Two unique biosand filters were constructed for this study in the workshop of the Tshwane University of Technology, Pretoria, South Africa, based on the biosand filter developed by David Manz at the University of Calgary, Alberta, Canada, in the early 1900s, with some modifications [17]. The components of a generic biosand filter include a concrete or plastic vessel filled with filter media and gravel with elevated piping that allows the filter to maintain a 5 cm layer of water above the sand surface to prevent it from drying out. The generic size dimensions of biosand filters comprise a height of 90 cm and a width of 30 cm with the fine sand media layer being 40–50 cm high. The first modification made to the design in this study was the use of a 25 L plastic bucket (height 41 cm and width 32 cm) and filter media at a height of 15 cm. The size of the filter was scaled down to ensure that the filter would not take up too much space in a small rural home. The use of clinoptilolite zeolite to replace the fine sand of the filter media was the second modification to one of the devices during this study. The construction of the two biosand filters is described below. A schematic diagram of these two biosand filters is provided in Figure 1.

The BF device was constructed from two 25 L buckets stacked on top of each other. A 20 mm hole was drilled 5 cm from the base of the bottom bucket, into which a tap was fixed for the collection of filtered water. The filter media, consisting of a 20 cm layer of fine sand (0.3 mm) and a 5 cm layer of gravel (5 mm), were packed into the top bucket. The filtering process in the BF takes place as follows: Contaminated water is poured into the upper bucket, passes through the filter media (fine sand), and the treated water accumulates in the lower bucket through the perforations in the base of the top bucket. Filtration occurs through the thick layer of fine sand while the gravel prevents the fine sand from getting into the collection vessel through the perforations [21] (Figure 3).

A ceramic candle filter (a hollow, dome-shaped cylinder, 10 cm high with a diameter of 10 cm) was obtained from Headstream Water Holdings (South Africa). It was wedged between two 25 L buckets by screwing the candle to the bottom of the upper bucket through the lid of the lower bucket, and a spigot was inserted 5 cm from the base of the lower bucket, as illustrated in Figure 4. The candle filter element was covered with a thick cloth to reduce the turbidity of the contaminated water and to trap debris (leaves and insects). Contaminated water is poured into the upper bucket and is filtered through the micropores (0.2 microns) of the dome-shaped candle filter. The filtered water accumulates in the lower bucket, where it is temporarily stored until use.

The silver-impregnated porous pot filter (Figure 5) used in this study was developed by the Tshwane University Water Research Group (Pretoria, South Africa) as part of a previous project of the Water Research Commission of South Africa (WRC Project No. K8/810). A mixture of ball clay, sawdust, paper fibre and silver nitrate solution (23.5 g) was moulded into a pot shape. The SIPP is similar to the ceramic silver-impregnated pot filter (CSF) described by Lantagne and co-workers [22]. However, the fabrication process the SIPP was modified to include AgNO₃ in the clay mixture prior to initial firing, instead of being coated with a 2 mL solution of colloidal silver (3.2% in 250 mL water) after firing, as described by Lantagne and co-workers [22]. The silver nitrate acts as a disinfectant due to the bacteriostatic properties of the nano-silver particles [23]. The complete water treatment system consists of the pot filter (maximum capacity 5–6 L) that is contained in a 10 L plastic receptacle (height 24 cm, diameter 26 cm) (Figure 5a,b), which is positioned on top of a 20 L collection bucket (height 33.5 cm, diameter 32 cm). The contaminated water is poured into the clay pot and slowly drips through the fine pores of the clay element into the collection vessel. A small spigot is used to withdraw water for drinking and to prevent the contamination of filtered water by dirty hands or utensils with which water is drawn (Figure 5b).

