Makongo area is located in the north-western 6˚45′27″S, 39˚11′42.72″E, urban zone of Dar es Salaam city, about 17 kilometers from the city center, as shown in Figure 1. The area is growing rapidly with residential and commercial buildings in Dar Es Salaam City. The topography of Makongo includes gentle slopes and a plateau. It also has large valleys and streams which drain water into the Mbezi river. These valleys inhibit close movement and increase the cost of transportation. Among the different methods of population projection techniques, the study considered the overall current situation of the targeted Makongo area. The population of Makongo is approximately 35,567 according to [17]. The mean yearly temperature recorded is 26.1˚C and approximately 1114 mm of precipitation descends.

Figure 1. Map of Makongo showing the study area.

Currently, Makongo town was supplied from Makongo booster station, which supplied an average capacity of 1029716.00 m³/year, but the delivered water to the customers is 666207.00 m³/year as per the Dar es Salaam Water Supply and Sanitation Service ([6]). The combined water supplied, consumed, and lost in Makongo is presented in Table 1.

Table 1.Combined water consumed, supplied, and water loss.

Source: ([6]).

Source: Open Street Map (Google Earth).

Figure 2. Existing Makongo and Goba water supply distribution networks.

The Water Supply Distribution Network (WSDN) of Makongo town is an integrated system designed to deliver water to consumers through a combination of branched and looped pipelines, extending approximately 188.55 kilometers in total length. The network originates at a pump station, which conveys water to a reservoir, from where it is distributed to consumers throughout the town.

The pipelines are constructed from three main materials: uPVC (unplasticized Polyvinyl Chloride), a corrosion-resistant plastic; HDPE (High-Density Polyethylene), a flexible and durable plastic; and steel, which offers high strength but requires protective treatment to prevent rusting. Pipe sizes vary according to function, with Nominal Diameters (DN) ranging from 50 mm for smaller local distribution lines to 400 mm for major transmission mains. This range ensures efficient water flow and reliable service to meet the town’s demand (see Figure 2).

Extended period simulation analysis was used to provide a specific pattern for giving demands with respect to hourly and daily variation in water demands, as shown in Equation 1.

where PF is the peak factor in a daily hour demand; Q_max is the maximum hourly demand (m³/hr); and Q_ave is the average daily demand (m³/hr).

The hydraulic model of the Makongo Water Supply system was developed in EPANET by translating physical network components and operational parameters into a digital simulation environment. The process integrated GIS data, CAD drawings from DAWASA, and field survey information. Pipeline maps and layouts in hardcopy or PDF formats were georeferenced in ArcGIS to align with topographical and satellite data. Where gaps existed, network elements were digitized manually in ArcGIS using high-resolution imagery and GPS data. This established the baseline model for subsequent calibration and hydraulic analysis.

Table 2. Lengths of pipes in the model before and after skeletonization.

Pipe skeletonization, following [10] guidelines, was applied to simplify the network while preserving hydraulic integrity. The criteria required retention of at least 50% of total pipe length, 75% of pipe volume, and all major transmission and distribution lines (DN50-DN400). Smaller service connections were excluded. This process reduced the modeled network length by 37,994 m (Table2). This level of simplification is acceptable as it strategically removes hydraulically insignificant elements while conserving the network’s core flow paths and storage capacity. As demonstrated by [20], such a skeletonized model can accurately replicate the hydraulic behavior of the original full network for planning-level analyses.

The EPANET hydraulic model of the Makongo distribution network was developed through a systematic process that translated both field and secondary data into simulation-ready inputs. The network layout was digitized using GIS and CAD drawings obtained from DAWASA, which were georeferenced against topographic maps and high-resolution satellite imagery. Junctions, reservoirs, and pumping stations were positioned using surveyed GPS coordinates, while pipeline alignments were verified through field reconnaissance.

