a new method. This innovative process emphasizes coagulation, filtration, and pumping. This shift demonstrates WASAC's commitment to enhancing its water treatment methods to ensure efficiency and effectiveness in providing safe and clean drinking water.

For coagulation, the majority of WTPs have adopted Sudfloc chemical as a substitute for a combination of aluminum sulfate, lime, and polymers. According to the WASAC technical department, Sudfloc proved to be efficient in coagulating high-turbidity raw water, and it demonstrated greater cost-effectiveness and higher volume efficiency in treating raw water. The omni filtration system was in place to provide the highest level of filtration and flexibility. Finally, the pumping stations utilized low-speed pumps and improved energy-efficient pumps.

Secondary datasets from WASAC, Rwanda Utilities Regulatory Authority (RURA), and the Rwanda Meteorology Agency (Meteo Rwanda) were collected, processed, and analyzed. WASAC provided daily WTP data covering the period from 1 July 2017 to 30 June 2020. This included daily volumes of the raw, treated, and supplied water along with information on chemical products used during the same timeframe. Additionally, WASAC provided details on the total costs incurred during the water treatment process. WASAC also provided data on the water quality, such as chemical products used, the turbidity of the raw water, and the turbidity of the treated water, as well as the residual-free chlorine and the power of hydrogen (pH) in the treated water during the study period.

This research also collected RURA quarterly statistical reports published in its website. These reports reflected the number of WASAC subscribers and the quantities of water that were treated, supplied, billed, and lost from the quarter that ended 30 September 2017 to 30 June 2022. While the data of the quarter July–September 2013 served as the reference, the analysis used quarterly data from 1 July 2017 to 30 June 2022. Since WASAC, as a case study, started operations in 2014, three years (2014–2017) were considered as a transition, and therefore, related data were not included in the analysis.

Meteo Rwanda also provided daily rainfall data from 2017 to 2020 for various stations, including Kigali, Kamembe, Gisenyi, Gikongoro, Kibungo, Byumba, Busogo, Bugarama, Ruhengeri, Gitega, Rubengera, Byimana, Kawangire, and Nyagatare. Each WTP was associated with the rainfall data from the nearest station, as indicated in Table 1.

The data collected were compiled and then analyzed.

This included raw water, treated water, supplied water, billed water, NRW, NRW as percentage of supplied water, total number of customers, percentage of billed supplied water, and NRW per subscriber from 1 July 2017 to 30 June 2022. This dataset helped compute annual water volume averages used for trend analysis to determine WASAC's capacity to meet the water needs of Rwandans.

Quarterly averages of the last three years (1 July 2019 to 30 June 2022) also helped identify the most critical quarters when water treatment, supply, and loss increase or decrease. This trend analysis made reference to the performance of the quarter that ended on 30 September 2013, just before the establishment of the WASAC.

This is for the fiscal year that ended 30 June 2020 (2019/2020), indicating the WTP, month, quarter, raw water, treated water, supplied water, NRW, water cost, chemical cost, energy cost, oil cost, staff cost, total direct cost, overhead, depreciation cost, total expenditure, rainfall, quality performance score, drinking water quality index (DWQI), province (location), unit cost, and raw water turbidity.

The fiscal year 2019/2020 was the last year where full operation could be guaranteed, and the data were available. Subsequent years faced issues related to the COVID-19 pandemic. These data were used to perform water production and supply efficiency analyses, as well as to determine the factors influencing water supply and quality. Random effects of the generalized least squares regression model were developed to analyze the panel data described above. Unlike the fixed effects model, the variation across WTPs was assumed to be random and uncorrelated with the predictors included in the models, which was confirmed with the Hausman test.

Rainfall was included in this analysis as a factor that could increase or decrease the volume of raw water (Wikipedia 2022). Overall, Rwanda experiences four climatic seasons, represented by the long rainy season (March to May) and the short rainy season (September to November). These seasons alternate with the long dry season (June to August) and the short dry season (December to February) (Climate Change Knowledge Portal 2021).

