Improving safe sanitation practices using groundwater transport modelling and water quality monitoring data

Contamination of groundwater as a result of poor sanitation is one of the major threats to drinking water resources throughout the world. The problem is exacerbated due to unpredictable rainfall patterns, prolonged dry seasons, and overexploitation of groundwater resources (Haddeland et al. 2014). The urban areas of many developing countries are at high risk due to a high population growth rate, increased urbanization, and abject poverty, along with poor sanitation and inadequate water supplies, which together result in the spread of waterborne diseases (WHO 1992). As the population in developing countries continues to increase, it is expected that the global reliance on groundwater and pit latrines will also increase (Graham & Polizzotto 2013; Ravenscroft et al. 2017).

In densely populated Dar es Salaam, where this study was conducted, the main source of city water is the Ruvu River. However, the flow in the Ruvu River is unreliable and the city is unable to supply water to all residents due to the decrease in water levels (Ngoye & Machiwa 2004; Alphayo & Sharma 2018; Miraji et al. 2019). As a result, more than 50% of residents rely heavily on groundwater for their daily use. It is estimated that annual withdrawals from the aquifer exceed 69 million m³ (Mtoni et al. 2011).

This study was conducted in one of the peri-urban areas of Dar es Salaam, Tanzania. As water is scarce, residents of this area rely on multiple water sources for household consumption (Ngasala et al. 2018). One of the most commonly used water sources is domestic wells, often privately or publicly owned, from which people can purchase water. The average density of wells in this area is 4.6 wells/km². The other sources include street vendors and treated surface water from Dar es Salaam Water and Sewerage Corporation (DAWASCO). Residents of the study area are low-income earners who spend a significant portion of their income to purchase water for domestic use and wastewater management (Ngasala et al. 2019b). While the cost of purchasing water from private well owners is high, residents often have few options other than this or purchasing water from street vendors. Water from most domestic wells in the study area is contaminated by household sewage due to the close proximity of the wells to poorly constructed septic tanks and pit latrines (Ngasala et al. 2019a). About 90% of the homes use unimproved pit latrines, which are not constructed to meet the definition of a properly engineered and constructed pit latrine. See Appendix A in the Supplemental Materials for more details.

Ngasala et al. (2019b) linked the cross-contamination of domestic water with practices involving mixing during collection and at the point of use in this area. The study focused on the water quality analysis of three different water sources used by the community (city water, domestic wells, and water vendors) and found that domestic wells have the highest level of contamination. Ngasala et al. (2019a) found that more than 65% of the drinking water wells were located within 15 m of a septic system, 18% were within 25 m, and 17% were within 35 m.

There is widespread use of groundwater modelling in hydrologic sciences, however, their use is less common in many developing countries. Very few studies have used groundwater models to estimate setback distances based on specific site conditions such as soil type and aquifer properties with the rate of transport of microbiological and chemical contaminants in the aquifer as suggested by Graham & Polizzotto (2013). As multiple contaminants such as bacteria, viruses, nutrients, and suspended solids relevant to human health are typically of concern, a holistic assessment calls for multi-component groundwater transport modelling. An additional complicating factor is a need for integrated modelling across multiple domains to include relevant processes such as runoff, infiltration, and transport in the vadose zone before contamination moves to the fully saturated groundwater domain.

While significant progress has been made in the development of integrated transport models (e.g., Bradford et al. 2014; Niu & Phanikumar 2015; Dwivedi et al. 2016), such efforts tend to be expensive and time-consuming due to the need for large datasets, computational time, and detailed site characterization. A novel aspect of the present work is that it combines groundwater flow and tracer transport simulations with water quality monitoring data collected using low-cost sensors to identify site-specific separation distances and to identify best practices for future developments in the region. Tracers are used to identify aquifer transport processes, characterize flow directions and travel times and quantify rates of mixing within subsurface systems. It is important to note that, simulated tracer concentrations were not directly used to identify distances but instead they were linked with observed levels of contaminants (E. coli, nitrate, and total dissolved solids (TDS)); therefore, the effects of adsorption, decay and all other relevant processes are already included in the data. A simulated tracer was used to link transport from the pit latrines (source of the tracer) to the drinking water wells and by definition, a conservative tracer is not degraded or retarded in the aquifer.

