This paper presents the results of a study of four full-scale upflow gravel filters that are part of full-scale multi-stage filtration. The study explored the design criteria, the operation and maintenance (O&M) practices, and the performance of the systems. Findings showed that most design criteria and O&M procedures are following the recommendations as presented in the literature but several diversions were also identified. Performance data showed that removal efficiencies were on the low side when compared to the literature, possibly because of the good influent quality water that was treated. Cleaning efficiency was analyzed and the overall conclusion is that an adjustment of the design criteria and O&M procedures is needed to enhance system performance. This includes drainage system design, surface cleaning by weir, and filter bed cleaning to allow a reduction in cleaning cycles and an improvement in operation control.

INTRODUCTION

Upflow gravel filtration (UGF) is an important component in multi-stage filtration (MSF) systems, particularly because it protects slow sand filters (SSF) from receiving high loads of suspended solids. The main development of UGF technology emerged in Colombia in the 1980s, where it was introduced first at technical and thereafter at full scale (Galvis et al. 1999). In 2005, more than 140 MSF systems existed in Colombia (Visscher 2006), and to date the number surpasses 200. In the Valle del Cauca region, about 25% of the rural water treatment plants use MSF.

A UGF consists of a box, or a series of boxes, filled with gravel where the water enters from below and flows out from the top (Figure 1). During this passage, impurities are retained in the filter. When filters are cleaned, accumulated solids are removed through gravity flow by opening the drainage valve. The gravel has a large surface area where particles can be retained by sedimentation (main removal mechanism) and attachment (Boller 1993; Galvis 1999), thus facilitating long filter runs. Operation of a UGF involves the control of the filtration velocity, the head loss over the filter and effluent water quality. O&M mainly comprises control of the filtration velocity, head loss, influent and effluent water quality as well as different types of cleaning procedures such as gravel surface cleaning and filter bed cleaning, which may be undertaken daily, weekly, monthly or even less frequently (Galvis et al. 1999).

Figure 1

Schematic overview of a UGF system with different gravel layers. , monitoring water quality parameter: turbidity, Escherichia coli, total coliforms, pH; head loss measure; : surface cleaning; : drainage during filter bed cleaning.

Figure 1

Schematic overview of a UGF system with different gravel layers. , monitoring water quality parameter: turbidity, Escherichia coli, total coliforms, pH; head loss measure; : surface cleaning; : drainage during filter bed cleaning.

Several systems already operate for a long time making it relevant to evaluate the robustness of design, operation and maintenance (O&M) procedures and performance of such systems, particularly because most systems are managed by local water committees. Therefore, these issues are addressed in this paper, as well as comparing practice with the criteria and procedures recommended in the literature.

MATERIALS AND METHODS

Approach

Four treatment systems were selected near to Cali, Colombia. These systems were selected because they represent different situations that together make up a large part of the UGF systems currently available in Colombia. Differences include social and community conditions (middle and low-income communities), gravity versus pumped systems (three gravity systems and one pumped system), surface sources with and without storage reservoir, and differences in O&M practices. The analysis explored the design criteria applied in the systems, the O&M procedures that are used and the UGF performance, including the treatment efficiency for turbidity, Escherichia coli, total coliforms and also total suspended solids (TSS) removal, hydraulic behavior and cleaning efficiency. The analysis was further based on multiple sources of evidence, e.g., observation, interviews, water analysis and the literature (Yin 1989). This enhances the validity of the findings by triangulation (Stake 1995), which consists of using a combination of methodologies to study the same phenomenon, thus making it possible to compare, enrich the interpretations, and contrast data from different sources. In this case, feedback from system operators was used to check whether performance and conditions during the short research period deviated from the normal situation. Although the research of a reduced number of systems has limitations for the generalization of findings, the four selected systems are still fairly representative for a much larger number of UGF systems that treat water from surface water sources in Colombia.

Design criteria were established by reviewing drawings and physical inspection. O&M procedures were reviewed by looking at operational instructions (if available), observation and interviews with operators. Water quality parameters were measured in the UGF (see Figure 1, points (1) and (2)); surface cleaning was observed (point 3) and filter bed cleaning was monitored (point 4). The flow was measured by a calibrated triangular weir installed in the inlet channel of the UGF units. Samples of the filter material were sieved to verify gravel size and porosity was determined following the procedure described by Ives (1990). Head loss (hf) over the UGF was measured daily over 7 days, covering a full cycle of operation between two cleanings.

