Biosand Filter as a Point-of-Use Water Treatment Technology: Influence of Turbidity on Microorganism Removal Efficiency

The number of people living without access to clean water can be reduced by the implementation of point-of-use (POU) water treatment. Among POU treatment systems, the domestic biosand filter (BSF) stands out as a viable technology. However, the performance of the BSF varies with the inflow water quality characteristics, especially turbidity. In some locations, people have no choice but to treat raw water that has turbidity above recommended levels for the technology. This study aimed to measure the efficiency with which the BSF removes microorganisms from well water and from fecal-contaminated water with turbidity levels of 3, 25, and 50 NTU. Turbidity was controlled by the addition of kaolin to water. Turbidity removal varied from 88% to 99%. Reductions in total coliform (TC) and Escherichia coli ranged from 0.54–2.01 and 1.2–2.2 log removal values (LRV), respectively. The BSF that received water with a higher level of turbidity showed the greatest reduction in the concentration of microorganisms. Additional testing with water contaminated with four bacterial pure cultures showed reductions between 2.7 and 3.6 LRV. A higher reduction in microorganisms was achieved after 30–35 days in operation. Despite the filter’s high efficiency, the filtrates still had some microorganisms, and a disinfection POU treatment could be added to increase water safety.


Introduction
Despite numerous investments, about 2.1 billion people in the world still consume drinking water from sources contaminated with feces [1], and about 2.3 billion people still lack basic health infrastructure. Furthermore, some 159 million people, mainly from rural areas and principally in developing countries, still drink water collected directly from surface sources and shallow, unsafe wells [1,2]. There are high rates of gastrointestinal and parasitic diseases, as well as malnutrition associated with the consumption of water contaminated by microbial pathogens [3,4].
Waterborne diseases remain the leading cause of mortality worldwide, registering more than 2.2 million deaths per year [5], with diarrhea being responsible for 1.5 million deaths [6]. Most cases occur in developing countries and affect children under five years of age [3]. Point-of-use (POU) water treatment technologies have been recommended as an effective solution for the provision of safe water in places where families do not have access to conventional systems for the treatment and supply of drinking water [7].

Filter and Media Preparation
The four filters used in the experiments were designed according to the Center for Affordable Water and Sanitation Technology (CAWST) [13]. The filters' containers were made of acrylics, a material that was available in the laboratory. The containers had 10 mm thickness, 300 mm inside diameter and 990 mm height. The containers were covered with low density polyethylene plastic to prevent algal growth on the walls. A diffuser plate was installed in each of the BSF by means of a trapezoidal polyethylene bucket with top and bottom diameters of 300 mm and 200 mm, respectively. The bucket was placed with its upper part at the top of the container. The bottom of the bucket was perforated with 2 mm holes, to ensure that feed water was evenly spread onto the filter top surface ( Figure 1).
As support layers, the filters were filled with 8 cm of gravel of size 6-12 mm, 7 cm of granular gravel of size 1-6 mm, and 6 cm of coarse sand of size 1 mm. Above this support base, it was placed 46 cm of fine sand with an effective size (D10) of 0.11 mm and a uniform coefficient of 0.16 mm. All fillings were sieved, washed, and dried, before being placed into the filters. Granulometric tests of coarse and fine sands were made according to the methodology described in NBR 11799 [19]. Standing water was kept 4.8 cm above filter surface. The outlet tube with 19 mm diameter was elevated at the same standing water level above the filter sand layer. This elevation enabled a layer of water to be maintained above the top of the sand, regardless of whether the filter was operating or not. When in operation, water coming from the perforated bucket maintained pressure to force the flow through the pores of the filter. The flow rate slowed down as the operation approached the resting period. When the water finally stopped flowing, the standing water layer had the same height as that of the end of the outlet tube. The average surface application rate was approximately 1.44 m 3 /m 2 /day. Water 2020, 12, x FOR PEER REVIEW 3 of 19 as that of the end of the outlet tube. The average surface application rate was approximately 1.44 m 3 /m 2 /day.