Several factors can contribute to contamination at a drinking water well, including transport from nearby sources of contamination (e.g., pit latrines) and contamination of the well casing by surface water (e.g., due to an improperly sealed well casing or casing at a non-complying depth). By following the evolution of tracer plumes originating from pit latrines using a numerical transport model and developing relationships between observed water quality and simulated tracer concentrations, contamination originating from nearby sources can be quantified while still relying on local conditions such as soils and geology to obtain data that can be used to drive decision-making. The approach can be used to identify setback distances for an acceptable level of risk by using well-known and well-established relationships between levels of fecal indicator bacteria such as E. coli and enteric pathogens (Sokolova et al. 2012; Korajkic et al. 2018, Holcomb & Stewart 2020).

In this work, the goal was to develop and demonstrate a method that could be used to determine site-specific separation distances based on the available information (i.e., soil type and aquifer properties) that can be obtained from well boring logs and the literature. This would be more protective of public health than relying on national and international standards and guidelines. We have expanded upon the work of Ngasala et al. (2019a), by using a numerical tracer and to determine its association with distances and measured contaminant levels (nitrate, E. coli, and TDS). We then compared our results with various well separation distances from the literature (Tables 1 and S1 in Supplemental Materials).

The study site is located in Dar es Salaam city, Tanzania (Figure 1). The area is densely populated (population density of 14,250/km²), with an average temperature of 22 °C and an average rainfall of 117 and 253 mm for short and long rains respectively (Mtoni et al. 2012). The site was surveyed to identify existing domestic wells and wastewater systems (pit latrines) and collect information such as construction method, size, depth, age of pit latrine, and sewage discharge schedule. Sixty-three domestic wells were located in the study area (as shown in Figure 1). The distances between each well and the closest pit latrine were measured. Field investigations revealed that out of 63 wells identified, 56% of domestic wells were located less than 15 m from pit latrines. The furthest distance was found to be 35 m and the shortest was 3 m with an average value of 17.1 m. Additionally, 24% of all wells were shallow wells with a depth of less than 15 m, and the rest were deep wells up to 120 m. Water samples were collected from 63 wells during the dry season (June–August 2016). The concentrations of nitrate, TDS, and E. coli were determined in all water samples obtained from the wells. Details of the sampling and analysis are provided in Appendix A of the Supplemental Materials.

GPS coordinates of each pit latrine and domestic well locations were recorded for mapping and modelling purposes. Additionally, we obtained official well records from the Drilling and Dam Construction Agency (DDCA) that included soil profiles and well logs. Groundwater recharge rates and rainfall patterns were obtained from the literature (Mjemah et al. 2009; Mtoni et al. 2011). Hydraulic head and groundwater flow direction data were calculated using the well log information provided and digital elevation model (DEM) data. Wells were divided into two groups as the water in the shallow and deeper aquifers tend to follow different flow paths (Wilcox et al. 2010). The two groups are wells that are less than 15 m deep (shallow) and wells that are more than 15 m deep (deep).

Two main aquifers underlie the city: an upper unconfined sand aquifer and a lower semi-confined sand aquifer that are separated by a clay aquitard (Mtoni et al. 2012). The majority of the city including the study area overlays a clay aquitard and a sand aquifer, which are the main sources of groundwater (Mjemah et al. 2009). Detailed descriptions of the hydrogeology of the study area and the three-dimensional conceptual model used are included in the Supplemental Materials (Figure S1).

MODFLOW-2000 (Harbaugh et al. 2000) and MT3DMS (Zheng & Wang 1999) were used to simulate flow and conservative solute transport in the fully saturated groundwater domain within the study area. Figure S2 shows the step-by-step flowchart procedure that was used for groundwater simulation. A MODFLOW numerical model was constructed from the conceptual model using appropriate aquifer properties, physical boundaries, and data from field observations. The conceptual model and grids were the same for both the groundwater flow and the transport models. The study area covered was about 36 km² and a horizontal model grid of 100 (x) 100 (y) cells was used with three layers in the vertical z-direction. The boundary conditions for the upper aquifer were the constant head boundaries in the north, east, and south of the domain associated with river stages. No flow boundary conditions were used in the western part of the domain.