Turbidity and TSS in the influent and effluent of UFG were measured daily during a 2-week site visit and were used to estimate TSS accumulation during the filter run. This was compared with the TSS measured in the cleaning process. Discharge during cleaning was measured at point (4) (Figure 1). To observe possible differences in cleaning efficiency, the standard cleaning procedure with shock loading by opening and quickly closing the drainage valve (some 10 times) was compared with an uninterrupted drainage process (which is easier for the operator).

Water quality analyses

Influent and effluent quality was monitored in the UGF units at each treatment plant, looking at the following parameters: TSS, turbidity, E. coli, total coliforms and pH (see Figure 1, points (1) and (2)), using Standard Methods (APHA AWWA & WPCF 2005). During filter bed cleaning, turbidity and TSS were measured and the correlation was verified.

Filter bed cleaning

To obtain more insight into filter bed cleaning, the following procedure was followed: (1) the surface area of each filter was measured (A); (2) the declining water level in the filter (Δh) was measured over time (t); and (3) the washing velocity was set by the expression Q = Δh*A/t (m3/s). The drop in the water level in the filter was measured in the inlet pipe until the filter was empty.

Hydraulic behavior of UGF units

The hydraulic behavior of the UGF units was established by applying tracer tests with sodium chloride, which makes it possible to determine the presence of dead zones resulting from the hydraulic design and possible permanent clogging. The concentration curve of the tracer was analyzed using the mathematical simulation models of Wolf-Resnick, the Morril index, and the model of completely mixed reactors in series (CMRS) (Sánchez et al. 2012).

Description of the UGF systems

The four full-scale MSF plants are described in Box 1. All systems include dynamic gravel filtration, except for El Retiro, which was selected because this system is preceded by a reservoir (4,000 m3) to prevent peak loads of suspended solids reaching the UGF. All plants have UGF in layers as a secondary filtration stage. The system in La Sirena was selected because it has two stages of UGF, both of which are in layers. This diverts from what is described in the literature as a two-stage UGF with two filtration stages of different gravel size with crushed gravel, which is different to the other systems that use river cobble. The plant in Arroyohondo was selected because it has a special feature in that it makes it possible to dose a coagulant prior to the UGF when turbidity is high, to stimulate coagulation and flocculation in the UGF (Sánchez et al. 2012). All plants have SSF as the final filtration stage. Golondrinas is a typical system located in a mountainous rural area with deforestation problems in the watershed. A summary of the treatment plant components is shown in Table 1.

Box 1
The four MSF systems included in this study

The MSF system in El Retiro replaced a conventional water treatment plant with rapid filtration in 1987; the system provides water to a better-off neighborhood with 500 inhabitants and a number of private schools. The system was financed through the tariff and is managed by a team of operators supervised by a users committee.

The MSF system in Arroyohondo replaced a compact conventional water treatment plant with rapid filtration. In this MSF, it proved possible to use coagulation and flocculation in combination with UGF, which enhances the flexibility to respond to variations in turbidity (Sánchez et al. 2012). The system was built in 2005 with financial resources raised by local organizations and communities. Today, it supplies water to 840 inhabitants.

The MSF system in La Sirena was built in 1988 in response to several cases of cholera that occurred in the community. Initially it only comprised SSF, but subsequently this was transformed into an MSF system to cope with the deterioration of water quality in the watershed. It provides water to 4,500 inhabitants of a low-income settlement. It is managed by a water committee and was built with support from central and local governments and a small grant from the Dutch Embassy.

The MSF system in Golondrinas is located in a remote, rural low-income community. It provides water to 2,500 inhabitants, is managed by a water committee and was built in 2005 with financial resources from central and local governments.

