The influence of three important parameters, namely, pause time, charge volume and influent turbidity on the performance of a biosand filter (BSF) was assessed by employing response surface methodology with a Box-Behnken design. Fifteen runs were conducted in triplicate using a full-scale BSF with a pore volume of 19.6 L, and the performance of the filter was assessed by monitoring total coliforms, Escherichia coli and turbidity of the effluent. Results showed that there was a significant improvement in the microbiological quality of filtered water when the pause time was increased from 12 to 36 h, while it deteriorated as charge volume was increased from 10 to 30 L. The effluent microbiological quality was unaffected by changes in the influent turbidity in the range of 10–50 NTU. Further, effluent turbidity was not significantly affected by the changes in any of the operating parameters, and it remained below 1.5 NTU in all the runs. At optimum conditions of pause time 23 h, charge volume 19.4 L and influent turbidity 32 NTU, bacterial removal exceeded 99.4%. The results show that charging of the filter with about one pore volume of water once a day would be the ideal operating strategy for a BSF.

INTRODUCTION

The use of contaminated waters for drinking is one of the principal pathways for diseases that kill millions and sicken more than a billion people each year. Technologies commonly employed in developing regions for treating drinking water are typically too expensive, too complex to be locally maintained and repaired, and incompatible with local customs and beliefs (Tellen et al. 2010). Thus self-sustaining, decentralized and affordable technologies that target the poorest and most vulnerable are needed. Access to safe drinking water can be accelerated by provision of affordable technologies for point-of-use water treatment and storage in combination with improved hygiene behaviour (Kubare & Haarhoff 2010). The World Health Organization (WHO) has also identified household water treatment technologies as an option for providing safe water to households (Sobsey et al. 2008). Several different types of systems are available for affordable household water treatment. Most of these involve the use of ceramic, charcoal, or sand filtration and/or chemical or solar disinfection.

One of the most promising and accessible technologies emerging for household drinking water treatment is biosand filtration. A biosand filter (BSF) is an intermittently operated slow sand filter (SSF) on a small scale, making it suitable for households. The filter consists of a bed of fine sand supported by a layer of gravel enclosed in a box with appurtenances to deliver and collect the water (Kubare & Haarhoff 2010). The filter is designed to operate in such a way that a batch, or dose, of influent water is added all at once, creating a pressure head that drives flow. The new dose enters the interstitial spaces of the media, displacing water from the previous dose (Young-Rojanschi & Madramootoo 2014a). The time between one dose and the following is referred to as the residence period or pause period (Jenkins et al. 2011). During the first few weeks of operation, the filter ripens as the schmutzdecke, or filter cake, develops and the filter's capacity to remove microorganisms improves (Young-Rojanschi & Madramootoo 2014b). Sobsey et al. (2008), based on different criteria, identified BSF as the most effective method among the five different household treatment technologies considered, and as having the greatest potential to become widely used and sustainable in the rural areas of developing countries.

Different parameters influencing the performance of a BSF include sand size, hydraulic head, pause time, maintenance practices, charge volume and influent turbidity (Jenkins et al. 2011). Several reports are available in the literature on the effects of these parameters. Baumgartner et al. (2007) showed that pause time and dosing volume significantly affected bacterial removal. They reported that a BSF operated with 12-h residence time gave significantly better total coliform removal compared to a filter operated with 36-h residence time. Jenkins et al. (2011) concluded that BSFs are typically designed to treat 20 L at a time with a maximum head of 17–29 cm. Also, increased residence time emerged as highly beneficial for improving microbial removal. On the contrary, Kennedy et al. (2012) demonstrated that increasing hydraulic residence time by reducing outlet diameter did not produce any significant improvement in faecal coliform removal. Elliott et al. (2008) reported that feed volumes greater than 50% of the filter volume showed decreased incremental removal for Escherichia coli and bacteriophages. A study by Nair et al. (2014a) concluded that doubling the daily charge volume from 20 to 40 L significantly reduced E. coli removal by at least one log10 unit. Young-Rojanschi & Madramootoo (2014b) operated a BSF continuously and intermittently to compare filter efficacy under both conditions. Although BSFs were developed for intermittent operation, the filters performed significantly better when operated continuously. However, both operational modes resulted in a significant removal of microbial indicators. Young-Rojanschi & Madramootoo (2014a) compared E. coli removal in filters operated with 1-, 2-, and 3-day residence periods, and found that filters with increased residence periods had lower dissolved oxygen (DO) concentrations and increased nitrite levels. However, no significant difference in E. coli removal was observed by extending the residence time to 3 days.