The water quality data were compiled in index ‘performance (discrepancy) score (PerfScore)’ (Mwitirehe Kipruto & Charles 2022). It was used to measure the extent to which the treated water complied with the quality requirements in terms of the standard values for turbidity, pH, and residual-free chlorine as shown in Table 2 (Rwanda Standards Board 2011; World Health Organization 2018).

The PerfScore was complemented with the DWQI, also computed on nine chemicals used by WTPs, which included Sudfloc 3870, Sudfloc 3895, Zetafloc 2350, aluminum sulfate, calcium hydroxide, sodium chloride, calcium hypochlorite, lime, polymer, twin oxide, and Hinco Alpha. In this method, the water quality rating scale, relative weight, and overall DWQI were calculated as follows (Tyagi et al. 2013; Akter et al. 2016):

• i. Calculate the quality rating scale:

  

  (3)

  where C_i and S_i indicate, respectively, the concentration of i parameter and standard value of i parameter.
• ii. Calculate the relative weight:

  

  (4)

  where the standard value of the i parameter is inversely proportional to the relative weight.
• iii. Finally, calculate the overall DWQI according to the following expression:

  

  (5)

The DWQI was used to measure the extent to which chemical inputs used in the water treatment process complied with acceptable limits set by competent authorities. The water treatment process involved the use of various chemicals that can be harmful to human life. This is the reason why the dosage should respect the acceptable limits set by World Health Organization, Rwanda Standards Board (RSB), or product manufacturers. DWQI values were, therefore, calculated and rated as per Table 3 (Tyagi et al. 2013; Akter et al. 2016).

Finally, the data from the literature review were incorporated into the analysis to complement the information collected from the data sources described above. More importantly, the data from the 2022 Rwanda Population and Housing Census helped determine the contribution of WASAC's water sources in providing Rwandans with access to improved drinking water sources compared to other sources of drinking water.

Over the period from 1 July 2017 to 30 June 2022, the volumes of supplied water increased alongside a rise in the number of subscribers. However, this increase did not lead to a significant reduction in the percentage of NRW. On average, out of the 54,254,239 m³ of water annually supplied by WASAC to urban and peri-urban areas in Rwanda, 22,558,230 m³ (i.e., 41.58%) was lost (Table 4). This implies that only 31,696,009 m³ of water was billed to customers. On average, each subscriber received 132 m³ of water annually.

During the same period, the increased water supply was pushed by increased extractions of raw water. Specifically, as of 30 June 2022, the raw water had increased by 1.4587 times 50,111,653 m³ of raw water extracted in 5 years. This situation also resulted in increased annual quantities of treated water. Out of 294,681,598 m³ of raw water (an average of 58,532,446 m³ per year), 95.75% were treated, and 96.14% of the treated water, or 92.06% of the raw water were supplied. Over the period of 5 years on average, 287,608 water subscribers had access to 158,480,044 m³ of water, equivalent to 551 m³ per subscriber.

On the other hand, the increase in treated and supplied water volumes did not consistently meet the demand. The average billed water per subscriber varied over the study period (Table 4). Figure 2 illustrates fluctuations in water production, supplies, and NRW between 1 July 2019 and 30 June 2022. On a quarterly basis, the number of subscribers continued to grow. Notably, the water production increased from 3 million m³ in Q3 2013 to 14–18 million m³ from Q3 2019 to Q2 2022, aligning with the surge in water demand.

Figure 2 also indicates that the percentage of NRW increased from 41 to 46% between 1 July 2019 and 30 June 2022. However, during the same period, the difference between supplied and treated water (NRW1*) decreased from 4 to 1%. In comparison to NRW, WASAC has effectively managed to minimize water loss during the treatment process (NRW1*).