Table 1

Treatment plant components

  Pretreatment system
 
SSF
 
Treatment plant Flow (L/s) Type A (m2Filter length (m) ɛ (%) A (m2vf (m/h) 
El Retiro 20 Reservoir 2,000 – – 480 0.15 
UGFL 28 1.6 44–46 
La Sirena 10 DyGF 9.0 0.6  240 0.20 
UGFS2 17.7 2.2 54–56 
Arroyohondo DyGF 5.4 0.60  72 0.15–0.30 
UGFL 10.6 1.05 38–42 
Golondrinas DyGF 8.1 0.60  216 0.15 
UGFL 23.1 1.1 42–43 
  Pretreatment system
 
SSF
 
Treatment plant Flow (L/s) Type A (m2Filter length (m) ɛ (%) A (m2vf (m/h) 
El Retiro 20 Reservoir 2,000 – – 480 0.15 
UGFL 28 1.6 44–46 
La Sirena 10 DyGF 9.0 0.6  240 0.20 
UGFS2 17.7 2.2 54–56 
Arroyohondo DyGF 5.4 0.60  72 0.15–0.30 
UGFL 10.6 1.05 38–42 
Golondrinas DyGF 8.1 0.60  216 0.15 
UGFL 23.1 1.1 42–43 

#: units number; A: area; ɛ: porosity; UGFS2: upflow gravel filtration in series with two stages; DyGF: dynamic gravel filtration; UGFL: upflow gravel filtration in layers.

All UGF units are made from reinforced concrete. Drainage systems, consisting of perforated polyvinyl chloride (PVC) pipes, are placed at the bottom of the structure, serving both to distribute the flow during filtration and to discharge the water during periods of cleaning.

RESULTS AND DISCUSSION

Design characteristics of the UGF systems

The design characteristics of the systems are shown in Table 2, which also includes the guideline values given by Galvis et al. (1999).

Table 2

Design criteria applied and design criteria recommended for each upflow gravel filtration

  Treatment plant
 
Criterion Guide El Retiro La Sirena Arroyohondo Golondrinas 
Design period (years) 8–12 15 15 15 15 
Period of operation (h/d) 24 24 24 24 24 
Filtration velocity (m/h) 0.3–0.6 0.64 0.67 0.45–0.9 0.7 
Number of stages      
UGFL  
UGFS 2–3    
Filter bed      
Length of gravel bed (m)      
UGFL 0.6–0.9 1.0  0.75 0.80 
UGFS 1.15–2.35  1.6a   
 Size (mm) 1.6–25 4.0–28 4.0–28 3.2–25 2.2–25 
Support bed      
 Length (m) 0.3 0.5 0.3 0.30 0.30 
Supernatant water height (m) 0.1–0.2 0.05 0.10 0.05 0.10 
Minimum static load of washing flow (m) 3.0 2.2 4.0 1.62 1.55 
Area per filtration unit (m2<20 28 17.7 10.6 23.1 
Initial washing velocity (m/h) >10 5.4 10.2 10.4 5.4 
  Treatment plant
 
Criterion Guide El Retiro La Sirena Arroyohondo Golondrinas 
Design period (years) 8–12 15 15 15 15 
Period of operation (h/d) 24 24 24 24 24 
Filtration velocity (m/h) 0.3–0.6 0.64 0.67 0.45–0.9 0.7 
Number of stages      
UGFL  
UGFS 2–3    
Filter bed      
Length of gravel bed (m)      
UGFL 0.6–0.9 1.0  0.75 0.80 
UGFS 1.15–2.35  1.6a   
 Size (mm) 1.6–25 4.0–28 4.0–28 3.2–25 2.2–25 
Support bed      
 Length (m) 0.3 0.5 0.3 0.30 0.30 
Supernatant water height (m) 0.1–0.2 0.05 0.10 0.05 0.10 
Minimum static load of washing flow (m) 3.0 2.2 4.0 1.62 1.55 
Area per filtration unit (m2<20 28 17.7 10.6 23.1 
Initial washing velocity (m/h) >10 5.4 10.2 10.4 5.4 

aTwo stages of 0.8 m.

Some design characteristics (filter length, number of stages and period of operation) are in line with Galvis et al. (1999). Gravel sizes, however, are different and the observed filtration velocities were all above the recommended levels (0.6 m/h). In El Retiro and Golondrinas, the minimum filtration area per upflow gravel filtration in layers (UGFL) unit is over 20 m2, which may influence the washing efficiency. Important differences also exist for the minimum static head (difference between supernatant water level and the outlet pipe in the drainage chamber (Ht, Figure 1)) only matches the criteria in La Sirena, which is part of the hydraulic design of drainage system, to ensure sufficient initial washing velocity. This velocity was low in two systems, showing deficiencies in the design of these two systems.

Rulers to measure flow rate and head loss were missing in all systems. The absence of these tools suggests that the operators and their supervisors did not grasp the importance of either flow control to avoid overloading or head loss measurement to follow the clogging process.