As the use of BSF is spreading, it is necessary to determine the conditions that give the optimum performance. The traditional approach to the optimization problem is the one-factor-at-time method, which has several disadvantages. Statistical design of experiments has been used to overcome these problems. The advantages of the design of experiments include the same amount of information in fewer experiments, estimation of interactions between the factors and development of empirical models. The graphical perspective of the design of experiments has led to the response surface methodology (RSM) (Montgomery 2010). This methodology has been successfully used for studying and modelling different water and wastewater treatment processes such as coagulation–flocculation (Ghafari et al. 2009), chlorination (Umar et al. 2011) and the advanced oxidation process (Zhang et al. 2012). While several studies have been reported on the influence of different parameters on the performance of BSFs, no studies have been reported on the use of RSM in BSF performance optimization.

The present study was aimed towards optimization of BSF operation. The influence of three parameters, namely, pause time, charge volume and influent turbidity, was studied using a full-scale BSF by a design of experiments approach to obtain optimum operating conditions that give efficient removal of contaminants. The performance of the filter was assessed by monitoring effluent total coliforms, E. coli and turbidity.

MATERIALS AND METHODOLOGY

BSF

The filter used in this study was designed based on the guidelines of the Centre for Affordable Water and Sanitation Technology (CAWST 2012), and were similar to the filters used by Nair et al. (2014a) (Figure S1 in the Supplementary Materials, available with the online version of this paper). Plastic containers bought from the local market were used in the study. The effective size (D10) and uniformity coefficient (D60/D10) of the fine sand medium were found to be 0.3 mm and 1.8 mm, respectively. The porosity of the fine sand was 39%. Water was present inside the filter before loading the sand to avoid the development of air pockets and short circuiting. A plastic diffuser plate was placed on the top of the filter, which was used to avoid disturbance of the top layer of the filter media during charging of the filter with water.

Influent water

Tap water collected from the tap of the Environmental Engineering Laboratory of the Civil Engineering Department of Sardar Vallabhbhai National Institute of Technology, Surat, India was used for the study. Since the tap water had no/very little bacterial content, spiking of the microorganism was done by adding sewage (0.1% v/v), which gave influent total coliform and E. coli concentrations of 2.0 × 104 ± 220 and 1,820 ± 258 MPN/100 mL, respectively in the tap water (MPN: most probable number). In order to get the required influent turbidity, natural clay was added. The tap water had the following characteristics before spiking with sewage: turbidity 0.5 NTU, pH 8.2, temperature 28 °C, dissolved oxygen 7.1 mg/L, alkalinity 70 mg/L as CaCO3, total hardness 80 mg/L as CaCO3.

Experimental design and filter operation

A Box-Behnken statistical experimental design (BBD) with the RSM was used to investigate the effects of three independent variables, namely, pause time, charge volume and influent turbidity on the performance of the filter. Effluent total coliforms, E. coli and turbidity were the response parameters. BSF is generally operated with a pause time of 24 h, and hence in the RSM design a range of 12–36 h was used. The pore volume of the filter used in the study was around 20 L, and hence for the charge volume a range of 10–30 L was used in the RSM study so as to understand the impact of higher and lower charge volumes on the performance. Since BSF cannot be used for raw water with high turbidity, a moderate influent turbidity range of 10–50 NTU was used in the experimental design. Table 1 presents the independent variables used in RSM along with their coded values. The BBD design with three replicates at the centre is presented in Table 2.