Feed Water
Filter influent consisted of water from a well (BSF1) and dechlorinated tap water that had been contaminated with filtered secondary effluent from a local wastewater treatment plant (BSF 2, 3, and 4). Dechlorination was achieved using sodium thiosulfate (Na2S2O3). The ratio of dechlorinated tap water to effluent was 9:1. Kaolin solutions in concentrations of 5, 50 and 90 mg/L were added to BSF 2, 3, and 4 to achieve turbidity of 3, 25, and 50 NTU, respectively. The air and water temperature could not be controlled. While the average room temperature ranged between 12 and 25 °C during the experimental work, water temperature varied between 17 °C and 18 °C.
During a three-month period of operation, filters were dosed daily with 12 L of influent water. In the first 45 days, filters operated continuously for 24 h followed by a resting period of 24 h. From days 46 to 90, the BSFs were fed for 24 h with a pause period of 48 h. Images of the grains of sand were obtained by scanning electron microscopy (SEM) on days 1, 45, and 90 of BSF operation. BSF4 continued to operate for an additional 30 days after closing BSF 1, 2, and 3.
From days 120 to 125, BSF4 received 20 liters of water with 50 NTU turbidity, free of chlorine and bacteria. In days 124 and 125, samples of filtered water were collected and analyzed. No colony forming units (CFU) were detected. After confirmation that BSF4 effluent was free of CFU, 20 L of dechlorinated tap water with a turbidity of 50 NTU was contaminated by adding Escherichia coli ATCC® 25922 (E. coli), Salmonella Typhimurium DT177 (S. Typhimurium), Enterococcus Faecalis ATCC® 29212 (E. faecalis), and Pseudomonas aeruginosa ATCC® 27853 (P. aeruginosa) to a final

Feed Water
Filter influent consisted of water from a well (BSF1) and dechlorinated tap water that had been contaminated with filtered secondary effluent from a local wastewater treatment plant (BSF 2, 3, and 4). Dechlorination was achieved using sodium thiosulfate (Na 2 S 2 O 3 ). The ratio of dechlorinated tap water to effluent was 9:1. Kaolin solutions in concentrations of 5, 50 and 90 mg/L were added to BSF 2, 3, and 4 to achieve turbidity of 3, 25, and 50 NTU, respectively. The air and water temperature could not be controlled. While the average room temperature ranged between 12 and 25 • C during the experimental work, water temperature varied between 17 • C and 18 • C.
During a three-month period of operation, filters were dosed daily with 12 L of influent water. In the first 45 days, filters operated continuously for 24 h followed by a resting period of 24 h. From days 46 to 90, the BSFs were fed for 24 h with a pause period of 48 h. Images of the grains of sand were obtained by scanning electron microscopy (SEM) on days 1, 45, and 90 of BSF operation. BSF4 continued to operate for an additional 30 days after closing BSF 1, 2, and 3.
From days 120 to 125, BSF4 received 20 liters of water with 50 NTU turbidity, free of chlorine and bacteria. In days 124 and 125, samples of filtered water were collected and analyzed. No colony forming units (CFU) were detected. After confirmation that BSF4 effluent was free of CFU, 20 L of dechlorinated tap water with a turbidity of 50 NTU was contaminated by adding Escherichia coli ATCC ® 25922 (E. coli), Salmonella Typhimurium DT177 (S. Typhimurium), Enterococcus Faecalis ATCC ® 29212 (E. faecalis), and Pseudomonas aeruginosa ATCC ® 27853 (P. aeruginosa) to a final concentration of approximately 2.5 × 10 7 CFU/mL. Each bacterium was first cultivated in Tryptic Soy agar for 24 h at 37 • C. Two milliliters of brain heart infusion (BHI) culture medium was inoculated with three colonies of each bacteria and incubated at 37 • C for 18 h. Each bacteria species was cultivated separately. BSF4 operated continuously while receiving the contaminated influent for five consecutive days with a resting period of 24 h. Samples from the filtrate were collected on days three, four and five, kept in 50 mL Falcon tubes and refrigerated. They were analyzed within a maximum of two hours after collection. They were submitted to five serial dilutions, as described in the literature [20]. The drop plate seeding technique was implemented to inoculate the selected agar plates, BEM (eosin methylene/blue agar-Levine), XLD (xylose deoxyclolate agar), BEA (Bili Scullin agar), cetrimide, and TSA (Tryptone Soy Agar) for E. coli, S. Typhimurium, E. faecalis, P. aeruginosa and TC, respectively. The plates were incubated at 35 • C and checked for CFU after 18 h. The non-growing plates remained in the incubator for additional 18 h. The number of colonies counted was multiplied by the respective dilution factor to calculate the concentration in CFU/mL. Three replicates were performed for each sample. Figure 2 presents the experimental timeline and influent water preparation, including the concentrations of kaolin that were added to achieve the target turbidity. concentration of approximately 2.5 × 10 7 CFU/mL. Each bacterium was first cultivated in Tryptic Soy agar for 24 h at 37 °C. Two milliliters of brain heart infusion (BHI) culture medium was inoculated with three colonies of each bacteria and incubated at 37 °C for 18 h. Each bacteria species was cultivated separately. BSF4 operated continuously while receiving the contaminated influent for five consecutive days with a resting period of 24 h. Samples from the filtrate were collected on days three, four and five, kept in 50 mL Falcon tubes and refrigerated. They were analyzed within a maximum of two hours after collection. They were submitted to five serial dilutions, as described in the literature [20]. The drop plate seeding technique was implemented to inoculate the selected agar plates, BEM (eosin methylene/blue agar-Levine), XLD (xylose deoxyclolate agar), BEA (Bili Scullin agar), cetrimide, and TSA (Tryptone Soy Agar) for E. coli, S. Typhimurium, E. faecalis, P. aeruginosa and TC, respectively. The plates were incubated at 35 °C and checked for CFU after 18 h. The nongrowing plates remained in the incubator for additional 18 h. The number of colonies counted was multiplied by the respective dilution factor to calculate the concentration in CFU/mL. Three replicates were performed for each sample. Figure 2 presents the experimental timeline and influent water preparation, including the concentrations of kaolin that were added to achieve the target turbidity.