Despite this improvement, the volume of water produced per customer has been inconsistent, fluctuating between 56 and 64 m³ due to the increase in NRW. This variability suggests challenges in consistently meeting customer demands for water supply. Moreover, the comparison of the treated and supplied water as well as the NRW indicates that during the dry season (Q3), there tended to be an increase in water treatment and relatively low levels of NRW. This will be discussed further later.

During the study period, WASAC provided an average of 14,563 m³ of water each quarter, as indicated in Table 5. The majority of this water (85.21%) was distributed to various regions, including Kigali City, Rubavu, Musanze, Rwamagana, Nyagatare, and Bugesera districts. Notably, Kigali City received the largest share, accounting for 60.57% of the total water supplied. Among areas outside of Kigali, Rubavu district received the highest volume of water, followed by Musanze, Rwamagana, Nyagatare, and Bugesera districts.

Moreover, when examining the average water volumes billed and lost per quarter, they were recorded at 8.25 and 6.31 million m³, respectively. Kigali City stood out with the highest average volumes of NRW per quarter, totaling 3.73 million m³, constituting 59% of the total NRW and 42.27% of the water supplied to the city.

Following Kigali City, the eastern province exhibited 1.03 million m³ of NRW, accounting for 16.35% of the total lost water and 47% of the water supplied in that province. The western province followed closely with 586,000 m³ of NRW, representing 9.28% of the total lost water and 45.59% of the water supplied in that region. The southern province reported an equivalent volume of NRW, but it represented 43.02% of the water supplied in the province. Lastly, the northern province registered 378,000 m³ of lost water, making up 5.99% of the total lost and 40.21% of the water supplied in that province.

The dashboard allowed a detailed analysis of each plant's water quality, taking into account pH, turbidity, and chlorine levels. For instance, Kadahokwa, ranked second for potable water, had the median pH of 7.00 (neutral), median turbidity at 0.49 nephelometric turbidity unit (NTU) (within the recommended range of 0.2–0.5 NTU), and median residual-free chlorine of 0.70 mg/l (slightly outside the RSB range but still recommended for potential fecal contamination). As for Mutobo, it scored 0.02 as a result of the median pH of 7.20, median turbidity of 0.02 NTU and median residual-free chlorine of 0.65 mg/l. However, Rwasaburo plant had the poorest score of 1.09 because all the actual median values in input variables were mostly outside the critical values. For example, its pH was 6.5, while the recommended range was between 6.5 and 8.5, and it had a median turbidity of 3.38 NTU (the highest) and residual-free chlorine of 0.9 mg/l, both of which were outside the recommended ranges.

In specific potable water production processes, the raw water often exhibited low turbidity. For instance, the raw waters of Cyunyu and Mutobo WTPs had median raw water turbidities of 0.49 and 0.02 NTU, respectively. These sources maintained their low turbidities even after treatment (0.46 and 0.02, respectively), indicating the highest quality of raw water (Figure 3). In contrast, other WTPs with initially high raw water turbidity experienced a significant reduction after treatment, as seen in the following examples:

• – Nzove, with a median turbidity of 956.21 NTU before treatment, decreased to 1.98.

• – Nyagatare, with a median turbidity of 395.90 NTU before treatment, decreased to 0.75.

• – Rwasaburo, with a median turbidity of 4.00 NTU before treatment, reduced to a score of 3.38.

• – Kimisagara, with a median turbidity of 170.65 NTU before treatment, attained a score of 1.01.

• – Gihira, with a median turbidity of 794.54 NTU before treatment, decreased to 2.31.

Generally, the water treatment process has significantly reduced the prior turbidity level. However, this analysis was not possible for pH and residual-free chlorine data because their raw water data were missing.