Key hydraulic parameters were defined using a combination of literature values and locally available operational records. Pipe roughness coefficients were assigned by material type according to recommended Hazen-Williams values: 140 for HDPE, 150 for uPVC, and 120 for steel. Nodal demand was estimated from DAWASA’s 2024 consumption records and projected household and commercial demands, adjusted with diurnal peak factors (morning = 1.45, evening = 1.20, off-peak = 0.65) to reflect actual usage patterns. Pump curves were generated from DAWASA operational data, with an average head-discharge slope of 0.82 m/(L/s). Reservoir head levels were fixed at 85.4 m (R1) and 78.7 m (R2), based on field measurements. Junction elevations were derived from an open-source (SRTM) 30-meter Digital Elevation Model (DEM).

This systematic parameterization ensured that the baseline model was both physically representative and hydraulically realistic. Calibration and validation were then undertaken as described in the following section.

To ensure the reliability of the EPANET hydraulic model used in analyzing the Makongo water distribution network, a calibration exercise was undertaken.

Model calibration was performed based on the results of the model hydraulic parameters; pressure measured in the selected nodes was used for calibration ([25]). This process involved comparing simulated pressure values generated by the model with field-measured pressure data collected at selected junction nodes within the network.

To validate the EPANET model, a separate dataset independent of the calibration process was collected through field measurements at selected junction nodes. Ten (10) representative nodes, J-1263, J-803, J-862, J-1205, J-1870, J-1770, J-1624, J-1804, J-1862, and J-2129, were monitored for pressure values using portable pressure gauges. Measurements were taken at two daily intervals, at 09:00 AM high demand period and 02:00 PM low demand period, for five consecutive days to improve the quality of the results, covering both peak and off-peak conditions. These nodes were chosen because they represent a range of elevations, demand categories (domestic, commercial, institutional), and hydraulic zones within the network, making them suitable for assessing the model’s predictive accuracy across different conditions.

The measured pressures were then compared with simulated pressures generated by the EPANET model under the same boundary conditions. Statistical indicators including the coefficient of determination (R²), Root Mean Square Error (RMSE), and Mean Square Error (MSE) were used to evaluate the model’s performance. By relying on independent field data, the validation process ensured that the model was not only calibrated to fit known conditions but also capable of accurately predicting system behavior under different operating scenarios.

The model statistical hydraulic parameters’ performance was evaluated using objective functions, including the coefficient of determination, using Equation (2) and Equation (3).

All measured and simulated pressure values obtained from the EPANET model and field surveys were exported into Microsoft Excel for statistical analysis. The following statistical techniques were applied:

2.7.1. Coefficient of Determination (R²)

R² was computed to evaluate the strength of the linear relationship between measured and simulated pressures. Using Excel’s built-in regression analysis tool, the correlation coefficient (R) was first obtained by plotting simulated versus measured pressures and fitting a linear trendline. The square of this correlation coefficient gave the R² value, which indicates how much of the variability in the measured data is explained by the model.

2.7.2. Mean Square Error (MSE)

MSE was calculated to quantify the average squared difference between the measured and simulated pressures. In Excel, this was implemented using a formula column where each error term (x_i − y_i)² was computed, summed, and divided by the number of observations (n).

and Root Mean Square Error (RMSE) is a frequently used measure of the differences between values predicted by a model or an estimator and the values observed. It is a standard way to measure the error of a model in predicting quantitative data.

The mathematical equation for RMSE is:

All evaluation statistics are embedded in the EPANET environment for calibration purposes. Calibration is a critical step in hydraulic modeling, as it enhances the model’s ability to accurately reflect real-world conditions and increases confidence in subsequent simulations used for planning and operational decisions ([19]).

3.1.1. Water Balance

The existing water balance in the study area was assessed by calculating the difference between water production and consumption. The total water loss in Makongo’s Water Supply Distribution Network accounts for 13.9% of total production. This relatively low loss is attributed to the system’s recent construction in 2018. At the beginning of a system’s design life, minimal water loss is expected; however, losses may gradually increase over time unless minor leakages are promptly repaired and a routine maintenance program is established and consistently implemented.