With the exception of the CCF, all the materials used in the construction of the devices were purchased locally. Table 2 illustrates the total cost of the materials used during the manufacturing process. The total manufacturing cost of the CCF amounted to R 501.15 (USD 64), while the total price of a complete SIPP filter is between R 240.56 (USD 30) and R 290.56 (USD 36). The following materials were used in the construction of the biological sand filters: two 25 L plastic buckets (R 25.58 each, USD 3.2), two spigots (R49.99, USD 6.2), 1 m clear tubing (R 24.99, USD 3.1), two insert elbows (R 3.79 each, USD 0.47), 1 thread tape (R 6.49, USD 0.81), 40 kg fine sand (R 28.00, USD 3.5), 40 kg gravel (R 30.24, USD 3.8), 40 kg coarse sand (R 34.82, USD 4.4) and 50 kg zeolites (R 155.35, USD 19.4). The clear tubing and the thread tape could be used to construct two filters, the fine sand 20 filters, gravel and coarse sand 40 filters, and zeolites five filters. This resulted in the cost of the BSF-S being R 133.16 (USD 16) and the BSF-Z being R 164.23 (< USD 20.00).

The selected devices were subjected to five trials in the laboratory for the removal of E. coli spiked into sterile saline deionised water. Escherichia coli ATCC 29552 was obtained from the American Type Culture Collection (Quantum Biotechnologies, RSA) and was maintained on nutrient agar (Biolab, South Africa) plates and incubated at 36 ± 1 °C for 24 h. One loop full of this bacterial culture was inoculated into 100 mL sterile nutrient broth (Biolab, South Africa) and incubated overnight (16 h) at 37 °C in a shaking incubator (Scientific Model 353, Lasec, South Africa) at a speed of 100 ± 10 rpm, after which 1 mL of the overnight culture was serially diluted in 9 mL sterile physiological water (0.9% w/v NaCl) and spread-plated onto chromocult coliform (CCA) agar (Merck) plates. The plates were incubated at 37 °C for 24 h and the resulting colonies were counted to express the initial bacterial concentrations as cfu/mL [24]. Aliquots of the E. coli culture corresponding to10⁶ cfu/mL were inoculated into 20 L (final volume) of sterile saline water (0.9%). The spiked water samples were shaken vigorously several times before being passed through the HWTS. Five trials were performed to evaluate the performance of these devices using synthetic water.

Sampling of water sources—Source water samples were collected from four different sites in Gauteng Province, between 27 September 2010 and 18 March 2011. Samples of surface water with a low (SWL) and high (SWH) turbidity were collected from the Apies River (Pretoria, Gauteng Province, South Africa) and from the Hartbeespoort Dam (Hartbeespoort, North West Province, South Africa), respectively. Samples of groundwater with a low (GWL) and high (GWH) turbidity were collected from boreholes in Delmas (on the border of the Mpumalanga and Gauteng provinces, South Africa) and Wallmannsthal (Gauteng Province), respectively. For this study, the water turbidity levels were categorised as 2 NTU to 18 NTU, 10 NTU to 40 NTU, 2 NTU to 10 NTU and 2 NTU to 15 NTU for the abovementioned test water sources, SWL, SWH, GWL and GWH, respectively. Test water samples were collected six times from each source in sterile 20 L plastic buckets. Water samples were also collected in sterile 1 L glass bottles in order to detect and enumerate the initial concentration of bacteria before treatment. The samples were transported to the laboratory within 6 h and assayed for microbiological contamination and turbidity level to determine the quality of the water before and after treatment [24].

Source water samples were filtered as follows through each device in the laboratory: 5 L/d for SIPP and 20 L/d for each of the remaining HWTS. Filtration was done taking into consideration that rural communities could use any available water source that is accessible to them during seasonal changes. Different volumes of filtrates were collected at 1 h intervals over the 3 h period of filtration with the assumption that enough purified water would have been produced in this period for drinking and cooking. This also allowed the researcher to establish whether significant differences in the reduction of microbial contaminants could be found at different interval times and to make the necessary recommendations in terms of safe drinking water. The total volumes of all water types filtered were 305 L and 1220 L for the SIPP and the other four devices, respectively.