All WTPs were rated A in terms of DWQI median values. The highest DWQIs were found in Ngenda, Shyogwe-Mayaga, Mpanga, Gihuma, and Kanyonyomba, but they were all still under excellent water quality (A rating) since they were all less than 25. Generally, DWQI values were at their highest levels during the rainy seasons, i.e., in April, October, November, and December. The dashboard easily distinguished two categories: one with the highest DWQI values and another with the lowest DWQI values. The highest DWQI category included Ngenda, Shyogwe-Mayaga, Mpanga, Gihuma, and Kanyonyombya WTPs. The lowest DWQI category encompassed Kanyabusage, Cyunyu, Mutobo, Nyamabuye, and Nyagatare WTPs.

The comparison between treated water volumes and DWQI showed no correlation. The southern province, hosting Mpanga WTP, had the highest median value of 8.4. The eastern, City of Kigali, northern, and western provinces followed with median values of 6.8, 3.9, 1.3, and 0.6, respectively. Under such conditions, the southern and eastern provinces were characterized by the highest median values in DWQI compared with other provinces. Moreover, a relationship between DWQI and PerfScore was suspected but it was not proven statistically significant (Pearson correlation = −0.005, t = −0.55, df = 12,127, p > 0.05). Nevertheless, high water quality productions in Cyunyu and Mutobo WTPs were confirmed by both PerfScore and DWQI.

The cost of producing 1 m³ of treated water varied by plant and over time. Direct and indirect costs were considered. The direct costs included the costs for drawing raw data (water cost), chemical costs, energy and oil costs, staff costs, and other fees such as motor and vehicle repair fees, communication, supplies, and travel costs. Indirect costs included the overhead cost and depreciation, set at 30 and 1.5% of the total direct cost, respectively. The cost per m³ at Shyogwe-Mayaga treatment plant averaged Rwandan Francs (Frw) 85.60, while it was Frw 97.32 at the Kanyabusage plant. Unit production costs for Mutobo, Kadahokwa, and Cyunyu, known for high-quality water, ranked among the cheapest (Frw 130.76, Frw 255.63, and Frw 283.23 per m³). Ngenda in Bugesera District emerged as the most expensive water producer (Frw 574.16 per m³).

Appendix 1 highlighted annual changes in average production costs from 1 July 2017 to 30 June 2020. Cyunyu, Kimisagara, Mpanga, Mutobo, Ngenda, and Nyamabuye costs increased yearly. Conversely, Gihuma, Nzove, Gisuma, Kanyonyomba, and Shyogwe-Mayaga costs decreased annually. Gihira, Kadahokwa, and Karenge water production costs remained relatively consistent throughout the study period. Kanzenze WTP was not included in the analysis as it started operations after 2022.

It is noteworthy that a profitability analysis for each WTP was not conducted as part of this study. This omission was due to the fact that financial statements for the WTPs were prepared centrally at the WASAC headquarters, utilizing the data provided by each respective WTP. Moreover, the revenue collection process was centralized, handled by WASAC rather than the individual WTPs. This centralized approach to financial management and revenue collection underscored the need for a holistic understanding of the financial landscape at the individual plant level.

Rainfall was included in this analysis as a factor that could increase or decrease the volume of raw water (Wikipedia 2022). Overall, Rwanda has four climatic seasons, represented by the long rainy season (March to May) and the short rainy season (September to November). These seasons alternate with the long dry season (June to August) and the short dry season (December to February) (Climate Change Knowledge Portal 2021).

The average quarterly rainfall between 1 July 2017 and 30 June 2020 (Appendix 2) indicated that all treatment plants experienced rainfall throughout the fiscal year, especially in Q4 (140.91 mm quarterly average), Q1 (121.91 mm), and Q2 (134.23 mm). However, the rainfall was low in Q3 (43.96 mm on average). The rainfall might prevent the water resources from depleting. Therefore, it could be one of the drivers of the water supply in Rwanda and could affect the quality of raw water.

On average, during that period, and as per Appendix 2, water treatments located in the southern province had the highest average levels of rain. They were followed by those located in the western, northern, and eastern provinces, respectively.