Statistical Analysis
The arithmetic mean and standard deviation were used for all data except microbiological data, for which the geometric mean was used. The percent reduction was calculated using source and filtered water (Influent-Effluent/Influent × 100) from each BSF for all water quality parameters. Logarithm removal values (LRV) and percent reductions were calculated for bacterial removal.
The Student's t-test (two-tailed) was used to evaluate the significance of the differences between the mean of most probable number (MPN) values detected before and after water filtration. A probability level of less than 0.05 (p-value < 0.05) was considered to be significant.
To analyze the influence of turbidity on the removal of total coliform (TC) and E. coli, an adjustment of mixed models was used. Separate means of responses were analyzed both for BSF and time (Time 1 before filtration and Time 2 after filtration), in addition to the interaction means between filter and time. Since the variable TC did not reach the assumption of normality, the log-normal distribution was used. Assumptions of normality are called mixed generalized linear models [21].

Statistical Analysis
The arithmetic mean and standard deviation were used for all data except microbiological data, for which the geometric mean was used. The percent reduction was calculated using source and filtered water (Influent-Effluent/Influent × 100) from each BSF for all water quality parameters. Logarithm removal values (LRV) and percent reductions were calculated for bacterial removal.
The Student's t-test (two-tailed) was used to evaluate the significance of the differences between the mean of most probable number (MPN) values detected before and after water filtration. A probability level of less than 0.05 (p-value < 0.05) was considered to be significant.
To analyze the influence of turbidity on the removal of total coliform (TC) and E. coli, an adjustment of mixed models was used. Separate means of responses were analyzed both for BSF and time (Time 1 before filtration and Time 2 after filtration), in addition to the interaction means between filter and time. Since the variable TC did not reach the assumption of normality, the log-normal distribution was used. Assumptions of normality are called mixed generalized linear models [21]. Table 1 presents the parameters that were analyzed, together with the methods and instruments used. The analyses were performed according to the methodologies of the Standard Methods for Examination of Water and Wastewater [22].  Table 2 presents the results for water quality variables monitored in the influent and effluent of the filters.  Figure 3 shows turbidity values measured in the influent and effluent of each filter along the experiment. Most filtrates had turbidity lower than 1.0 NTU. Following day 30, the turbidity in filtrates remained approximately constant, except for some lower values measured after day 80. The ripening of the filters may have occurred after day 30, but this was not confirmed by SEM images. The removal rates measured in this study were comparable with values described in the literature [12,23,24].  Figure 4 shows the influent and effluent pH values for BSFs throughout the experiment. Filtrates from BSF2-4 had consistently lower pH values than the influent, showing that some acid-forming reaction was occurring within the filter medium. Reductions in pH values were also observed by Kennedy et al. [12].  Figure 4 shows the influent and effluent pH values for BSFs throughout the experiment. Filtrates from BSF2-4 had consistently lower pH values than the influent, showing that some acid-forming reaction was occurring within the filter medium. Reductions in pH values were also observed by Kennedy et al. [12]. Figure 5 shows the alkalinity concentrations in the filters' influent and effluent. BSF1 had an average influent water alkalinity of 103 mg/L CaCO 3 , and did not show a significant decrease throughout the entire operation. BSF2, BSF3 and BSF4 showed a reduction in alkalinity concentrations of approximately 50%. This suggests that an acid-forming reaction occurred within the filter media.