Analysis of water source quality—Faecal coliforms and presumptive E. coli were determined by membrane filtration techniques using mFC agar (Merck SA) and Chromocult agar (CCA; Merck SA) plates that were incubated at 44.5 °C and 37 °C, respectively, for 24 h [24,25]. All tests were performed in triplicate. Following the enumeration, five characteristic colonies for presumptive E. coli were randomly selected from different plates and transferred onto chromocult media by the streak-plate method and incubated at 37 °C for 24 h. The colonies were further purified by the same method at least three times using nutrient agar (Biolab, SA) before Gram staining. Oxidase tests were then performed on those colonies that were Gram negative. The API 20E kit was used for the oxidase-negative colonies and the strips were incubated at 37 °C for 24 h. The strips were then analysed and bacterial species were identified using APILAB Plus bacterial identification software (BioMérieux, Marcy l’Etoile, France). The bacterial counts were calculated and expressed in cfu/100 mL.

Five colonies of each test bacteria presumptively identified as E. coli were subcultured onto selective media three times before being used for molecular identification. An overnight culture of the reference E. coli strain (ATCC 29552) and each presumptive E. coli colony was prepared in nutrient broth at 37 °C. Bacterial cells (1 mL culture) were concentrated by centrifugation at 10,000× g for 5 min in a microcentrifuge (Heraeus Pico 21, Thermo Scientific) and then suspended in 200 μL sterile Milli-Q water. DNA was extracted with the ZR Fungal/Bacterial DNA Kit^TM (ZYMO Research, Pretoria, South Africa) according to the procedures provided by the manufacturer.

Molecular identification of E. coli was done by amplification of the uidA gene that encodes for β-D-glucuronidase using primers that are specific (Table 3) for this gene in E. coli. The PCR amplification of the target DNA was carried out in a thermal cycler (MJ Mini^TM Personal Thermal Cycler, Biorad SA) using 200-μL PCR tubes and a reaction mixture volume of 50 μL. A reaction mixture was prepared, containing 20 ng of template DNA, 25 μL 2× Dream Taq^TM PCR master mix (10× Dream Taq^TM buffer, 2 μM dnTP mix and 1.25 U Dream Taq^TM polymerase) and a 10 μM concentration of each PCR primer (synthesised by Inqaba Biotechnical Industries (Pty) Ltd, Pretoria, South Africa) and was made up to 50 μL with ultra-pure nuclease-free water. The following cycling parameters were used: denaturation of template DNA at 94 °C for 2 min, followed by 25 cycles of denaturation at 94 °C for 1 min, annealing of template DNA at 58 °C for 1 min and an extension time of 1 min at 72 °C for the primers. After the last cycle the samples were kept at 72 °C for 2 min to complete the synthesis of all strands [26,27].

An aliquot of 10 μL of PCR product was electrophoresed through a 1.5% agarose (Merck, SA) gel stained with 4.5% ethidium bromide (Merck, SA) to determine the size of the product, which was visualised under UV light in a InGenius L Gel documentation system (Syngene). A negative control and a positive control were also included in the PCR run [26]. The negative control consisted of all PCR reagents, except for the template DNA, and the positive control was the DNA of the reference E.coli strain at two different concentrations 10 μL and 5 μL.

For the SIPP, CCF and BSF-S devices, the flow rates were measured by recording the volume of water collected from these devices after a period of 1 h. This was done over 3 h to obtain a triplicate reading. Contaminated water was passed through the HWTS devices and accumulated in the collection vessels (bottom buckets of BF, CCF and SIPP and 10 L sterilised plastic buckets for BSF-S and BSF-Z) for 1 h, after which samples (first hour collection) were taken for analysis. The collection vessel was allowed to collect filtered water for one more hour, after which the water sample was taken again for analysis (second hour collection). The third hour collection was obtained in the same manner as the first and second hour collections. The flow rate of the BSF-Z was measured by recording the volume of water collected in one minute immediately after the water had been poured into the filter; this was done at hourly intervals over a period of 3 h. A volume of 5 L was initially filtered during the first and second hour followed by filtration of the remaing 10 L in the third hour. The flow rate of the BF was measured by recording the time it took to filter 20 L of water. This was converted to L/h. Flow rate is recorded as L/h in this study.