Being unsafe for human consumption, raw water should be treated to remove contaminants. That is, the marginal effect of 1 m³ of raw water on treated water is about 0.96 m³. For every 1 m³ of raw water that is treated, other things being equal, there is a loss of 0.04 m³. Moreover, the marginal effect of 1 m³ of treated water on the supplied water is 0.99. In fact, less than 1% of treated water is not supplied because it is used or lost during the treatment process. The big loss happened during the supply of water in different parts of the country. Therefore, for an additional m³ of supplied water, its marginal effect on the billed water was 0.5835 (i.e., 58.35%). In other words, only 58.04% of the treated water or 55.62% of the raw water could reach the targeted beneficiaries.

Moreover, when accounting for the water production costs, some WTPs experienced greater efficiency gains than others. The random effects generalized least squares regression model (Appendix 3) was developed to examine whether there are significant differences in total costs per m³ (unit cost) of treated water among WTPs. This analysis considered factors such as quality scores (PerfScore) and the time of the year (month) to assess variations.

Considering the quality of water and the treatment time (month), the average production costs per 1 m³ of water in Kimisagara, Nzove, Kadahokwa, Cyunyu, and Gisuma plants did not differ significantly. However, the production cost per unit in Kimisagara plant was significantly higher than in Ngenda, Nyamabuye, Karenge, Kanyonyomba, and Rwasaburo by 337.38, 247.61, 193.20, 176.35, and 155.11 Frw, respectively. On the other hand, Kanyabusage, Gihira, and Mutobo WTPs recorded significantly higher average production costs of 196.54, 157.11, 114.77 Frw than Kimisagara WTP. Shyogwe-Mayaga WTP was not included in the analysis since it has many missing values. Appendix 3 indicates that compared to other months of the year in the fiscal year 2019/2020, the lowest production costs per 1 m³ were incurred in April and February. The costs incurred in October, November, December, and January were not significantly different from the costs incurred in April at a 5% significant level, which can be the case at a 10% significant level. The highest production costs were registered in July and August.

The volumes of the raw water and the quality of the water production are the key determinants of the water supply. As per Appendix 4, one additional cubic meter of raw water significantly increases the quantity of supplied water by 0.986 m³, while one additional unit of the PerfScore (i.e., when the quality compliance decreases) significantly reduces the quantity of supplied water by 127.92 m³, other things being equal. The production costs as well as the turbidity of raw water can also affect the volume of the supplied water at a 10% significant level. Accordingly, one additional NTU on turbidity is likely to reduce the volume of water supply by 4.38 m³, accounting for the effects of other variables in the model. Similarly, one additional Rwandan franc incurred in the water treatment process of 1 m³ is likely to increase the water supply by 90 cm³, other things being equal.

Appendix 4 also indicated that the average supplied water in Karenge and Kimisagara WTPs is not significantly different. However, the average volume of water supplied by Karenge WTP was 135,248.40 m³ lower than the water supplied by Nzove WTP but significantly higher than any other remaining plant.

The determinants of the quality of the water production are the turbidity of the raw water and the costs of the chemical inputs (Appendix 5). The water quality model indicates that one additional Rwandan franc spent on chemical products significantly increases the quality of water (reducing the score by 0.000000682) by controlling the effects of the turbidity of raw water and the individual characteristics of the WTPs. Similarly, one additional NTU of the turbidity of the raw water, controlling the effects of other variables of the model, significantly increased the quality score (thus reducing the quality of water) by 0.018.

Moreover, other things being equal, the quality of water (expressed in PerfScore) in Kimisagara, Kadahokwa, Gihira, Mutobo, Cyunyu, Mpanga, Muhazi, Nyagatare, and Kanyonyomba was not statistically different. It was significantly different from the quality of water produced in Karenge, Nzove, Gihuma, Rwasaburo, Nyamabuye, and Ngenda.