Table 1

Independent variables with their coded values

Variable Factors Coded factor level 
−1 
Pause time (h) 12 24 36 
Charge volume (L) 10 20 30 
Influent turbidity (NTU) 10 30 50 
Variable Factors Coded factor level 
−1 
Pause time (h) 12 24 36 
Charge volume (L) 10 20 30 
Influent turbidity (NTU) 10 30 50 
Table 2

Design matrix along with response values

Run Pause time (h) Charge volume (L) Influent turbidity (NTU) Effluent total coliforms (MPN/100 mL) Effluent E. coli (MPN/100 mL) Effluent turbidity (NTU) Effluent pH Effluent DO (mg/L) 
12 30 30 204 46 1.5 7.95 5.4 
12 10 30 47 11 0.8 8.15 5.0 
12 20 10 170 37 1.1 7.93 4.8 
12 20 50 94 40 0.9 8.23 4.7 
24 10 10 24 0.8 8.14 4.4 
24 30 50 130 24 1.1 8.06 5.0 
24 10 50 28 11 1.1 8.17 4.3 
24 30 10 93 29 1.0 8.16 5.0 
24 20 30 35 11 1.0 8.18 4.4 
10 24 20 30 45 0.9 7.65 4.4 
11 24 20 30 52 11 0.9 7.72 4.5 
12 36 20 50 50 11 0.9 7.81 3.7 
13 36 30 30 104 21 1.4 7.95 4.7 
14 36 20 10 52 15 1.53 7.77 3.8 
15 36 10 30 11 0.95 7.75 3.5 
Run Pause time (h) Charge volume (L) Influent turbidity (NTU) Effluent total coliforms (MPN/100 mL) Effluent E. coli (MPN/100 mL) Effluent turbidity (NTU) Effluent pH Effluent DO (mg/L) 
12 30 30 204 46 1.5 7.95 5.4 
12 10 30 47 11 0.8 8.15 5.0 
12 20 10 170 37 1.1 7.93 4.8 
12 20 50 94 40 0.9 8.23 4.7 
24 10 10 24 0.8 8.14 4.4 
24 30 50 130 24 1.1 8.06 5.0 
24 10 50 28 11 1.1 8.17 4.3 
24 30 10 93 29 1.0 8.16 5.0 
24 20 30 35 11 1.0 8.18 4.4 
10 24 20 30 45 0.9 7.65 4.4 
11 24 20 30 52 11 0.9 7.72 4.5 
12 36 20 50 50 11 0.9 7.81 3.7 
13 36 30 30 104 21 1.4 7.95 4.7 
14 36 20 10 52 15 1.53 7.77 3.8 
15 36 10 30 11 0.95 7.75 3.5 

Fifteen runs were conducted under different operating conditions as given in Table 2. Before conducting the RSM-based runs, the filter was operated for 30 days by feeding daily 20 L of sewage-spiked water. Since it is known that filter efficiency is affected by filtration rate, great care was taken to ensure that all the runs were conducted at similar flow rates, and the flow rate was maintained in a narrow range of 0.28–0.31 L/min throughout the 15 runs. Three different pause times, 12 h, 24 h and 36 h were used in the RSM runs. For 12-h runs, the filter was charged twice daily. For 24-h and 36-h pause times, the successive charges were added after a gap of 24 h and 36 h, respectively. While in the case of 10-L and 20-L charge volumes, the filter was charged with the entire volume, for the 30-L charge volume, the second half of the charge was added 20 min after the first half was added so as to prevent excessive head and high filtration rate. The flow rate was determined from the time to collect 1 L of filtered water immediately after each charge. Each run was conducted thrice and the average values were used in the RSM calculations.

In order to evaluate the contribution of the three independent variables, experimental data were fitted to the following second order polynomial model using Design Expert 8.0 software (Stat-Ease Corporation, Minneapolis, USA): 
formula
1
where y represents predicted response, β0, βi, βii and βij are regression coefficients for intercept, linear, quadratic and interaction coefficients respectively, xi and xj are independent variables, ɛ is the error associated with the experiments and k is the number of variables studied. The regression analysis of the experimental data was prepared using Design Expert 8.0.