Input parameters for running the steady-state MODFLOW simulation include time steps, sources, and sinks such as rivers, wells, recharge, hydraulic parameters, and boundary conditions. Three layers were simulated; fine sand (thickness=60 m), clayey sand (thickness=20 m), and sandy clay (thickness=30 m). The aquifer materials for each layer are classified as described in the Hydrogeology section above and as shown in Figure S1 in the Supplemental Materials. The average hydraulic conductivity values used for fine sand, clayey sand, and sandy clay were 15.14, 0.15, and 0.0022 m/day respectively, with a porosity value of 0.3 for all three layers (Roberts 1984; Mjemah et al. 2009; Sarki et al. 2014). Vertical and horizontal hydraulic conductivity anisotropy values were estimated to be 4 and 1 respectively (Mualem 1984; Chapuis & Gill 1989). The starting head was set at 60 m. The average annual recharge value for Dar es Salaam is 184 mm/year (0.0005 m/day) (Mtoni et al. 2011). Hydrogeological properties used were soil types, hydraulic conductivity, porosity, and observed hydraulic heads. A summary table of input parameters is provided in Supplemental Materials (Table S2).

Lastly, the model was run to simulate groundwater heads and flows. Model calibration was performed using the observed heads obtained from field data. More than 90% of all wells used for this study are considered public wells because residents purchase water from the well owners. Based on the information collected during the site visit, the duration of pumping varies among wells, usually, between 5 and 6 hours continuously per day with one or two hour-intervals. Considering the long periods (5–6 h) of pumping and the proximity of wells to latrines, the direction of groundwater flow due to the natural hydraulic gradient is expected to have minimal effect on well contamination. Although some of the wells were surrounded by more than one pit latrine, the distance was measured from the latrine located nearest to a well. Therefore, all wells analyzed are located downgradient from the source (pit latrines).

Heads and fluxes were computed by MODFLOW during the flow simulation, then read by MT3DMS to simulate tracer transport. The longitudinal ⁠, transverse horizontal ⁠), and vertical dispersivity ⁠) values were estimated to be 20, 2 m, and 0.2 m respectively for all three layers based on Engesgaard et al. (1996). A conservative numerical tracer with an arbitrary initial concentration (C₀) of 100 mg/L was released into the aquifer from each of the 63 pit latrines. The contaminant transport model was run for 1825 days (5 years). The plume migration from the source region was observed at the nearest monitoring well.

The predicted heads obtained using MODFLOW were compared to the observed heads to assess the accuracy of the model and appropriateness of the initial input parameters. Both the goodness of fit, R² and the root mean square error (RMSE) were calculated to assess model performance. The results from the transport model were used to determine the relationships between the tracer, distance from the nearest pit latrine, and the contaminants. SPSS version 24.0 (IBM Corp 2016) was used for all statistical analyses.

Bivariate correlations were used to determine the relationships between simulated tracer, measured distance, and contaminant levels. Pearson's r correlation was used to characterize the linear relationships between the tracer concentration and measured distance from the well to the nearest pit latrines for shallow wells and deep wells as well as for the following combinations: (1) tracer concentration with nitrate, (2) tracer concentration with E. coli and (3) tracer concentration with TDS. Results from the Pearson's r correlation were used to find the relationship between contaminants and distance by using a linear regression model.

The adjusted coefficient of determination (R²) was used to measure the accuracy of linear models to identify the percentage of variance in the input (s) (Kvålseth 1985). The significance of the relationship of the combination was indicated when p-values were less than 0.01.

To understand how contaminant concentrations vary with distance and to identify site-specific separation distances based on the concentrations of measured variables, new values of fitted distances for each domestic well were obtained using the regression equation between distance and simulated tracer concentration. The simulated tracer values were used as independent variables to obtain new fitted distance (separation distance) values based on Equations (1)–(4). The recommended separation distance was selected from the highest value for nitrate and E. coli for concentration levels that met WHO guidelines.

Figure 2 shows the computed hydraulic heads in all three model layers. Figure S3 in the Supplemental Materials shows a flow vector map corresponding to these heads. The R² value was 0.95 between observed and simulated heads and the RMSE was 1.3 m, which indicates good overall agreement and model performance.