A portable turbidity meter (2100P Hach) was used to determine the level of turbidity in water samples before and after filtration. The percentage turbidity reduction achieved by each of the filter devices was calculated using the following equation:

The bacterial removal efficiency was obtained by comparing the concentrations of target organisms before and after treatment. Enumeration of faecal coliforms and presumptive E. coli after treatment was done by standard methods as mentioned above [24,25]. The Log bacterial reductions were calculated using the equation below and were converted to percentage killed [28]:

The present part of the study is a continuous investigation on the performance of SIPP in removing pathogenic organisms from contaminated water sources. The preliminary experimental studies were conducted in our laboratory by Momba and co-authors [29,30]. During this initial study, the authors conducted a series of analytical and mechanical characterization tests on the SIPP, which included the XRF analysis that confirmed the presence of Ag, the break strength and the porosity of the filter. These authors found that the silver leached from the SIPP pot was at a level of between 0.5 and 0.6 mg/L [29], higher than the WHO recommendation of 0.1 mg/L [31]. The Ag elution was greatest in the early stages (within the first 5 L) but appeared to begin to stabilize by 10 L.

For the present study, the SIPP filter was soaked in 20 L deionised water overnight prior to use. The concentration of the silver in the filtered water was monitored at one-hour intervals over a three-hour period to determine the Ag elution by the SIPP after filtering a total volume of 305 L. The first, second and third filter runs were performed with deionised water, groundwater and surface water, respectively. The Spectro Acros ICP spectrometer (Spectro, RSA) was used to detect and determine the concentration of silver in each water sample.

The statistical software package used to analyse the data was Stata V10 (StataCorp LP, College Station, Texas, USA). Data obtained for flow rates, turbidity and microbial contaminants after treatment were subjected to One-Way Analysis of Variance (ANOVA) to compare more than two groups. Comparisons were made between the treatment means of each device per water source to determine whether there were significant differences between treatments. Where differences were observed, pair-wise comparisons or post hoc tests were performed and the Wilcoxon Rank-Sum Test was used to compare the two groups. The interpretation was performed at 95% confidence limit. The tests for relationships between flow rates and turbidity or between turbidity and bacterial removal for each home water treatment were carried out using the Pearson Correlation index.

Generally, poor E. coli removal efficiency was noted when filtering synthetic water through the BSF-S, but total removal was observed in the first hour of the fifth trial (Figure 9). With surface water sources, total removal of faecal coliforms was initially noted during the third trial after filtering a volume of low-turbidity surface water (SWL) up to 800 L through the BSF-S and also within the second and third hours during the fourth trial (Figure 9a). Complete removal of E. coli was noted within the first and third hours after filtering up to 840 L (4th trial) of this test water source through the BSF-S (Figure 8b). Complete removal of faecal coliforms was observed during the first two trials and during the sixth trial after filtering highly turbid surface water to make up total volumes of 1120 L, 1140 L and 1220 L respectively (Figure 9a). No E. coli were detected during the first three treatment trials (Figure 9b).

Filtration of low-turbidity groundwater (GWL) up to total volumes of 880 L, 900 L, 960 L and 980 L using the BSF-S device resulted in the complete removal of faecal coliforms during the first two trials and also during the fourth and sixth trials, respectively (Figure 9a). There was a complete removal of E. coli during the third, fourth and sixth trials after filtering up to 920 L, 940 L and 960 L of GWL, respectively (Figure 9b). Total removal of faecal coliforms occurred during the last three trials after filtering highly turbid groundwater (GWH), corresponding to total filtered volumes of 1060 L, 1080 L and 1100 L, and also in the second and third hours after filtering a total volume of 1040 L (Figure 9a). No E .coli concentrations were noted in any of the trials, except for the first hour of the first trial (Figure 9b).