Analyses

During the filter runs, influent and effluent samples were analysed for different water quality parameters which included dissolved oxygen concentration, turbidity, temperature, pH and different microbiological parameters. Composite effluent samples were collected during each filter run. Concentrations of total coliforms and E. coli were determined by using the multiple tube technique (MPN method). Ten-fold serial dilution was used with a 5 tube MPN series. Results from only three consecutive dilutions were used to determine the MPN, and the results are expressed as MPN/100 mL. All the tests were conducted in accordance with the techniques described by Standard Methods (APHA 1998). Turbidity was measured using a turbidimeter (Hach 2100P), pH was determined using a pH meter (Hanna pH 209) and DO was measured using a DO meter (ESICO, India).

RESULTS AND DISCUSSION

Model development and validation

Experimental results of the 15 filter runs are presented in Table 2. The observed data were used to compute the models using a second-order polynomial as represented by Equation (1). The models for effluent total coliforms and E. coli concentrations in terms of coded factors were determined as: 
formula
2
 
formula
3
where A = pause time, B = charge volume and C = influent turbidity.

The analysis of variance (ANOVA) technique was used to check the adequacy of the developed models at the 95% confidence level (Tables S1 and S2 in the Supplementary Materials, available with the online version of this paper). The significance of each factor and interactions between them were checked with the help of a Fisher test. The larger the magnitude of the F-value and correspondingly the smaller the ‘p > F’, the more significant are the corresponding model and the individual coefficients (Montgomery 2010). If the ‘p > F’ value is below 0.05, then the model is significant at the 95% confidence interval. The model can also be evaluated by the lack of fit (LOF) test. The LOF test measures the model failure in representing data points in the experimental domain by comparing the residual error to the pure error from the replicated experimental design points (generally the central points in the experimental design) and should be insignificant. If this ratio is greater than the tabulated F-value, it is concluded that there is evidence of a LOF and the model needs improvement (Massart et al. 1998).

From the ANOVA tables (Tables S1 and S2) it is observed that the models for effluent total coliforms and E. coli are significant, i.e. the p + F value for both the models is <0.050. For both the models A, B and A2 are significant model terms. Equations (2) and (3) were developed after eliminating statistically insignificant terms based on the p-value of the coefficient for those terms. The LOF p-value of >0.050 in both the responses indicates insignificant LOF, and LOF should be insignificant for a good model.

The coefficient of determination, R2, determines overall efficiency of model prediction. In this study though R2 and R2adjusted show good correlation with each other, the R2adjusted of both the models are not very close to the corresponding R2 values as one might normally expect. This could be due to inclusion of an insignificant variable, influent turbidity, in the model. Influent turbidity being the main process variable, its exclusion was not possible. ‘Adequate precision’ measures the signal-to-noise ratio. A value greater than 4 is desirable. Adequate precision of both the models is >4, indicating that the model can be used to navigate the design space (Anderson & Whitcomb 2007).

Normal probability plots of residuals (Figure S2, Supplementary Materials, available online) were constructed in order to check the normality assumption. If the residuals are normally distributed, the plot of the residuals will follow a straight line with some moderate scatter indicating that the model is well fitted (Anderson & Whitcomb 2007). No strong departure from straight lines was observed, suggesting normal distribution of the data. A plot of experimental response values versus the predicted response values also helps to evaluate the model (Figure S3, Supplementary Materials, available online). The points should be split evenly along the 45° line. Groups of points above or below the line indicate areas of over- or under-prediction (Anderson & Whitcomb 2007). These figures show good agreement between predicted and actual responses.

Analysis of the results

Figure 1 presents the main effect plots of pause time, charge volume and influent turbidity on effluent turbidity, effluent total coliforms and effluent E. coli generated by the Design Expert software. As can be seen, there was a significant improvement in microbiological water quality as the pause time increased from 12 h to 36 h, with both total coliforms and E. coli showing reduced levels in the effluent. Sufficient time is required for biological and physico-chemical processes to decay or consume the pathogens before the next batch addition (Elliott et al. 2011). Hence, higher pause time would result in better quality of treated water. Jenkins et al. (2011) reported that increasing residence time can increase bacterial removal by 0.050–0.063 log10 units per hour. Baumgartner et al. (2007), however, reported that too long an idle time may reduce the viability of the biological layer as microorganisms rely on the nutrients from the periodic inflow of source water for survival. Further, it should be noted that longer idle/residence time may fail to fulfil the water production requirements of the household. Hence an optimum pause time should be selected. In the present study, microbiological water quality improved with longer pause times, while effluent turbidity was not affected by variation in pause time (Figure 1). CAWST recommends operating BSFs by dosing between one and four times per day, with a minimum residence period of 1 hour and a maximum of 48 hours based on the assumption that residence periods longer than 48 hours will lead to nutrient depletion and starvation of the biolayer (CAWST 2012).
Figure 1

Effect of pause time, charge volume and influent turbidity on effluent E. coli, total coliform and turbidity.