Operation and maintenance as practiced in the systems

O&M procedures were compared with the procedures proposed in the literature (Table 3). All systems were operated based on visual inspection of the water, closing the inlet if the operator observes that the turbidity is too high. Flow velocity, head loss and turbidity were not measured.

Table 3

Qualitative comparison of applied and recommended operation and maintenance activities

  Treatment plant
 
Activity recommended El Retiro La Sirena Arroyohondo Golondrinas 
Daily operation 
 Flow measurement and adjustmenta No No No No 
 Turbidity measurement Yes No Yes No 
 Head loss measurementb No No No No 
 Remove any floating material Yes Yes Yes Yes 
 Record of turbidity Yes No Yes No 
Weekly maintenance 
 Cleaning walls of the inlet and outlet chamber Yes Yes Yes Yes 
 Hydraulic filter cleaning (filter draining)c Yes Yes Yes Yes 
 Restarting the UGF Yes Yes Yes Yes 
 Checking of filter cleaning efficiency Yes Yes Yes Yes 
Monthly maintenance 
 Gravel surface cleaningd Weekly Weekly Weekly Weekly 
 Implement normal cleaning Weekly Weekly Weekly Weekly 
Less frequent 
 Gravel bed removing, cleaning and put back into the unit No No No No 
  Treatment plant
 
Activity recommended El Retiro La Sirena Arroyohondo Golondrinas 
Daily operation 
 Flow measurement and adjustmenta No No No No 
 Turbidity measurement Yes No Yes No 
 Head loss measurementb No No No No 
 Remove any floating material Yes Yes Yes Yes 
 Record of turbidity Yes No Yes No 
Weekly maintenance 
 Cleaning walls of the inlet and outlet chamber Yes Yes Yes Yes 
 Hydraulic filter cleaning (filter draining)c Yes Yes Yes Yes 
 Restarting the UGF Yes Yes Yes Yes 
 Checking of filter cleaning efficiency Yes Yes Yes Yes 
Monthly maintenance 
 Gravel surface cleaningd Weekly Weekly Weekly Weekly 
 Implement normal cleaning Weekly Weekly Weekly Weekly 
Less frequent 
 Gravel bed removing, cleaning and put back into the unit No No No No 

aVisual adjustments, but there are no records.

bVisual inspection of the water level is done in the inlet chamber to verify the maximum level, but no record is made.

cHydraulic filter cleaning was performed with successive closures of the fast drainage valve. This is a butterfly valve which facilitates operation.

dDone as part of weekly maintenance.

Weekly cleaning was applied in all systems but operators added the envisaged monthly surface cleaning and carried this out before filter bed cleaning. All operators followed the procedure as suggested in the literature, which entails interrupting the outlet and inlet flows to the unit while maintaining a layer of supernatant water on top of the gravel bed. Surface cleaning was then done manually with a shovel, stirring the surface layer of the filter to remove solid material adhering to the gravel. The supernatant water was discharged with the released solids. In the two systems with orifices, the water discharge is low and much lower than the two systems with overflow weirs, which may result in the removal of fewer solids.

The cleaning procedure by filter drainage also matches the procedures indicated in the literature. Filter units were filled to 20 cm above the gravel bed by opening the inlet valve, thus increasing the static head at the start of the cleaning. During drainage, the butterfly valve on the drain pipe was quickly opened and closed (approximately 10 times). The filter was then filled again from the top and drained, and thereafter put back into operation.

The envisaged occasional extraction and washing of the gravel has never been done in any of the systems according to the operators, and one system has been operating for over 15 years with only weekly cleanings. Table 4 presents additional O&M data for each system. Differences exist in the maximum turbidity levels that operators accept before closing the inlet to avoid turbidity peaks reaching the system. Operator judgement is based on visual inspection (no measurement); interestingly, when water samples were taken it turned out that their visual assessment was quite in line with the indicated levels (Table 4). Frequency and duration of interruptions are low, thus not affecting the continuity of the overall system.