Poor removal efficiency was observed when filtering synthetic water source through the BSF-Z, as E. coli concentrations after treatment ranged between 2 Log₁₀ and 6 Log₁₀. However, complete removal of E. coli and faecal coliforms was observed during the first and third trials, respectively, after filtering SWL of total volumes of 800 L and 760 L (Figure 10a,b). During the filtration of highly turbid surface water, no faecal coliforms were detected during the first, second and sixth trials after filtering total volumes of 1120 L, 1140 L and 1220 L, respectively, while E. coli were completely removed during the first three trials (Figure 10a,b).

Trials with low-turbidity groundwater sources indicated no faecal coliforms during the first two and last two trials after adding total volumes of 880 L, 900 L and 960 L, 980 L, respectively, to the BSF-Z (Figure 10a). Total removal of E. coli was noted during the second, third and fourth trials after filtering total volumes of 900 L, 920 L and 940 L of low-turbidity groundwater, respectively, as well as in the second and third hours of the first trial (Figure 10b). Adding up to 1060 L, 1080 L and 1100 L of highly turbid groundwater to the filter resulted in the total removal of faecal coliforms during the last three trials. As indicated in Figure 10a, this bacterial group could also not be detected during specific hours of the first three trials after filtration of up to 1000 L, 1020 L and 1040 L. No E. coli were detected during the last four trials after filtering total volumes of 1040 L, 1060 L, 1080 L and 1100 L, as well as during the second and third hours of the first two trials (Figure 10b).

In general, the trend in the removal of target bacterial pathogens in BSF-S and BSF-Z was towards increased bacterial removal with repeated filtration of raw water sources. This shows that either the development of the biological layer (Schmutzdecke) or the increase in the retention time with the filter media as the filtration rate decreases due to head-loss accumulation has an impact on bacterial removal [40,41,42,43]. Studies by Kaiser et al. [38] and Campos and his colleagues [40] have also shown that bacterial removal efficiency improves with repeated filter use and increasing time of filtration, due to filter maturation.

The removal of E. coli from synthetic water source by BF was found to be poor, as can be seen in Figure 11. With surface water sources, total removal of faecal coliforms was observed in the second and fourth hours of the second and third trials after filtering total volumes of 780 L and 800 L, respectively (Figure 11a). No E. coli were detected during the second and third trials after filtering up to 800 L and 820 L. The complete removal of E. coli and faecal coliforms was achieved during the second trial and in the first and third hours of the fourth trial, respectively, after filtering total volumes of 1140 L and 1180 L (Figure 11a,b).

During the first two trials and last two trials with low-turbidity groundwater, there was total removal of faecal coliforms after filtering up to 880 L, 900 L and 960 L, 980 L respectively of this water source through the bucket filter (Figure 11a). Complete removal of E. coli was observed during the fourth and sixth trials after filtering up to total volumes of 920 L and 980 L, respectively. No E. coli were detected in the second and third hours of the fourth trial (Figure 11b). During the treatment of highly turbid groundwater, faecal coliforms and E. coli were totally removed during the last two trials after filtering total volumes of 1080 and 1100 L (Figure 11a,b).

Overall, the bucket filter showed the poorest removal of faecal coliforms [2–3 log₁₀ (99–99.9%)] and E. coli [1–3 log₁₀ (90–99.9%)], as no biological layer develops on the top surface of the sand bed in a rapid sand filter to enhance the removal of bacterial pathogens (Table 6). The bucket filter does not have a resting water level as the biological sand filters have, and dries up. Thus, conditions in a bucket filter are not favourable for the development of a biological layer [9,34]. This clearly explains its poor performance in bacterial removals compared to the biosand filters. As Sobsey [9] reported that bucket filters are not expected to reduce bacteria by more than 90%, approximately 50% reduction was expected. However, the BF filter in this study removed upto 99.9%. The high performance of this rapid sand filter can be attributed to the particle size of sand used (0.3 mm), which is smaller than the size reported by Sobsey (1–3 mm) [9].