Alkalinity
Water 2020, 12, x FOR PEER REVIEW 8 of 19  Figure 5 shows the alkalinity concentrations in the filters' influent and effluent. BSF1 had an average influent water alkalinity of 103 mg/L CaCO3, and did not show a significant decrease throughout the entire operation. BSF2, BSF3 and BSF4 showed a reduction in alkalinity concentrations of approximately 50%. This suggests that an acid-forming reaction occurred within the filter media.  Figure 6 presents data on ultraviolet light absorption at a wavelength (UV254) of 254 nm. Organic compounds with an aromatic structure or conjugated double carbon bonds are absorbed by UV254 [25]. Filtered absorbance values were reduced by 72%, 62%, 68% and 73%, with respect to influent absorbance for BSF1 to BSF4, respectively. The reductions showed that filters were retaining dissolved organic matter. Lynn et al. [23] measured a reduction in absorbance of 35% in BSF-lower  Figure 6 presents data on ultraviolet light absorption at a wavelength (UV 254 ) of 254 nm. Organic compounds with an aromatic structure or conjugated double carbon bonds are absorbed by UV 254 [25]. Filtered absorbance values were reduced by 72%, 62%, 68% and 73%, with respect to influent absorbance for BSF1 to BSF4, respectively. The reductions showed that filters were retaining dissolved organic matter. Lynn et al. [23] measured a reduction in absorbance of 35% in BSF-lower than the observed in this study.

Nitrate
Nitrate concentrations in filtrates from BSF were consistently lower than those in influent, with reductions varying from 30-53%. The reductions in nitrate suggested that denitrification occurred in the lower parts of the filters, where an oxygen deficit could be present. This observation is compatible with the pH reduction and alkalinity consumption that occurred during filtration. In previous studies, the simultaneous occurrence of nitrification/denitrification in BSF has been suggested due to the availability of oxygen at the top layer, and its deficiency at lower depths [26,27]. From day 0 to day 45, the TC reduction efficiencies for BSF2, BSF3 and BSF4 were 64.2%, 69.2% and 92.0%, respectively. From day 45 onwards, the operation cycle of the filters changed from a 24 h-24 h to a 24 h-48 h resting period. In Figure 7, it is possible to see the improvement in TC removal-98.4%, 98.1% and 99.8% for filters BSF2, BSF3 and BSF4, respectively. The higher efficiency in TC reduction implies that the BSF worked better with the change of the pause period from 24 h to 48 h. The improvements could also be related to the filter ripening.

Nitrate
Nitrate concentrations in filtrates from BSF were consistently lower than those in influent, with reductions varying from 30-53%. The reductions in nitrate suggested that denitrification occurred in the lower parts of the filters, where an oxygen deficit could be present. This observation is compatible with the pH reduction and alkalinity consumption that occurred during filtration. In previous studies, the simultaneous occurrence of nitrification/denitrification in BSF has been suggested due to the availability of oxygen at the top layer, and its deficiency at lower depths [26,27]. wastewater effluent. In general, rises and declines in TC LRV in BSF2, 3 and 4 occurred together and independently from BSF1. BSF4, which had the highest level of turbidity (50 NTU), performed better than the other BSF in terms of coliform reduction. Its efficiency was comparable to the typical interval of 93-99% documented by other researchers [8,[28][29][30][31].