Tracer breakthrough curves generated for several wells are shown in Figure S4 in the Supplemental Materials. The shortest distance the plume traveled to reach the monitoring well with steady-state tracer concentration of 100 mg/L was 3 m, corresponding to a travel time of 4.5 days. The longest distance the plume traveled to the well was 35 m, corresponding to a travel time of 76.4 days, which had the steady-state tracer concentration of 10.2 mg/L ⁠. The average groundwater velocity was 3.63 m/day.

Figure 3 shows the results of the tracer transport simulation in the top layer after 90 days (3 months), 365 days (1 year), and 730 days (2 years). After 90 days, the transport model showed the plumes moving towards the east side of the study area, which is the direction of natural gradient flow. At 1,825 days (5 years), the tracer transport model showed that 44% of the wells had tracer concentrations that exceeded 50% of the initial concentration. While the wells that were located upgradient of the pit latrine were anticipated to have lower contaminant levels, due to the proximity of wells from the pit latrines in our study area, dispersion and mixing induced by pumping contributed to contamination of wells on the upgradient side of the pit latrines in our model. All domestic wells were within 35 m of a pit latrine.

Figure S5 in the Supplemental Materials shows the plume transport for all three layers after 2 years for a well that was located about 9 m from the pit latrine. After 2 years (730 days), the well was predicted to have a tracer concentration of 90 mg/L and the plume in the first top layer traveled more than 175 m. The predicted plume travel distance decreased to about 90 m in the second layer and about 55 m in the third layer.

Water samples collected and tested from all 63 domestic wells were found to have nitrate levels between 0 and 46 mg/L as N with an average of 18 mg/L (Figure 4(a)). The concentrations of nitrate in nearly 65% of these wells were above the WHO drinking water guideline of 10 mg/L. More than 90% of these wells were contaminated with E. coli (Figure 4(b)). TDS concentrations were between 71 and 1,910 mg/L with an average of 861 mg/L (Figure 4(c)). The WHO guideline for TDS is 600 mg/L (WHO 2017).

Nitrate, E. coli, and TDS concentrations detected through sampling and analysis at the drinking water wells were used for bivariate correlation analysis with the tracer from the transport model. Pearson's r correlations from three combinations of contaminants: (1) tracer with nitrate, (2) tracer with E. coli,and (3) tracer with TDS revealed that nitrate and E. coli are positively correlated with tracer with Pearson r values of 0.80 and 0.79 respectively. The tracer with TDS showed a weak positive correlation with r of 0.23, suggesting that these parameters are not well correlated. A summary of the r values for tracer and contaminants is provided in Table S3 in the Supplemental Materials.

Bivariate correlation results between tracer with distance showed strong negative correlations with r values of −0.96 and −0.76 for shallow and deep wells respectively. Correlative comparisons showed that nitrate and distance as well as E. coli and distance exhibit strong negative correlations with r values of −0.95 and −0.93 respectively for shallow wells and −0.85 and −0.83 respectively for deep wells. Correlative comparisons for TDS and distance showed a weak negative correlation with r values of −0.41 and −0.17, for shallow and deep wells respectively as shown in Table S4 in the Supplemental Materials.

WHO drinking water guidelines were used as the threshold to compare contaminants with the tracer concentration at the well to determine the acceptable separation distance from the linear regression analyses. According to WHO guidelines, the acceptable concentration of nitrate is 10 mg/L as N, for E. coli it is 0 CFU/100 mL and for TDS it is 600 mg/L. Results based on the actual testing showed that only 32% of the wells tested for nitrate met the WHO guidelines, 6% for E. coli, and only 17% for TDS. The model results showed the lowest tracer concentration detected in wells that met WHO standards was 0.01 (1% of C₀) and the highest was 0.41 (41% of C₀) with an average of 0.19 (19% of C₀). Linear regression was used to fit the tracer concentration with distance for shallow wells and deep wells. Results showed shallow wells had a higher R² value (0.92) compared to deep wells (0.62) as shown in Figures S6 and S7 in Supplemental Materials. When linear regression was used to fit the tracer concentration for all wells combined with distance, nitrate, E. coli, and TDS, results showed R² values of 0.71, 0.65, 0.62, and 0.05 respectively as shown in Table S5 in Supplemental Materials. Figure 4 and Figures S8, S9, and S10 in Supplemental Materials show the linear regression model results of the tracer as a dependent variable with distance, nitrate, E. coli, and TDS as independent variables.