While poor E. coli removal efficiency was observed during filtration of synthetic water through the CCF (Figure 12), no faecal coliforms were observed during the 3rd, 4th and 6th trials after filtering low-turbidity surface water through this CCF to make up total volumes of 800 L, 820 L and 860 L, respectively. As illustrated in Figure 12a, faecal coliforms could also not be detected during the second and third hours of the second trial after filtration of up to 780 L. Filtration of total volumes of 820 L and 840 L low-turbidity surface water during the fourth and fifth trials resulted in the complete removal of E. coli after the second and third hours of treatment. With high-turbidity surface water, no faecal coliforms could be detected during the first two trials and also during the fifth trial after filtering up to 1120 L, 1140 L and 1200 L of water respectively in total (Figure 12b).

With groundwater sources, complete removal of faecal coliforms was achieved during the first two trials and during the last two trials after filtering up to total volumes of 800 L, 900 L, 960 L and 980 L of low-turbidity groundwater, respectively (Figure 12a). During the last four trials, no E. coli were observed in the filtered water samples (Figure 12b). Faecal coliform bacteria were not observed during the last three trials after filtering up to 1080 L, 1080 L and 1100 L intake GWH source (Figure 12a). With the exception of the first hour of the first trial, no E. coli were detected after treating this type of water source using the CCF device (Figure 12b).

This device consistently produced high-quality water that did not contain any traces of faecal coliform or E. coli during each hourly interval for each day of the filter run throughout the study period (Table 7). The device removed between 2.5 Log₁₀ and 6.6 Log₁₀ of these coliform bacteria from contaminated water. The results for SIPP are presented in Table 7, as no bacteria were detected in any of the treated water samples.

The average faecal coliform and E. coli removal and percentage removal efficiency by HWTS devices from the various water sources are summarised in Table 6. The SIPP was the only device that consistently produced drinking water complying with the recommended limits set by the South African National Standard (SANS) 241 Drinking Water Specification [16], regardless of the quality of the source water treated. The percentage removal of E. coli observed in the other four devices when treating synthetic water was >99.9%. The removal of target bacteria from the environmental water samples ranged between 2–4 Log₁₀ (99.9% to >99.99%), 1–4 Log₁₀ (90% to >99.99%), 1–3 Log₁₀ (90% to 99.9%) and 2–3 Log₁₀ (90% to 99.9%) for the BSF-S, BSF-Z, BF and CCF, respectively (Table 7). All devices removed E. coli within similar ranges or in some cases above the ranges reported by Elliott et al. [35], Baumgartner [42] and Laurent [43]. Biosand filters have been reported to remove E. coli in the rate ranging between 85% and 98% [35,40,41]. Chemically modified rapid sand filters may remove between 80% and 95% [43]. The removal of bacteria by the SIPP and CCF was observed to be >99% for the duration of the study. The performance of these two filters was within the ranges reported by previous investigators. The ceramic candle filter and silver-coated ceramic pot filters have been shown to remove E. coli in ranges between 80% and 99%, and between 95% and 100%, respectively [44,45].

As stated above, the preliminary investigation on the performance of SIPP in removing pathogenic organisms from synthetic water sources was conducted in our laboratory by Momba and co-authors [29]. Escherichia coli spiked in synthetic water were used as a prototype organism and the authors included a control porous pot that was not impregnated with AgNO₃. It was revealed that only the Ag impregnated pot was significantly effective in removing this organism [30]. The more efficient bacterial performance of the SIPP as compared to the control porous pot was attributed to the imbedded Ag nanoparticles in the micropores [29]. The silver leached from the SIPP pot was at a level of between 0.5 and 0.6 mg/L, higher than the WHO recommendation of 0.1 mg/L [31]. The Ag elution was greatest in the early stages (within the first 5 L) but appeared to begin to stabilize after filtering up to 10 L of contaminated synthetic water. The SIPP, however, displayed a more gradual reduction for Ag released into the water [29].