Figure 1

Effect of pause time, charge volume and influent turbidity on effluent E. coli, total coliform and turbidity.

It is evident from Figure 1 that as the charge volume increased from 10 to 30 L the microbial concentrations in the effluent increased. This might be due to the difference in hydraulic loading between the different operating conditions. When a high charge volume is used, water is pushed through the biofilm and sand at a faster rate, resulting in less exposure to both bacterial predation and adsorption. At lower charge volume sufficient exposure time was obtained, as the residence time in the filter is increased, which permits the bacterial removal mechanisms. Effluent turbidity was not significantly affected by the charge volume.

Increased influent turbidity did not significantly influence bacterial removal (Figure 1). Effluent turbidity also remained unaffected by the variations in influent turbidity as it increased from 10 NTU to 50 NTU. Jenkins et al. (2011) reported that increased influent turbidity would result in improved bacterial removal. However, in the present study, increased influent turbidity did not improve bacterial removal in a significant way. According to WHO, drinking water turbidity should not exceed 5 NTU, and for chlorination to be effective, a turbidity of <1 NTU is recommended (WHO 1997). In the present study, in all the runs effluent turbidity values were around 1 NTU, though the influent turbidity varied in the range of 10–50 NTU.

In RSM the visualization of the regression model can be obtained by surface response plots and contour plots. The three dimensional graphical representation of the fitted models is termed a response surface while the two dimensional plot is known as a contour plot. These graphical plots are helpful for understanding the nature of the response at different factor levels (Nair et al. 2014b). Figure 2(a) and 2(b) present the effects of different variables and their interactive effects on effluent total coliforms and effluent E. coli based on Equations (2) and (3). As found in the main effect plots (Figure 1), it can be observed from these figures that an increase in pause time and decrease in charge volume significantly increased total coliforms and E. coli removal, while influent turbidity did not produce a significant effect on the effluent quality.
Figure 2

Response surface plots for (a) effluent total coliform concentration and (b) effluent E. coli concentration.

Figure 2

Response surface plots for (a) effluent total coliform concentration and (b) effluent E. coli concentration.

Optimization

Response surfaces can be analysed to obtain the maximum or minimum responses and the corresponding optimum conditions. With multiple responses, the optimum conditions can be met when all the parameters concurrently meet the desirable criteria (Ghafari et al. 2009). A desirability-function-based approach was used in this study. The desirability function in the ideal case should be equal to 1 but in practical cases should be close to 1 (Montgomery 2010). The Design Expert software provides five options – none, maximize, minimize, target and within range – for choosing the desired goal for each variable and response. Desired goals for pause period, effluent total coliforms and effluent E. coli were set as minimize, charge volume was set as maximize and influent turbidity was set as ‘in range’. The best solution satisfying the above criteria obtained using the Design Expert software is given in Table 3. It can be seen that a charge volume of about 20 L, which is close to the pore volume of the filter media (19.6 L), and a pause time of about 1 day gave the optimum conditions. This is similar to the observations by Elliott et al. (2011). In order to maximize virus reductions in BSF, they suggested that the ratio of maximum volume in a single charge to pore volume of the filter media bed should preferably be no greater than 1:1 (Elliott et al. 2011). In the present study, at optimum conditions, total coliform and E. coli removal were 99.7% and 99.4%, respectively. It may be noted that higher bacterial removals than these were obtained on different runs (for example, runs 5 and 15 – Table 2). Since optimization was done by maximizing charge volume and minimizing pause time, comparatively lower bacterial removals were achieved at optimum conditions.