In Rwanda, 95.80 and 76.80% of households in respectively urban and rural areas have access to improved drinking water sources (Table 6). Globally, 82.30% of Rwandans have access to improved drinking water, which is a significant increase from the 72.00% access rate in 2012 (National Institute of Statistics of Rwanda 2012; Ministry of Finance & Economic Planning 2023). In urban areas, per Table 6, the western province (75.40%) and the southern province (78.40%) have lower access percentages than the City of Kigali (97.40%). The lowest percentages of access to improved water sources were seen in the districts of Nyamagabe (59.00%) in the southern province as well as in Rutsiro (61.00%) and Karongi (64.00%) in the western province (Ministry of Finance & Economic Planning 2023). The same source indicates that unprotected springs or wells account for the majority (11%) of private households' access to unsafe drinking water, followed by rivers and surface water (6.00%).

The improved sources of drinking water included water from pipes, public taps, tube wells or boreholes, protected springs or wells, rain, and bottled water (Ministry of Finance & Economic Planning 2023). Table 7 indicates the categories of customers that WASAC served as well as their numbers and percentage increased from June 2020 to June 2022. During that period, WASAC registered an increase of almost a quarter of the customers. From 230,190 as of 30 June 2020, WASAC reached 287,608 customers as of 30 June 2022 made of 263,708 residential connections (91.69%), 16,779 non-residential connections (5.83%), 6,799 public standpipes (2.36%), and 322 industrial connections (0.11%). Under such conditions, WASAC served only 27.35% of the 964,287 private households (residential connections) in Rwanda. Therefore, the remaining 54.95% of households that had access to improved water in urban areas could not be served by WASAC.

The private households that were not served by WASAC may have used wells, boreholes, protected springs, rain, or bottled water. Although spring water is seen as being aesthetically suitable for home use, the existence of poorly designed pit latrines, inadequate solid waste management, and poor spring protection may cause pathogenic microorganisms to contaminate the water (Haruna et al. 2005). Moreover, private wells can become contaminated by chemicals, naturally occurring toxic substances, or pathogenic organisms that can cause illness in children (Alan et al. 2023).

This section critically analyzes the research findings on water quality, reliability, and cost-effectiveness of the WASAC drinking water. Comparing results with existing knowledge, the discussion proposes targeted interventions for enhanced water service delivery.

The outcomes of this research underscored the pivotal significance of water quality management in Rwanda. The examination of WTPs' efficacy, gauged through both PerfScore and DWQI metrics, yielded profound insights into the nuanced variations in water quality across different facilities. Notably, Mutobo and Cyunyu WTPs consistently exhibited exceptional water quality, as indicated by both performance measures. Conversely, Gihuma WTP consistently fell into the category of lower water quality according to both assessment tools.

However, intriguingly, other WTPs, such as Kadahokwa, demonstrated divergent classifications – rated as having the second rank of clean water by PerfScore but considered of relatively lower quality by DWQI. Similarly, Kanyabusage held the top rank for water cleanliness according to DWQI, while it was categorized among the least clean water facilities by PerfScore. These disparities underscore the imperative for in-depth investigations into the operational dynamics and water quality management practices of each WTP. Such scrutiny is essential to gain a comprehensive understanding of the factors influencing water quality and to formulate targeted strategies for improvement where needed.

A meticulous analysis of factors influencing water quality, including chemical costs and raw water turbidity, contributed granularity to the extant knowledge base. However, it is pertinent to note that the study did not statistically establish disparities in water quality across regions or temporal variations. The regional and temporal variables were not included in the model because they were not adding any value to the analysis. As an illustration, the water dispensed in Kigali City, emanating from three distinct WTPs – Kimisagara, Karenge, and Nzove – exhibited statistically significant differences.

This underscored the necessity for targeted interventions in accordance with recommendations articulated in the prevailing scientific literature (Wyatt 2010; Huttinger et al. 2015; Karamage et al. 2016; Kirby et al. 2016; Mukanyandwi et al. 2019; Herschan et al. 2023). Such nuanced examinations are imperative to elucidate the intricate determinants affecting water quality and to guide targeted measures for enhancement, aligning with the prevailing scientific discourse.