Total Coliform Reduction
To analyze the influence of turbidity on the removal of microorganisms, and to be able to differentiate the results according to BSF, an adjusted generalized mixed model was used with statistical analysis system (SAS) statistical software. The total coliforms variable did not reach the assumption of normality, and a log-normal distribution was used to generate the mixed generalized linear model. Figure 8 shows the results in terms of the variation in the mean value of the TC concentration of the four BSF at Time 1 (before filtration) and Time 2 (after filtration). It can be seen in this graph that the variation in TC concentrations during the two periods of BSF1 did not change significantly. However, the mean values of the responses of the other BSFs showed differences, meaning that there is statistical evidence to state that at least one filter average, time and filter interaction with time differed from the others. From day 0 to day 45, the TC reduction efficiencies for BSF2, BSF3 and BSF4 were 64.2%, 69.2% and 92.0%, respectively. From day 45 onwards, the operation cycle of the filters changed from a 24 h-24 h to a 24 h-48 h resting period. In Figure 7, it is possible to see the improvement in TC removal-98.4%, 98.1% and 99.8% for filters BSF2, BSF3 and BSF4, respectively. The higher efficiency in TC reduction implies that the BSF worked better with the change of the pause period from 24 h to 48 h. The improvements could also be related to the filter ripening.
In day 43, there were TC LRV drops in BSF 2, 3 and 4. In the same day, there was a rise in LRV for BSF1. There was no specific reason that could justify the decreases. Coliforms analyses have some variability, but it was unlikely to be the explanation.
In days 9 to 39, and 43 to 65, there were drops in TC LRV in BSF1. This filter, fed with well water, had influent with lower TC concentrations than BSF2, 3 and 4, which were contaminated with treated wastewater effluent. In general, rises and declines in TC LRV in BSF2, 3 and 4 occurred together and independently from BSF1. BSF4, which had the highest level of turbidity (50 NTU), performed better than the other BSF in terms of coliform reduction. Its efficiency was comparable to the typical interval of 93-99% documented by other researchers [8,[28][29][30][31].
To analyze the influence of turbidity on the removal of microorganisms, and to be able to differentiate the results according to BSF, an adjusted generalized mixed model was used with statistical analysis system (SAS) statistical software. The total coliforms variable did not reach the assumption of normality, and a log-normal distribution was used to generate the mixed generalized linear model. Figure 8 shows the results in terms of the variation in the mean value of the TC concentration of the four BSF at Time 1 (before filtration) and Time 2 (after filtration). It can be seen in this graph that the variation in TC concentrations during the two periods of BSF1 did not change significantly. However, the mean values of the responses of the other BSFs showed differences, meaning that there is statistical evidence to state that at least one filter average, time and filter interaction with time differed from the others. Water 2020, 12, x FOR PEER REVIEW 12 of 19 Figure 8. Analysis of variance for repeated measures by mixed models of total coliforms (TC).
As seen in Table 3, there is statistical evidence to state that at least one filter interaction with time (BSF*Time) differed from the others due to the fact that p is less than 0.05. Since some of the average values for the response variable (TC) of the BSF interactions with time were different from the others, a post-hoc test with the Tukey Kramer adjustment was performed, to determine where these differences were.
As shown in Table 4, for Time 1 (before filtration), BSF1 was significantly different from all other BSF, since all p values were less than 0.05. This was to be expected, since the water that was treated by BSF1 had a different origin than the others, namely, a well. At Time 1, there were no significant differences in the means of BSF2, BSF3 and BSF4, which were the filters that received influent water with similar concentrations of microorganisms: 1706, 1589 and 1779 MPN/100 mL, respectively.
For Time 2 (filtered water), the means of the response variables between BSF1 and other BSFs were not significant, because p values were always greater than 0.05. The interaction between BSF2 and BSF3, which had TC concentrations of 99 and 123 MPN/100 mL, respectively, was not considered significant (p > 0.05). However, the mean concentration of BSF4, 17 NMP/100 mL, was significantly different from that of BSF3. As seen in Table 3, there is statistical evidence to state that at least one filter interaction with time (BSF*Time) differed from the others due to the fact that p is less than 0.05. Since some of the average values for the response variable (TC) of the BSF interactions with time were different from the others, a post-hoc test with the Tukey Kramer adjustment was performed, to determine where these differences were.
As shown in Table 4, for Time 1 (before filtration), BSF1 was significantly different from all other BSF, since all p values were less than 0.05. This was to be expected, since the water that was treated by BSF1 had a different origin than the others, namely, a well. At Time 1, there were no significant differences in the means of BSF2, BSF3 and BSF4, which were the filters that received influent water with similar concentrations of microorganisms: 1706, 1589 and 1779 MPN/100 mL, respectively.
For Time 2 (filtered water), the means of the response variables between BSF1 and other BSFs were not significant, because p values were always greater than 0.05. The interaction between BSF2 and BSF3, which had TC concentrations of 99 and 123 MPN/100 mL, respectively, was not considered significant (p > 0.05). However, the mean concentration of BSF4, 17 NMP/100 mL, was significantly different from that of BSF3.    From day 0 to 35, the average E. coli reduction efficiencies for BSF2, BSF3 and BSF4 were 78.4%, 89.7% and 92.7%, respectively. After day 35, the mean reductions for the same BSFs increased to 97.9%, 99.4% and 99.8%, respectively. These results showed that the BSFs worked better from day 35 onwards. According to the CAWST [13], the time required for the biofilm to mature is 30 days. It is worth noting that BSF 3 and 4, which had higher influent turbidity, were more efficient than BSFs with lower turbidities. BSF4, in particularly, with a turbidity of 50 NTU, performed better than From day 0 to 35, the average E. coli reduction efficiencies for BSF2, BSF3 and BSF4 were 78.4%, 89.7% and 92.7%, respectively. After day 35, the mean reductions for the same BSFs increased to 97.9%, 99.4% and 99.8%, respectively. These results showed that the BSFs worked better from day 35 onwards. According to the CAWST [13], the time required for the biofilm to mature is 30 days.