Table 4

Summary of operating, monitoring and maintenance conditions

  Treatment plant
 
Variable Arroyohondo El Retiro La Sirena Golondrinas 
Operational parameters 
 Maximum turbidity (NTU) at inlet (before closing) 30 20 50 60 
 Filter run (d) 
 Operation velocity (m/h) 0.5–1.0 0.64 1.0 0.6 
 Years of operation 15 
Monitoring parameters 
 Number of interruptions per year 11 No 11 15 
 Maximum duration interruption (h) No 
 Maximum head loss in UGF (m) 0.15 0.10 0.20 0.25 
 Head loss before weekly cleaning (m) <0.05 <0.05 <0.05 <0.05 
Maintenance activities 
 Required time (min) 59.8 137.2 116 168.9 
 Peron-h/m2 0.083 0.082 0.109 0.122 
 Discharge method for surface cleaning Weir and channel Weir and channel Orifice Orifice 
  Treatment plant
 
Variable Arroyohondo El Retiro La Sirena Golondrinas 
Operational parameters 
 Maximum turbidity (NTU) at inlet (before closing) 30 20 50 60 
 Filter run (d) 
 Operation velocity (m/h) 0.5–1.0 0.64 1.0 0.6 
 Years of operation 15 
Monitoring parameters 
 Number of interruptions per year 11 No 11 15 
 Maximum duration interruption (h) No 
 Maximum head loss in UGF (m) 0.15 0.10 0.20 0.25 
 Head loss before weekly cleaning (m) <0.05 <0.05 <0.05 <0.05 
Maintenance activities 
 Required time (min) 59.8 137.2 116 168.9 
 Peron-h/m2 0.083 0.082 0.109 0.122 
 Discharge method for surface cleaning Weir and channel Weir and channel Orifice Orifice 

The total time for all maintenance activities was observed and divided by the surface area of the unit (operator h/m2), to be able to compare systems. Maintenance time is highest in Golondrinas and La Sirena, mainly as a result of low drainage velocity during surface cleaning.

Water quality

Water quality monitoring is very limited and only concerns the end product (outflow SSF). In El Retiro, E. coli is monitored daily. Monthly monitoring of the effluent of the SSF is done in El Retiro and Arroyohondo (measurements: turbidity, color, pH and E. coli). No monitoring is applied in the other two systems.

During the site visits, the water quality at the inlet and outlet of the UGF and the outlet of the SSF was additionally monitored for a period of 2 weeks (Table 5). The mean turbidity level of the effluent of the UGF was less than 5 NTU. In all cases, the turbidity of the effluent of the UGF was below 10 NTU, which is the guideline value of inflow water to the SSF units (Galvis et al. 1999; Di Bernardo & Sabogal 2008). The best turbidity removal was obtained in El Retiro. La Sirena showed the worst performance, possibly due to the high filtration velocity and the type of filter material (crushed gravel with a higher porosity, a larger shape factor (8.7) and lower sphericity (0.69)) (Di Bernardo & Sabogal 2008).

Table 5

Water quality

 Parameters (statistics)
 
 Turbidity (NTU)
 
E. coli (log CFU/100 mL)
 
Total coliform (log CFU/100 mL)
 
Treatment plant Stage Av. SD E % Av. SD Red. Av. SD Red. 
La Sirena Influent 1.70 0.58 1.92 0.30  3.65 0.27  
Eff. UGF 1.40 0.41 16 1.60 0.27 0.30 3.44 0.27 0.21 
Eff. SSF 0.26 0.06 80 0.0 0.0 1.60 1.16 0.78 1.30 
El Retiro Influent 4.01 3.16  2.70 0.40  3.89 0.19  
Eff. UGF 1.70 0.99 55 2.23 0.50 0.47 3.61 0.30 0.28 
Eff. SSF 0.40 0.20 71 0.0 0.0 2.20 1.50 1.50 2.10 
Arroyohondo Influent 2.50 1.30  3.35 0.22  4.32 0.45  
Eff. UGF 1.70 0.23 36 3.12 0.23 0.23 4.10 0.45 0.26 
Eff. SSF 0.18 0.06 89 0.0 0.0 2.60 0.85 0.37 3.21 
Golondrinas Influent 5.70 2.60  1.92 0.16  2.80 0.51  
Eff. UGF 3.70 1.30 40 1.37 0.21 0.55 2.49 0.54 0.31 
Eff. SSF 0.60 0.17 78 0.0 0.0 1.37 0.89 0.22 1.60 
 Parameters (statistics)
 
 Turbidity (NTU)
 
E. coli (log CFU/100 mL)
 
Total coliform (log CFU/100 mL)
 