This research has illuminated the challenges associated with maintaining a reliable and consistent water supply, evident in the observed fluctuations in water production, supplies, and NRW. The quarterly data analysis has drawn attention to significant seasonal and spatial variations in water supply, highlighting the necessity of implementing effective strategies to address demand fluctuations and ensure sustained water reliability. However, the random effect model of the water supply did not account for seasonal variation and rainfall due to statistically significant collinearity with both raw water turbidity and raw water volumes.

As a result, raw water volumes and turbidity variables were kept in the model and they were proven to be statistically significant, holding other factors constant. This was further affirmed by the highest and lowest levels of water production and NRW observed during the dry season (third quarter of the year when the rainfall is at its lower level). Consequently, the quantity of water produced and supplied per customer reached its peak during this period.

However, the levels of water produced and supplied were at their lowest during the first or second quarter of the year, corresponding to the rainy seasons. During the rainy season, raw water quality suffered due to increased turbidity and contamination. Producing low quality of raw water under these circumstances might be costlier, and its extraction and treatment processes could be challenging, potentially impacting the quantity of supplied water. Therefore, these findings emphasize the importance of considering both temporal and spatial factors in the development of comprehensive water management strategies to enhance reliability.

The examination of water production costs per treatment plant provided a comprehensive understanding of cost variations among different WTPs. While existing literature has recognized the importance of cost optimization for the affordability of water supply services (Wyatt 2010), this research added specificity by identifying the most cost-effective and expensive WTPs. The variation in production costs over time, as presented in the study, underlined the dynamic nature of cost management strategies employed by different plants. For example, Shyogwe-Mayaga, Kanyabusage, Mutobo, and Gihira plants consistently maintained lower production costs, averaging between $0.086 and $0.133 per m³ from 1 July 2017 to 30 June 2020. In contrast, Nyamabuye, Ngenda, and Kanyonyomba incurred higher costs, ranging between $0.481 and $0.574 per m³ during the same period.

Although the model of water production cost per cubic meter did not incorporate spatial variations due to the lack of relevance in the corresponding variables, as indicated by the Akaike information criterion and Bayesian information criterion, a noteworthy geographical pattern emerged. All WTPs with the highest production costs, as listed above, are situated in the eastern province. Conversely, two out of the four WTPs listed among the lowest production cost per cubic meter, namely Kanyabusage and Gihira, are located in the western province. Additionally, Mutobo is positioned in the northern province, while Shyogwe-Mayaga is situated in the southern province.

The research strongly advocates for targeted interventions to address the identified challenges in water quality, reliability, and cost-effectiveness. These interventions align with previous studies suggesting that reducing water losses can significantly contribute to improving water availability in the WASAC operational area (Karamage et al. 2016). The implementation of a Supervisory Control and Data Acquisition (SCADA) system and the ongoing geographic mapping project are highlighted as key initiatives to enhance monitoring, reduce leakages, and improve overall efficiency (WASAC Ltd 2019).

The study emphasized the importance of community engagement and education as part of targeted interventions. By involving communities in water safety practices and promoting awareness of safe water storage and hygiene, the research suggests a holistic approach to improving water quality and reliability. The proposed capacity-building initiatives and the integration of ongoing projects, such as the intelligent water system, into these efforts, demonstrate a forward-looking approach to addressing water-related challenges in Rwanda.

In relation to the WASAC mandate to enhance the water utility's viability and autonomy (WASAC Group 2023), this research identified significant challenge that must be addressed. The absence of regular profitability reports per WTP poses a hurdle in assessing whether these plants operate sustainably or incur losses. Despite the finance department's capability to produce such reports on demand, the complexity of the process and the requirement for additional resources hinder seamless execution.