Similar observations were also noted in the current study. The SIPP filter was found to be the most efficient device in removing faecal coliforms and E. coli from all test water source samples, as it reduced the maximum E. coli concentration of up to 6.6 log₁₀ in spiked water to 0 cfu/100 mL (Table 6, Table 7). This corroborated the findings of Van Halem and co-workers [44], who determined that the highest possible reduction of E. coli by the silver-impregnated ceramic pot filter was 7 log₁₀. The SIPP filter showed elevated silver concentrations in the filtrate from the first and the second run with synthetic water (Table 8). Although the concentration of the Ag in the filtrate gradually dropped to 0.22 mg/L after filtration of groundwater and surface water sources up to a total volume of 320 L, this value was still above the WHO recommended limits [31]. Though SIPP filter displayed high levels of Ag elution above the level recommended by the WHO, independent studies have indicated the relative non-toxicity of Ag and its health benefits [46].

The bacterial removal efficiency of the SIPP filter in the present study can be due to straining of bacteria through the fine pores of the clay pot [22,23]. It is also possible that the removal efficiency could have been enhanced by the AgNO₃ that was embedded in the clay during the manufacturing process. The longer disinfection capacity of the SIPP could be attributed to the presence of the silver ion since independent studies have corroborated the effect of Ag in a water purification application, irrespective of substrate [47,48,49,50]. Silver has bacteriostatic properties and inactivates bacteria by disrupting the disulphide bonds of proteins in the cell membrane or by inhibiting DNA replication [23,44,51]. It is thus possible that the silver is contributing to the removal of bacteria. However, in future a study involving the comparison of the performance of the SIPP to a non-silver clay pot should be included to give a clear indication of whether the silver enhances the performance of the clay pot filter. Future studies should also include a long-term study whereby the concentration of silver in water filtered by SIPP is monitored on a weekly basis over 12 months or more. This would also indicate how the SIPP performs as the silver in leached out.

Due to its highly effective performance, the SIPP was statistically compared to the other selected devices. The removal of faecal coliforms and E. coli by the CCF and BF was found to be significantly lower than that of the SIPP (p < 0.05) (Table 9). This was expected, as the first two devices lack a disinfectant that is able to inactivate bacteria at a rate comparable to that of the SIPP device. In the CCF, bacterial pathogens are removed through mechanical trapping and adsorption [45,52]. Mechanical trapping takes place in two ways. The first is through the surface filtration, whereby particles that are too large to pass through the pore are trapped on the surface of the candle. The second is through the depth filtration, where particles that penetrate the candle filter are trapped throughout the depth of the candle wall [52]. In the BF, the removal of particles, including bacteria, occurs through physical trapping with deeper penetration into the sand bed [9]. However, the microbiological quality of water filtered through the BF (rapid sand filter) has been revealed to be poor, as no biological layer develops on the top surface of the sand bed [9,53] to remove bacterial pathogens. The statistical analysis showed that both the BSF-S and BSF-Z performed similarly to the SIPP in removing the indicator bacteria from environmental water samples (p > 0.05). However, the experimental data indicated that SIPP was the best device overall, as the average removal of these bacteria from environmental water samples by the BSF-S and BSF-Z devices ranged between 96% to 99.9% while removal of bacteria by the SIPP was >99.99% throughout the study. The average removal of faecal coliforms ranged between 2.2 to 3 log₁₀ or 99.2% to >99.9% for BSF-S and 3 to >3.7 log₁₀ or 99.9% to 100% for BSF-Z, whereas the average removal of E. coli from raw water ranged between 1.6 to 4 log₁₀ (96% to 100%) for BSF-S and between 1.3 to 4 log₁₀ (93% to 100%) for BSF-Z (Table 6).