Treatment plant Stage Av. SD E % Av. SD Red. Av. SD Red. 
La Sirena Influent 1.70 0.58 1.92 0.30  3.65 0.27  
Eff. UGF 1.40 0.41 16 1.60 0.27 0.30 3.44 0.27 0.21 
Eff. SSF 0.26 0.06 80 0.0 0.0 1.60 1.16 0.78 1.30 
El Retiro Influent 4.01 3.16  2.70 0.40  3.89 0.19  
Eff. UGF 1.70 0.99 55 2.23 0.50 0.47 3.61 0.30 0.28 
Eff. SSF 0.40 0.20 71 0.0 0.0 2.20 1.50 1.50 2.10 
Arroyohondo Influent 2.50 1.30  3.35 0.22  4.32 0.45  
Eff. UGF 1.70 0.23 36 3.12 0.23 0.23 4.10 0.45 0.26 
Eff. SSF 0.18 0.06 89 0.0 0.0 2.60 0.85 0.37 3.21 
Golondrinas Influent 5.70 2.60  1.92 0.16  2.80 0.51  
Eff. UGF 3.70 1.30 40 1.37 0.21 0.55 2.49 0.54 0.31 
Eff. SSF 0.60 0.17 78 0.0 0.0 1.37 0.89 0.22 1.60 

Eff: effluent; Av: average; SD: standard deviation; E: efficiency; Red: reduction.

The best removal efficiency for E. coli was found in the UGF units in El Retiro and Golondrinas with 66 and 72%, respectively. These plants were operated with a relatively constant flow, following the guidelines. The other systems had larger flow variations and lower removal efficiencies.

Hydraulic behavior of UGF units

Table 6 summarizes the results of the tracer tests for each treatment plant. Results show that the UGFs corresponded to a ‘dual system’ with plug flow and mixed flow while also presenting dead zones, which is consistent with the UGF behavior as reported by Galvis (1999). The UGF in El Retiro had the best performance with the largest portion of plug flow and the lowest fraction of dead zones. The highest fraction of dead zones was found in the systems that do not have a weir (La Sirena and Golondrinas), which suggests that the limitations in surface cleaning had a negative effect on the hydraulic behavior.

Table 6

Results based on the analysis for the Wolf and Resnick and CMRS models

UGF unit Plug flow (%) Mixed flow (%) Dead zone (%) CMRS Morril index RT (min) Washing velocity (m/h) Weir 
El Retiro (vf = 0.6 m/h) 48 50 10 2.5 94 5.4 Yes 
La Sirena (vf = 0.6 m/h) 40 62 6–7 2.8 120 10.2 No 
Arroyohondo (vf = 0.6 m/h) 37 60 2.8 59 10.4 Yes 
Golondrinas (vf = 0.6 m/h) 20 65 15 3–4 4.5 65 5.4 No 
UGF unit Plug flow (%) Mixed flow (%) Dead zone (%) CMRS Morril index RT (min) Washing velocity (m/h) Weir 
El Retiro (vf = 0.6 m/h) 48 50 10 2.5 94 5.4 Yes 
La Sirena (vf = 0.6 m/h) 40 62 6–7 2.8 120 10.2 No 
Arroyohondo (vf = 0.6 m/h) 37 60 2.8 59 10.4 Yes 
Golondrinas (vf = 0.6 m/h) 20 65 15 3–4 4.5 65 5.4 No 

RT: residence time.

The dead zones suggest that some permanent accumulation of solids occurred in the gravel bed. This accumulation was more severe in the systems with limitations in surface cleaning.

Cleaning behavior in UGF units

The TSS concentration during drainage (Figure 2) showed four zones: (1) a first peak of TSS during a high washing velocity; (2) a low concentration of TSS during declining washing velocity; (3) a peak in TSS during low washing velocity; and (4) a low concentration of TSS and a low flow. The first peak results from the high initial flow, which quickly dragged particles to the drainage system. Thereafter, the velocity reduced and fewer particles were dragged. The second peak is most likely the result of air being pulled into the gravel bed, which helped to disturb the particles that remained on top of the grains. Earlier reports on filter cleaning (Wolters 1988; Cinara & IDRC 1993) only reported the first peak. In Arroyohondo, two identical UGF units with the same operation time (7 days) and equal influent water quality were cleaned at the same time: one with shocks and the other only draining the filter. The behavior in terms of TSS removal was very similar in the two units, which suggests that shock loading, by quickly closing the drainage valve, did not have an effect on the TSS removal pattern (Figure 4(d)). This confirms the suggestion of Mataix (2004) and Collins et al. (1994) that stirring of the deposits does not happen because the energy is dissipated by deformation of the pipe and by the viscosity of the water.