Water production costs, recorded daily and reported weekly to the corporate planning unit, create difficulty in accurately associating billed or lost water with the production day, month, or WTP where treatment occurred. The lack of a proper management information system further complicates efforts to match revenues with the timing and location of water supply and treatment.

Applying a depreciation cost of 1.5% to total direct costs, rather than acquisition costs or market values, may compromise the capacity to maintain and upgrade infrastructure, particularly for small WTPs. This creates a continual dependence on the Government of Rwanda for technical support, oversight, and funding to cover production costs and infrastructure maintenance.

Additionally, WASAC's involvement in rural water initiatives, deviating from its mandate to serve urban and peri-urban areas, introduces complexities. The implementation of donor-funded development projects and rural sector services through the Single Project Implementation Unit (SPIU) raises questions about the impact on both rural and urban water production and supply. The management of rural water infrastructure by private operators, under contracts with districts and RURA, while WASAC offers technical support, adds a layer of complexity.

This research could not conclusively determine the positive or negative impact of WASAC's involvement in rural water management on overall water production and supply. There is a need to evaluate whether the private–public partnership strategy employed in rural water management could serve as a viable solution for both urban and peri-urban water supplies. This evaluation would shed light on whether entrusting WASAC with the management or oversight of rural water would result in improved outcomes.

This research has provided a comprehensive analysis of water quality, reliability, and cost-effectiveness in Rwanda's urban and peri-urban areas. The findings underscore the critical role of continuous monitoring and adjustment of treatment processes to ensure consistent water quality. Mutobo and Cyunyu WTP demonstrated exemplary performance in maintaining high water quality, emphasizing the importance of targeted interventions to enhance water service delivery.

The challenges identified in maintaining water reliability highlight the need for strategies to manage demand fluctuations and reduce water losses. Moreover, the examination of production costs among different WTPs emphasizes the dynamic nature of cost management strategies. Finally, based on the research findings, several recommendations are proposed to enhance water management in Rwanda.

Firstly, there is a pressing need for the implementation of targeted interventions to address the challenges identified in water quality, reliability, and cost-effectiveness. This includes the NRW management through the continued investment in technology, such as smart metering, to improve efficiency.

Secondly, WASAC should focus on strengthening its financial autonomy by addressing the challenges associated with the production of WTP profitability reports and the application of depreciation costs. The development of a robust management information system would facilitate accurate tracking of water supply, billing, and treatment, contributing to improved financial viability.

Thirdly, the Government of Rwanda and WASAC should critically evaluate the impact of WASAC's involvement in rural water management, considering its mandate to serve urban and peri-urban areas. This assessment should explore the potential benefits and drawbacks of extending WASAC's oversight to rural water infrastructure.

Lastly, community engagement and education should be integral components of water management strategies. Public awareness programs on water safety practices, safe storage, and hygiene can contribute significantly to improving water quality and reliability. Integrating ongoing projects, such as the intelligent water system, into these community-focused initiatives can enhance their effectiveness.

This work could not have been possible without the financial support of the African Center of Excellence in Data Science (ACEDS) of the University of Rwanda. The same gratitude also goes to WASAC that accepted to be the case study and, therefore, to provide data used in this article.

All the authors have given their consent for this publication, and believe that the research aligns with the journal's focus and meets all submission guidelines. The consideration for publication is highly appreciated.

J.M. contributed to conceptualization, data curation, formal analysis, methodology, project administration, resources, investigation, visualization, and also wrote the original draft, reviewed, and edited the manuscript. C.W.K. contributed to project administration, supervision, validation and also wrote, reviewed, and edited the manuscript. C.R. contributed to project administration, supervision, validation and also wrote, reviewed, and edited the manuscript.

The research was funded by the University of Rwanda, African Center of Excellence in Data Science (ACE-DS), College of Business and Economics (CBE) Gikondo, Kigali, Rwanda.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.