The removal efficiencies obtained from this study are much higher than those reported by Stauber et al. [41] and Ngai et al. [53] for biosand filters, namely between 60% and 100%. This implies that the modified biosand filters BSF-S and BSF-Z are more efficient. The more efficient performance of the BSF-S can be attributed to the fine sand (0.15 mm) used which enhanced straining of particles due to the formation of finer pores as compared to conventional biosand filters [13,17]. The high performance of the BSF-Z can be a result of the antimicrobial properties of zeolites reported by Princz et al. [18], Kallo and Ming [19] and Widiastuti et al. [20]. Statistical analysis showed that there was a significant difference (p < 0.05) in the removal of E. coli from surface water of high turbidity by BSF-S and BSF-Z. The BSF-Z showed higher removal efficiency than BSF-S (Figure 9 and Figure 10). However these biosand filters performed similarly in removing E. coli from all groundwater samples and from surface water of low turbidity (p > 0.05) and faecal coliforms from all test water samples (Table 9). The removal of bacteria from raw water by biosand filters occurs by a combination of physical and biological processes [54]. The biological process involves a predation process, whereby algae and protozoa that develop on the surface of the top layer of sand within four to six weeks of filter use consume incoming bacteria; these organisms, along with slime or biofilm, comprise the Schmutzdecke or biological layer [41,53,54]. A great improvement in the removal of indicator bacteria was noted in both the BSF-S and BSF-Z as contaminated raw water was progressively treated. The bacterial removal efficiency was higher during filtration of surface water of high turbidity (SWH) compared to surface water of low turbidity (SWL) which was the first environmental water sample to be treated (Figure 9 and Figure 10). This can be attributed to the maturation of the biological layer that occurs with increased filter use [35,38]. The biological layer is also responsible for the decline in flow rate as the pores become clogged due to the accumulation of dirt in the sand bed. This increases the contact time between filter media and contaminated water thereby enhancing the performance of biosand filters over time [54]. Statistical analysis was also performed to compare the bacterial removal efficiency of the one-, two- and three-hour intervals of filtration for each filter. No significant differences (p > 0.05) were found between these hourly intervals across all five devices and the multiple water sources filtered. This finding implies that water for drinking can be collected at any time from these devices. However, it should be kept in mind that only water from the SIPP can be collected and consumed directly after filtration. The Water from the CCF, BSF-S, and BSF-Z filters would require a disinfection step to render their water safe for consumption.

Turbidity is a measure of the clarity of a liquid and was once considered mainly an aesthetic characteristic of drinking water [36,55]. However, some studies have shown that controlling turbidity is a key step in eradicating pathogens from drinking water, as microorganisms can be embedded or attached to the surface of larger suspended particles. Therefore, the removal of the larger particles could result in the removal of pathogenic microorganisms [56]. Correlations between turbidity reduction and bacterial reductions were performed to determine the relationship between the turbidity and the bacterial concentration of a water sample. The results from our study show that there is a direct relationship (as the turbidity was reduced, the indicator bacteria were also reduced) between turbidity and bacterial removal as strong positive correlations were observed between the removal of E. coli and turbidity from groundwater of low turbidity by the BSF-S (r = 0.761) and BSF-Z (r = 0.780). LeChevallier and Norton [55] and Fox [56] who reported that water samples with a lower turbidity are associated with lower microbial counts and vice versa substantiate these findings. However, strong negative correlations were observed between the removal of turbidity and removal of faecal coliforms and E. coli from SWH and GWH by the BSF-S (r = −0.721), BSF-Z (r = −0.796) and CCF (r = −0.784), respectively (Table 10). These findings imply that an indirect relationship existed between the turbidity and bacterial concentration (as the turbidity was reduced by these devices, the concentration of faecal coliforms or E. coli increased and vice versa). However, from the graphs it can be seen that 1–3 log (90%–99.9%) reduction of the microorganisms was achieved (Figure 9a, Figure 10a). It is possible that the turbidity value remained high even though high bacterial removal efficiencies were observed. This can be explained by the fact that organic and inorganic materials, slit, clay and soluable coloured organic particles largely contribute to the turbidity of a water source, as opposed to bacteria or viruses, which are small and translucent [55].