Escherichia coli
It is worth noting that BSF 3 and 4, which had higher influent turbidity, were more efficient than BSFs with lower turbidities. BSF4, in particularly, with a turbidity of 50 NTU, performed better than all other BSFs. The E. coli reductions observed in this study compare well with the typical reduction interval of 93-99% documented by other researchers [8,[29][30][31].
To analyze the influence of turbidity on the removal of E. coli and to be able to differentiate the results with respect to BSF, an adjustment of the generalized mixed models was used. The E. coli variable reached the assumption of normality, so it was decided to use the gamma distribution to generate the mixed generalized linear model.
For E. coli, the results were slightly different from the results found for total coliforms. Figure 10 shows the variations in the average values of the E. coli concentrations of the four BSFs at Time 1 (before filtration) and Time 2 (after filtration). The variation in the concentration of E. coli during the two periods of BSF1 did not change significantly. As mentioned before, the well water had low concentration of E. coli, with a geometric mean of 15 MPN/100 mL.  [8,[29][30][31].
To analyze the influence of turbidity on the removal of E. coli and to be able to differentiate the results with respect to BSF, an adjustment of the generalized mixed models was used. The E. coli variable reached the assumption of normality, so it was decided to use the gamma distribution to generate the mixed generalized linear model.
For E. coli, the results were slightly different from the results found for total coliforms. Figure 10 shows the variations in the average values of the E. coli concentrations of the four BSFs at Time 1 (before filtration) and Time 2 (after filtration). The variation in the concentration of E. coli during the two periods of BSF1 did not change significantly. As mentioned before, the well water had low concentration of E. coli, with a geometric mean of 15 MPN/100 mL.  Table 5 shows that there was no significant difference for the interaction of BSF and Time (BSF*Time) for the response variable E. coli concentration. There were significant differences between Time 1 and Time 2, which was expected since the influent water quality (Time 1) greatly differed from those of the BSF filtrates (Time 2). It is possible that there was no significant difference in the interaction of BSF and Time, because all BSFs had similar removal efficiencies. This might be due to the low E. coli concentration in all filtrates: 1, 22, 4, and 3 MPN/100 for BSF1, BSF2, BSF3 and BSF4, respectively.   Table 5 shows that there was no significant difference for the interaction of BSF and Time (BSF*Time) for the response variable E. coli concentration. There were significant differences between Time 1 and Time 2, which was expected since the influent water quality (Time 1) greatly differed from those of the BSF filtrates (Time 2). It is possible that there was no significant difference in the interaction of BSF and Time, because all BSFs had similar removal efficiencies. This might be due to the low E. coli concentration in all filtrates: 1, 22, 4, and 3 MPN/100 for BSF1, BSF2, BSF3 and BSF4, respectively.  Figure 11 shows the concentrations of bacteria in raw and filtered water in BSF4. There were significant reductions for all bacteria. The CFU of P. aeruginosa was not detected in the filtered water in all repetitions of the tests. The logarithmic removal units for E. coli, S. Typhimurium, E. faecalis, and TC were 2.7, 3.6, 2.8 and 3.0, respectively. The log removals for these bacteria were similar to those observed for E. coli in BSF4 after day 35 (Figure 9).  Figure 11 shows the concentrations of bacteria in raw and filtered water in BSF4. There were significant reductions for all bacteria. The CFU of P. aeruginosa was not detected in the filtered water in all repetitions of the tests. The logarithmic removal units for E. coli, S. Typhimurium, E. faecalis, and TC were 2.7, 3.6, 2.8 and 3.0, respectively. The log removals for these bacteria were similar to those observed for E. coli in BSF4 after day 35 (Figure 9).