To understand how E. coli and nitrate concentrations vary with distance and to identify acceptable separation distances based on the concentrations of these variables, new values of fitted distances for E. coli and nitrate were obtained using (a) the regression equation between distance and simulated tracer concentration (Figure 5) and (b) regression equations between the simulated tracer and observed E. coli and nitrate concentrations. New linear regression results with fitted distances are shown in Table S6 in Supplemental Materials. R² values for nitrate and E. coli are 0.65 and 0.63 respectively but for TDS the model failed to explain the variance in the data with R² of 0.05. The scatter plots supporting results from Table S6 are shown in Figures S11, S12, and S13 in Supplemental Materials. From the fitted distance results, the shortest separation distances of wells that met WHO guidelines for all three contaminants were identified as shown in Figure 6.

For shallow wells, the shortest fitted distances for nitrate, E. coli, and TDS were found to be 23.7 m, 33.2 m, and 7.4 m respectively (Figure 6). For deep wells, the shortest fitted distances for nitrate, E. coli, and TDS were found to be 18.8 m, 25.6 m, and 7.1 m (Figure 6). Results for TDS do not show a large variation between fitted distances for TDS for all wells and those that were within the WHO guideline due to the weak correlation mentioned earlier. The recommended separation distance was selected from the highest value between shallow wells and deep wells for nitrate and E. coli for concentration levels that met WHO guidelines. As such, the shortest distance was from shallow wells which was found to be 33.2 (∼34 m).

The bivariate correlations and the linear regression analysis are consistent with the statistical analysis reported by Ngasala et al. (2019a), which indicated that it was plausible that the three contaminants (nitrate, E. coli, and TDS) originated from pit latrines, providing further evidence that nitrate and E. coli originate from the same source while TDS does not. Since the study area is peri-urban and there are no agricultural activities or known animal waste dumping, the only likely source of nitrate and E. coli is human waste. The weak correlation with TDS is likely due to the TDS resulting from sources other than sewage. A likely source is a saltwater intrusion since the study area is located within ∼7 km of the Indian Ocean.

The statistical analyses presented herein are consistent with those of Ngasala et al. (2019a) who found that there was a strong negative correlation between distance and nitrate and E. coli but a weak correlation for TDS, suggesting that TDS levels may not be related to distance from pit latrines. Linear regression results from both shallow and deep wells showed higher R² values of 0.92 and 0.62 respectively. Higher R² values and strong correlations for distance and tracer suggest that contaminants are originating from nearby pit latrines and the risk of domestic well contamination increases as the distance to the pit latrine decreases.

Similar studies from different geographical locations that assessed impacts of pit latrines on groundwater quality showed a variation of site-specific separation distances based on several factors including soil type and aquifer properties (Table 1). Our findings support those of Kiptum & Ndambuki (2012) who used MODFLOW modelling with the particle tracking tool (PMPATH) to determine that the shortest distance a well should be placed from a latrine in Langa, Kenya. They found that a separation distance of 48 m was necessary to ensure that fecal bacteria levels were zero and nitrate concentrations were less than the WHO guideline of 50 mg/L. They also calculated an average particle velocity of 1.2 m/day for the aquifer. The longer it takes for a particle to travel, the more effective is filtration, adsorption, and bacterial decay, thereby reducing the contamination of wells. In Marondera, Zimbabwe, Dzwairo et al. (2006) found that pit latrines were impacting groundwater quality at separation distances of up to 25 m based on the seven water quality parameters they investigated. Two studies conducted in different regions of South Africa with different soil conditions reported that the safe separation distance for water abstraction on the site with fine sandy soil was 20 m based on fecal coliform and nitrate concentrations; whereas in the region where nitrogen compounds were investigated, the separation distance to prevent groundwater pollution was less than 12 m (Still & Nash 2002; Vinger et al. 2012).