For comparison, [11] reported average water losses of approximately 49% in Dar es Salaam’s older infrastructure, primarily due to aging pipelines, illegal connections, and poor maintenance. The 13.9% loss in Makongo is significantly lower, likely reflecting the newer infrastructure. According to the World Bank ([9]), well-maintained systems typically experience NRW levels of 15% - 20%, whereas developing countries often exceed 30% - 40%. The Makongo figure of 13.9% falls below this benchmark, indicating efficient operation and recent commissioning.

[16] found water losses of 30% - 50% in urban Ugandan systems, recommending early leak detection, real-time monitoring, and district metered areas (DMAs) for control. This supports the observation that routine maintenance and timely repairs are critical to sustaining efficient operation as the system ages. Thus, while the Makongo system currently performs well, its long-term efficiency will depend on the consistent application of maintenance, leak monitoring, and NRW management practices.

3.1.2. Network Layout

The existing network was digitized into EPANET and consisted of 1253 nodes, 1406 pipes/links, 1 pump station, and 2 storage tanks. Pipe diameters ranged from 50 mm to 400 mm, and pipe materials included HDPE and steel. Water demand was assigned to nodes based on a per capita consumption rate of 110 liters per capita per day (lpcd) and a population density of 350 persons/hectare. These demand values were distributed across the network based on demographic and land use data from [17]. As shown inFigure 3, the existing Makongo water supply network consists of 1253 nodes, 1406 pipes, one pump station, and two storage tanks.

Figure 3. Existing Makongo water supply network (as built).

To evaluate the performance of the Makongo water distribution system, a hydraulic simulation was conducted to examine how water demand patterns influence pressure, flow, and overall network reliability. While steady-state analysis provides a snapshot of hydraulic conditions at a specific moment, it does not account for daily fluctuations in consumption. Therefore, an Extended Period Simulation (EPS) was employed to model the system over 24 hours, capturing the diurnal variations in water demand observed in Makongo. This approach allowed for assessment of peak demand impacts on service delivery, identification of potential low-pressure zones, and support for operational optimization and water loss reduction.

Model Calibration and Validation

The hydraulic model of the Makongo water distribution system was validated by comparing measured and simulated pressures at 10 key junctions J-1263, J-803, J-862, J-1205, J-1870, J-1770, J-1624, J-1804, J-1862, and J-2129. Measured pressures ranged from 36 to 98 meters, while simulated pressures varied from 21.44 to 120.81 meters. The largest deviation was 20.15 meters at Junc J-1870, and the smallest was 4.7 meters at Junc J-803 (Figure 4). These differences fall within acceptable engineering limits. The close agreement indicates that model parameters including pipe roughness, demand patterns, and elevations were appropriately defined and adjusted during calibration.

Figure 4. Simulated and measured values.

To further evaluate model performance, a correlation analysis was conducted between measured and simulated pressures at selected junctions (Table 3). The coefficient of determination (R²) was calculated as 0.89, while the Mean Squared Error (MSE) and Root Mean Square (RMS) were 734 and 10.26 m, respectively. These results indicate a strong agreement between observed and simulated values, confirming the reliability of the model for further analyses, including system optimization, scenario testing, and the assessment of intermittent supply impacts.

Table 3. Calculation of root mean square as per Equation (2).

Figure 5 illustrates the scatter plot of measured versus simulated pressures, with a fitted trendline. The strong linear relationship, with R² = 0.89, indicates that over 89% of the variation in simulated pressures is explained by measured values. The points closely align along the 1:1 trendline, demonstrating minimal deviation and high consistency between field measurements and model outputs.

Comparable results have been reported in other hydraulic modeling studies. [13] achieved an R² of 0.95 during EPANET-based calibration of a municipal water network in Australia. [23] reported R² values exceeding 0.96 in automatic calibration studies, while [18] recorded R² values of 0.97 in real-time modeling. These findings reinforce the robustness of the current calibration process and its alignment with internationally accepted practices.

Overall, the calibration and validation confirm that the model reliably represents the Makongo water distribution network and can be confidently used for performance analysis, operational planning, and water distribution efficiency assessments.

Figure 5. Correlation analysis was conducted between measured and simulated pressure.