Scanning Electron Microscopy
The top layers of sand from the BSF4 were visualized by SEM to monitor the growth of biofilms on the surfaces of the grains. At day 0, the grains were clean, without any material deposited ( Figure  12a,b), after 45 days of operation the surface layer of sand was still clean, but a very small deposit could be noticed. On day 90, some grains showed formation of a small layer on their surfaces, which could be the initial development of biofilm. The SEM images showed little biofilm formation on the grain surfaces, which is not in accordance with the literature on either slow sand filters or biosand filters [10,13].

Scanning Electron Microscopy
The top layers of sand from the BSF4 were visualized by SEM to monitor the growth of biofilms on the surfaces of the grains. At day 0, the grains were clean, without any material deposited (Figure 12a,b), after 45 days of operation the surface layer of sand was still clean, but a very small deposit could be noticed. On day 90, some grains showed formation of a small layer on their surfaces, which could be the initial development of biofilm. The SEM images showed little biofilm formation on the grain surfaces, which is not in accordance with the literature on either slow sand filters or biosand filters [10,13].

Conclusions
Four biosand filters were tested to evaluate the influence of turbidity on their ability to remove microorganisms present in feces. One filter received well water while the other three received water contaminated with treated wastewater with varying turbidity. Filters operated for 90 days and were capable of providing filtrates with turbidity that were almost always less than 1 NTU.
There were decreases in pH and alkalinity in filters contaminated with treated wastewater. Additionally, the nitrate concentrations were reduced, suggesting that denitrification might have occurred in the filter depth. The UV254 nm absorbance, which is related to the presence of organic matter, showed average removal efficiencies between 62 and 73%.
BSF had a reduction in total coliforms ranging from 0.54 to 2.01 LRV. BSF4, which received water with 50 NTU, showed the highest removal efficiency throughout the experiment, reaching up to 3 LRV at its end. It was found statistical evidence to state that the efficiency of BSF4 was significantly

Conclusions
Four biosand filters were tested to evaluate the influence of turbidity on their ability to remove microorganisms present in feces. One filter received well water while the other three received water contaminated with treated wastewater with varying turbidity. Filters operated for 90 days and were capable of providing filtrates with turbidity that were almost always less than 1 NTU.
There were decreases in pH and alkalinity in filters contaminated with treated wastewater. Additionally, the nitrate concentrations were reduced, suggesting that denitrification might have occurred in the filter depth. The UV 254 nm absorbance, which is related to the presence of organic matter, showed average removal efficiencies between 62 and 73%.
BSF had a reduction in total coliforms ranging from 0.54 to 2.01 LRV. BSF4, which received water with 50 NTU, showed the highest removal efficiency throughout the experiment, reaching up to 3 LRV at its end. It was found statistical evidence to state that the efficiency of BSF4 was significantly greater than those from BSF1 (well water), BSF2 (3 NTU) and BSF3 (25 NTU). Reductions in E. coli ranged from 1.2 to 2.2 LRV. BSF4 also showed a higher removal efficiency throughout the experiment.
The experiment in which the BSF4 influent, which had higher turbidity, was contaminated with the bacterial species S. Typhimurium, E. faecalis, E. coli, and P. aeruginosa had filtrates with 2.7-3.6 LRV.
During most of the experimental period, the filter that received influent water with higher turbidity, induced by kaolin addition, consistently showed filtrates with lower concentrations of microorganisms. It is possible that kaolin particles provided a surface to which organisms adhered, with subsequent particle retention during filtration. The presence of aluminum atoms in the kaolin chemical structure may have a secondary role as coagulant, thus helping to remove microorganisms.
The biosand filter showed its capability to remove fecal microorganisms under different turbidity conditions. This is a reliable point-of-use water treatment. However, in some cases, filtrates still had some microorganisms. As an additional step for increasing microbiological safety, a complementary point-of-use disinfection step can be used.