Not surprisingly, our simulation results show that wells that had a depth of 15 m or less were more contaminated and distance from the latrine had a greater influence on water quality in those wells than on that in deep wells. These results are consistent with the correlations between contamination levels and depth reported by Ngasala et al. (2019a). Results also showed that some of the deep wells were also impacted by contamination. As found by Glanville et al. (1997) and Wilcox et al. (2010) deeper wells do not guarantee a high-quality water supply if contaminants that are introduced at upgradient locations in the flow system reach greater depths in the aquifer. Deep wells can be contaminated through cross-contamination from shallow aquifers through breaches in the aquitard, along or across long well screens, or around aquitard edges (Santi et al. 2006). Our assumption in this study is that contamination is caused by transport from upstream sources of contamination (e.g., pit latrines). However, several other factors can contribute to contamination in deep wells, including contamination of the well casing by surface water due to an improperly sealed well casing or casing at a non-complying depth, improper agricultural practices, and broken or cracked well casings.

In this paper, we expand upon the work of Ngasala et al. (2019a, 2019b) that assessed the water quality from various sources and then evaluated the factors affecting water contamination in domestic wells. We developed and demonstrated a method that could be used to determine site-specific separation distances based on the available information (i.e., soil type and aquifer properties). Using numerical simulation and water quality data, we estimated that the site-specific separation distance that drinking water wells should be placed from pit latrines is 34 m in this peri-urban area of Dar es Salaam. While public wells can be drilled deeper to prevent the possibility of contamination, source water protection is essential to prevent the migration of pollutants into the deep well.

Based on our results, the national and international standards, guidelines, and codes for private well construction and separation distances i.e., listed in Table S1 in Supplemental Materials, are insufficient to protect drinking water quality in the unsewered and overpopulated peri-urban areas of Dar es Salaam. One of the challenges for implementing drinking water protection measures in this region is that government regulations do not explicitly give local entities the authority to regulate well construction or separation distances. As noted by Wilcox et al. (2010), since the literature has not covered site-specific variables such as soil types, geology, or groundwater flow directions, these findings are not surprising. In Tanzania, groundwater is heavily exploited not only in urban areas but also in rural areas where both shallow and deep wells are common. The major benefit of the approach we have used in this study is that it can be readily applied to other peri-urban areas or in regions with different hydrologic characteristics to provide a scientific basis for designing private water supplies and determining separation distances.

Larger separation distances can aid in the reduction of well contamination by allowing for longer distances for the adsorption and filtration of suspended contaminants (Kiptum & Ndambuki 2012) and decay of bacteria. Additionally, wells should be properly lined, sealed, and covered to prevent surface water from infiltrating into the well. For the wells that are already too close to pit latrines, the water must be treated before consumption using affordable and easy to use household treatment methods (Ngasala et al. 2020). The installation of decentralized water supply systems, such as community wells, is recommended wherever possible. To ensure that these systems are properly maintained and used, the formation of community water committees comprised of community members as well as local government leaders is essential. The capital costs associated with decentralized systems for both water and wastewater can be significantly lower than centralized systems which will results in savings for individual families and to the agencies responsible for water distribution networks (Ali 2010).

Proper wastewater and sewage management will also help protect water sources from contamination. We highly recommend the proper construction of the pit latrines and septic tanks, compliance with best engineering practices and with international and national codes and standards to reduce the infiltration and exfiltration of sewage into groundwater and overflow into surface waters. To conclude, groundwater transport simulation is a useful tool to understand contamination within an aquifer system in urban overpopulated areas of developing countries. Although the model only considered point sources, non-point sources could also contribute to groundwater contamination. Our approach (linking simulated tracer concentrations with water quality data) can be extended to include non-point sources, seasonal variations in rainfall (i.e., both rainy and dry seasons) as well as non-conservative transport of multiple contaminants within an integrated modelling framework to develop simpler statistical relationships to aid management efforts.

We are grateful to the Environmental Science and Policy Program at Michigan State University (ESPP-MSU) and the Miriam J. Kelley African Scholarship Grant Program at Michigan State University for financial support. Special thanks to Atuganile Ngasala for her assistance in manuscript preparation.

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