Figure 2

TSS behaviors on time during filter bed cleaning in UGF units. (a) UGF La Sirena, (b) UGF Golondrinas, (c) UGF El Retiro, (d) UGF Arroyohondo.

Figure 2

TSS behaviors on time during filter bed cleaning in UGF units. (a) UGF La Sirena, (b) UGF Golondrinas, (c) UGF El Retiro, (d) UGF Arroyohondo.

The highest washing velocities during UGF cleaning were obtained in Arroyohondo and La Sirena and these were in line with those reported by Wolters (1988) and Galvis (1999), but low in comparison to the range of 60–90 m h−1 found by Pardón (1989). For the other two systems, values were much lower, probably due to hydraulic limitations in the drainage system.

The effectiveness of filter bed cleaning was checked by analyzing the quantity of TSS removed during cleaning. Results were compared with turbidity data and showed to have a good linear correlation (TSS = 0.16 (turbidity) +0.138; R2 = 0.93; n = 16). Based on this correlation, the TSS concentration in the drainage water for both drainage cycles was calculated using the data from Figures 2 and 3. Results of the accumulated removal are shown in Figure 4. Furthermore, the mass balance of TSS was established based on TSS values in influent and effluent for the same 7 days to calculate the total amount of solids retained in the filter over the filter run (dotted horizontal line in Figure 4).

Figure 3

Washing velocities in UGF units.

Figure 3

Washing velocities in UGF units.

Figure 4

TSS evacuated during filter bed cleaning on time for each UGF. (a) UGF La Sirena (mean TSS 1.8 mg/L), (b) UGF Golondrinas (mean TSS 2.0 mg/L), (c) UGF El Retiro (mean TSS 0.8 mg/L), (d) UGF Arroyohondo (mean TSS 1.3 mg/L).

Figure 4

TSS evacuated during filter bed cleaning on time for each UGF. (a) UGF La Sirena (mean TSS 1.8 mg/L), (b) UGF Golondrinas (mean TSS 2.0 mg/L), (c) UGF El Retiro (mean TSS 0.8 mg/L), (d) UGF Arroyohondo (mean TSS 1.3 mg/L).

Figure 4 shows that on average, in all UGFs, about 90% of the retained solids were removed by two drainage cycles. The other 10% is expected to be removed by surface cleaning as it has not been necessary to remove the gravel for washing; in three UGFs, the second filter drainage removed a larger quantity of solids than the first. A possible reason may be the relatively low TSS concentration in the influent (0.8–2 mg/L), resulting in a low volume of deposits in the filter.

In a way, the lower initial washing velocity may have facilitated the solids removal. While the peak velocity was lower, a higher velocity was sustained for a longer period (Figure 3), which contributed to solids removal over a longer period of time (Figure 4(b) and 4(c)).

While Pardón (1989) indicates the need for frequent cleaning to avoid permanent clogging of the filters, our findings related to weekly cleaning suggest that cleaning frequency can be even lower. The systems only developed a small head loss after 1 week (<0.05 m). Furthermore, gravel was not removed for cleaning in any of the UGFs because of advanced clogging, and one system had been in operation for 15 years. Hence, it is relevant to explore the cleaning cycles in more detail since reduced frequency reduces the workload of the operator, and reduces water loss, which may be particularly relevant in pumped systems. Longer periods between cleaning may also have a positive effect on treatment efficiency by allowing more biomass development in the filters.

CONCLUSIONS

This paper presents the results of a study of four full-scale UGFs that are part of full-scale MSF systems. The study explored the design criteria that were applied, the way O&M procedures are carried out, and the performance of the systems, including filter bed cleaning. This study shows that in general, the design characteristics of the systems follow the literature with the exception of the drainage system and flow velocities; in two cases this resulted in lower washing velocities than recommended in the literature. Performance data showed that removal efficiencies were on the low side when compared to the literature, possibly because of the good quality influent water that was treated. Head loss and flow measurement are not possible in the systems due to the lack of measurement tools in the UGFs. A weir should be included in the design criteria of UGFs to facilitate water drainage during surface cleaning. Operators follow, to a fair extent, the recommended O&M procedures but they do not: take samples to monitor water quality, measure head loss, or control the flow velocity. Shock loads did not influence cleaning efficiency of the lowly loaded filters, implying that this practice can be replaced by just twice draining the UGFs, thus facilitating the work of the operator. Head loss build up in 1 week was low, suggesting that fewer cleaning cycles may be needed. Results show that the procedures applied for filter bed cleaning are effective despite some limitations found in the drainage systems and low washing velocity. About 90% of the retained solids were removed in two drainage cycles; the remaining 10% is probably removed during surface cleaning of the gravel bed. Adjustment of the design criteria and O&M procedures is needed to enhance system performance. This includes drainage system design, surface cleaning by weir, and filter bed cleaning to allow a reduction in cleaning cycles and to improve operation control.