The hydraulic performance of the existing water supply system was analyzed by evaluating pressures and velocities throughout the network. The skeletonized network considered in this study includes 1404 pipes of various materials, 1250 junctions, 5 pumps, and 2 water source reservoirs, as shown in Figure 6. Following model calibration, current water production was assessed in relation to the observed demand patterns in the Makongo and Goba areas.

The results align with previous studies. [15], using EPANET to evaluate a Portuguese urban water network, observed low-pressure zones at high-elevation nodes and high velocities in undersized or corroded pipes. [8] highlighted that growing urban water distribution networks often experience increased pressure variability, necessitating pipe re-sizing and re-zoning to maintain reliability. In Dar es Salaam, [11] reported that pressure loss and pipe friction from aging infrastructure significantly impacted water delivery, consistent with the findings in Makongo-Goba. Furthermore, [3] demonstrated that optimizing pipe diameters and pressure zones through hydraulic modeling can reduce water losses by 12% and achieve 15% energy savings, underscoring the importance of pressure-velocity analysis as conducted in this study.

Figure 6. Map of pressure in the existing distribution system at peak hour demand.

Table 4. Hydraulic pressures before optimization.

The hydraulic model results for node pressures during peak hour demand, presented in Table 4, provide critical insights into the performance of the existing water supply system. By comparing allowable, minimum, and maximum pressures at each junction, the analysis identifies areas where pressure may fall below or exceed recommended limits, highlighting potential service reliability issues. Maintaining pressures within the acceptable range is essential to ensure adequate water delivery to consumers, prevent pipe damage from excessive pressures, and minimize water losses. The peak-hour assessment specifically reflects the network’s ability to meet the highest demand periods, indicating whether pumps, reservoirs, and pipeline capacities are sufficient. These results serve as a basis for identifying low-pressure zones, prioritizing network upgrades, and optimizing operational strategies to enhance system performance and ensure consistent, safe water supply throughout the service area.

It was observed that 493 nodes (39%) in the existing water supply system satisfied the recommended pressure range of 15 - 60 m, as specified in the Ministry of Water Design Manual ([14]). [7] emphasize that high-pressure zones (>60 m) can lead to pipe bursts, joint failures, and increased leakage. Even a small proportion of over-pressured nodes, such as the 42 nodes (3%) identified in this system, can significantly amplify water losses if left unmanaged. Conversely, a substantial number of nodes 757 (61%) were found to be below the recommended pressure range. This indicates that large portions of the study area, particularly the low-pressure nodes, may not receive sufficient water during peak hour demand (PHD), notably around 9:00 AM.

The hydraulic model results for pipe velocities at PHD are presented in Table 5, showing the allowable, minimum, and maximum velocities in the network. These results are critical for identifying areas with potential flow constraints, assessing the risk of pipe erosion or sedimentation, and supporting decisions for network optimization and operational management.

Table 5. Hydraulic velocity before optimization.

Studies of urban water distribution systems (WDSs) in developing countries have shown that under-designed networks often suffer from low velocities and non-uniform pressures, leading to service disruptions and low efficiency ([21]). They highlighted the importance of optimizing pipe diameters and upgrading critical network links.

Analysis of the Makongo distribution network indicates that the majority of pipes approximately 89% operate at very low velocities below 0.6 m/s. This pattern is typical in systems with intermittent supply, where long periods without flow result in minimal water movement in dead-end or low-demand areas once supply resumes. Such conditions increase the risk of sediment buildup, stagnation, and microbial growth. About 7% of the pipes, mainly in main distribution lines or high-demand areas, exhibit moderate velocities between 0.6 and 1.5 m/s, sufficient to maintain water quality. A small fraction, roughly 4%, experiences high velocities above 1.5 m/s, typically during peak usage or in critical pipeline segments.

The model results revealed that the existing system’s pressure and velocity distribution are insufficient to meet the full water demand of the study area. Low velocities contribute to inadequate flow, siltation, and increased pipe wear, which exacerbate leakage and pipe bursts, further reducing water delivery to consumers. These findings underscore the need for optimizing pipe diameters and re-evaluating the network design to improve flow, reduce non-revenue water, and ensure reliable water supply.