Table 3

Optimum conditions for total coliforms and E. coli removal

Pause time (h) Charge volume (L) Influent turbidity (NTU) Effluent total coliforms (MPN/100 mL) Effluent E. coli (MPN/100 mL) Desirability 
23 19.4 32 43 11 0.68 
Pause time (h) Charge volume (L) Influent turbidity (NTU) Effluent total coliforms (MPN/100 mL) Effluent E. coli (MPN/100 mL) Desirability 
23 19.4 32 43 11 0.68 

Results of the study showed that increased pause time is highly beneficial for improving effluent microbial quality. The optimization study suggests that a user should charge the filter once a day with a volume of close to the pore volume of the filter. For a household with larger water requirements, it would be advisable to construct a larger filter with higher pore volume keeping the 24-h pause time. It should be mentioned that only three variables influencing the performance of the BSF were included in the present study. Other parameters such as influent bacterial concentration, filtration rate and filter maturity can also be studied using the methodology. Results of these tests would produce a basis for more rational design and better operating practices, and would result in optimized performance of the BSF.

CONCLUSIONS

In this study, optimization of the BSF was carried out considering three important parameters, namely, pause time, charge volume and influent turbidity. Optimization was done using RSM with BBD employing 15 filter runs. Results showed that while the microbiological quality of the treated water was affected by pause time and daily charge volume, aesthetic quality measured as effluent turbidity was unaffected by pause time, daily charge volume and influent turbidity. Increase in pause time and decrease in charge volume resulted in improved bacterial removal. No effluent parameters were significantly affected by the increased influent turbidity. At optimum conditions of 23-h pause time and 19.4 L charge volume, total coliform and E. coli removal of 99.4% and 99.7%, respectively, were obtained.

REFERENCES

REFERENCES
Anderson
M.
Whitcomb
P.
2007
DOE Simplified: Practical Tools for Effective Experimentation
,
2nd edn.
Productive Press
,
New York
.
APHA
1998
Standard Methods for the Examination of Water and Wastewater
,
20th edn
,
American Public Health Association/American Water Works Association/Water Pollution Control Federation
,
Washington, DC
.
Baumgartner
J.
Murcott
S.
Ezzati
M.
2007
Reconsidering ‘appropriate technology’: the effects of operating conditions on the bacterial removal performance of two household drinking-water filter systems. Environ Res. Letters
2
(
2
),
024003
.
CAWST
2012
Biosand Filter Manual: Design, Construction, Installation, Operation, and Maintenance. Centre for Affordable Water and Sanitation Technology, Calgary, Canada
.
Kennedy
T. J.
Hernandez
E. A.
Morse
A. N.
Anderson
T. A.
2012
Hydraulic loading rate effect on removal rates in a biosand filter: a pilot study on three conditions
.
Water Air Soil Pollution
223
,
4527
4537
.
Kubare
M.
Haarhoff
J.
2010
Rational design of domestic biosand filters
.
J. Water Supply: Res. Technol–Aqua
59
,
1
13
.
Massart
D. L.
Vandeginste
B. G. M.
Buydens
L. M. C.
De Jong
S.
Lewi
P. J.
Smeyers-Verbeke
J.
1998
Handbook of Chemometrics and Qualimetrics: Part A
.
Elsevier
,
Amsterdam, The Netherlands
.
Montgomery
D. C.
2010
Design and Analysis of Experiments
,
7th edn
,
Wiley India Pvt. Ltd
,
New Delhi, India
.
Nair
A. T.
Ahammed
M. M.
Davra
K.
2014a
Influence of operating parameters on the performance of household slow sand filter
.
Water Sci. Technol: Water Supply
14
,
643
649
.
WHO
1997
Guidelines for Drinking-Water Quality, 3rd Edition, Incorporating First and Second Addenda
.
World Health Organization
,
Geneva, Switzerland
.
Young-Rojanschi
C.
Madramootoo
C.
2014a
Comparing the performance of biosand filters operated with multiday residence periods
.
J. Water Supply: Res. Technol–Aqua
64
,
157
167
.
Young-Rojanschi
C.
Madramootoo
C.
2014b
Intermittent versus continuous operation of biosand filters
.
Water Res.
49
,
1
10
.

Supplementary data