ACKNOWLEDGEMENTS

The authors are grateful for the support of the communities and the plant operators: Alfonso Escandón and Eulogio Mósquera (Arroyohondo); Gabriel Murillo and Libardo Ortiz (El Retiro); Ever Tosse (La Sirena) and Leonardo Meneses (Golondrinas). Thanks also go to Carlos Mejia and Diana Posso who were involved in the fieldwork, and to Universidad del Valle – Cinara Institute for their support.

REFERENCES

REFERENCES
APHA, AWWA & WPCF
2005
Standard Methods for the Examination of Water and Wastewater
.
American Public Health Association
,
Washington, DC, USA
.
Boller
M.
1993
Filter mechanism in roughing filters
.
Water SRT-Aqua
42
(
3
),
174
185
.
Cinara and International Development Research Centre, IDRC
1993
Proyecto limpieza hidráulica de filtros gruesos, reporte final (Project of Hydraulic Cleaning for Coarse Filters, Final Report)
.
Cali, Colombia
, p.
110
.
Collins
R. M.
Westersund
C. M.
Cole
J. O.
Roccaro
J. V.
1994
Evaluating Roughing Filtration Design Variables
.
AWWA Research Foundation
,
Durham, New Hampshire, US. 179 pp
.
Di Bernardo
L.
Sabogal
L. P.
2008
Selecao de Tecnologias de Tratamento de 'Agua (Selection of Water Treatment Technology)
.
Editora LDIBE LTDA
,
Sao Carlos, Brasil
,
I, 878 pp
.
Galvis
G.
1999
Development and Evaluation of Multistage Filtration Plants: An Innovative, Robust and Efficient Water Treatment Technology
.
PhD Thesis
,
CEHE, University of Surrey
,
Guildford, Survey, UK
, p.
228
.
Galvis
G.
Latorre
J.
Visscher
J. T.
1999
Filtración en Múltiples Etapas, Tecnología Innovativa Para el Tratamiento de Agua (Multistage Filtration Innovative Technology for Water Treatment)
.
Universidad del Valle, Instituto Cinara
,
Cali, Colombia &
International Water and Sanitation Centre, IRC, The Netherlands. UNESCO, United Nations Office for Science and Culture
,
197 pp
.
Ives
K. J.
1990
Testing of filter media
.
J. Water SRT Aqua
39
,
144
151
.
Mataix
C.
2004
Mecánica de Fluidos y Maquinas Hidráulicas (Fluid Mechanics and Hydraulic Machines)
.
Oxford University Press
,
Mexico
Alfaomega group editor, second edition
,
660 pp
.
Pardón
M.
1989
Treatment of Turbidity Surface Water for Small Community Supplies
.
PhD Thesis
.
University of Surrey
,
Guildford, Surrey, UK
.
Sánchez
L. D.
Marin
L. M.
Visscher
J. T.
Rietveld
L. C.
2012
Low-cost multi-stage filtration enhanced by coagulation-flocculation in upflow gravel filtration
.
Drink. Water Eng. Sci.
5
,
73
85
.
Stake
R. E.
1995
The Art of Case Study Research
.
SAGE Publications
,
London
.
Wolters
H.
1988
Roughing Filtration a Literature Study
.
MSc Thesis
,
Delft University of Technology, The Netherlands
, p.
173
.
Visscher
J. T.
2006
Facilitating Community Water Supply Treatment, from Technology Transfer to Multi-Stakeholder Learning
.
IRC International Water and Sanitation Centre
,
Delft, The Netherlands
,
257 pp
.
Yin
R. K.
1989
Case Study Research: Design and Methods
.
Applied Social Research Series, Vol. 5
,
Sage
,
Newbury Park, California, USA
.