The optimization of the pipe diameters of the Water Supply system was carried out based on commercially available pipe sizes. The optimized pipe diameters skeleton in the study area is illustrated in Figure 7, whereby pipe optimization was undertaken during the PHD (9:00 AM) at the nodes, since it satisfied the required demands in the distribution systems, which means that other water demands were also fulfilled ([1]).

All hydraulic parameters before and after optimization were analyzed using nonparametric tests since the data populations were not normally distributed. The Wilcoxon signed-rank test was applied, and the results showed statistical significance (p < 0.05) for pressure. This indicates that the optimization process had a measurable impact on these parameters. The significance of the optimization was further analyzed based on desired performance criteria and their relative importance, as discussed in subsequent sections. Figure 7 illustrates the optimized pipe diameter skeleton of the study area’s water supply system.

Figure 7. Optimized pipe diameter of the study area water supply system.

Pressure

A statistical comparison of node pressures before and after optimization was conducted using the Wilcoxon signed-rank test, with the results presented in Table 6. This non-parametric test was selected to evaluate whether the optimization process produced significant changes in pressure distribution across the network. The results indicate a statistically significant improvement in pressures (p < 0.05), confirming that the optimized pipe diameters effectively increased low-pressure node values and reduced the occurrence of over-pressured nodes. These improvements enhance the overall reliability of the water supply system, ensuring that all areas receive adequate pressure during peak demand periods and reducing the risk of pipe bursts or excessive leakage. Consequently, the statistical analysis validates the effectiveness of the optimization strategy in improving hydraulic performance and supporting sustainable water delivery.

Table 6. Wilcoxon signed-rank test results for pressure at nodes.

Using 20 critical nodes, the sum of positive ranks (W⁺ = 210) far exceeded the sum of negative ranks (W⁻ = 0), with a test statistic W = 0, well below the critical value of 52 at α = 0.05. Normal approximation yielded a Z-score of −3.92 and a p-value < 0.001, confirming that pressure increases were not due to chance.

3.5.1. Optimized Pipe Diameters

Pipe diameter optimization for the Makongo water supply system was conducted using a simulation-based approach with EPANET with objective function of minimize head loss, maximize pressure with key constraints of minimum pressure of 5 m, minimum velocity of 0.6 m/s. Commercially available pipe sizes were evaluated to meet peak demand while maintaining pressures within 20 - 80 m and velocities between 0.6 - 2 m/s, and to minimize construction and operational costs. Multiple iterative simulations were performed using EPANET’s optimizer, and each configuration was assessed against hydraulic constraints.

The final optimized network, shown in Figure 7, successfully met peak hour demand at 9:00 AM, ensuring reliable supply throughout the day. Statistical analysis using the Wilcoxon signed-rank test confirmed significant pressure improvements (p < 0.05), demonstrating that the optimization process effectively enhanced overall network performance.

3.5.2. Pressure and Velocity Improvements

A detailed statistical evaluation of nodal pressures (Table 7) confirmed the effectiveness of optimization. 95% of the nodes after optimization had a pressure within the recommended range (15 - 60 m). These improvements enhance network reliability during peak demand, reducing the risk of pipe bursts, overpressure, and non-revenue water (NRW).

Table 7. Pressure at the Nodes after optimization.

Velocity analysis showed reductions in high-flow sections and increased flow in low-velocity segments. Approximately 12% of pipes achieved improved velocities, helping prevent stagnation, reduce siltation, and maintain water quality. Statistically, these velocity improvements were validated (Z = −6.13, p < 0.00001), demonstrating that diameter optimization and layout adjustments contribute to better hydraulic performance and operational efficiency.

Comparative studies reinforce these results. [22] observed pressure improvements of up to 40% at key nodes during peak hours, while [4] reported a 12% reduction in pipe velocities post-optimization, improving energy efficiency and reducing pipe failure risk. [5] found that increasing pressure by 10 m at low-pressure nodes reduces water loss by 6% in aged systems. The Makongo study confirms these findings and couples hydraulic modeling with rigorous statistical validation, providing a scientifically robust benchmark for water distribution optimization in Tanzanian urban contexts.