Designing Iron-Amended Biosand Filters for Decentralized Safe Drinking Water Provision

Abstract

There are ongoing efforts to render conventional biosand filters (BSF) more efficient for safe drinking water provision. One promising option is to amend BSF with a reactive layer containing metallic iron (Fe⁰). The present communication presents some conceptual options for efficient Fe⁰-amended BSF in its fourth generation. It is shown that a second fine-sand layer should be placed downwards from the Fe⁰-reactive layer to capture dissolved iron. This second fine-sand layer could advantageously contain adsorbing materials (e.g. activated carbons, wooden charcoals). An approach for sizing the Fe⁰-reactive layer is suggested based on 3 kg Fe⁰ per filter. Working with the same Fe⁰ load will ease comparison of results with different materials and the scaling up of household BSF to large scale community slow sand filters (SSF).

Abbreviations:

BSF, biosand filter; RZ, reactive zone; WHO, World Health Organization

Introduction

Safe drinking water may derive from surface water or groundwater. Surface water is often polluted with pathogens (e.g. bacteria, viruses). Groundwater is mainly contaminated/polluted by inorganic species (e.g. arsenic, iron, nitrate, uranium). Both surface water and groundwater may be turbid (physical pollution) and contain organic and inorganic contaminants from both natural and anthropogenic origins 1–5. Accordingly, at any location available water may contain biological, chemical and physical contamination. The World Health Organization (WHO) guidelines for drinking water quality 6 outline a preventive management framework for safe drinking water 2, 7–9. The WHO guidelines 6 entail: (i) health based targets, (ii) system assessment from source through treatment to the point of consumption, (iii) operational monitoring of the control measures in the drinking water production, (iv) management plans documenting the system assessment and monitoring plans and (v) a system of independent surveillance that verifies that the above are operating properly.

Whenever a water is polluted, e.g. after the WHO Guidelines 6, it should be rendered safe before consumption. Universal appropriate treatment technologies should be able to efficiently remove all three classes of contamination. For rural and peri-urban areas in the developing world, the treatment system should ideally occur in a single-stage filtration process at household or small community level 2, 7, 10–23. From the available technologies, household biosand filters (BSFs, intermittent slow sand filters) have been tested the most 2, 24–36. Currently, at least 500 000 people are using BSFs to provide safe drinking water 28, 30, 32. However, it has been traceably shown that BSF do not remove all of the pathogens from water. Accordingly, several attempts are tested to improve the efficiency of conventional BSF 14, 16, 20, 27, 28, 37–45. For example, Baig et al. 16 used locally available biomass in their innovative BSF and reported on significant efficiency enhancement. On the other hand, amending BSF with metallic iron (Fe⁰) has been proven beneficial for water treatment 27, 28, 38–40, 45, 46. However, as with most innovations, the early development of Fe⁰-amended BSF is marked by empirical designs 32. Therefore, the Fe⁰-amended BSF technology should now be translated into rational engineering design criteria.

The present communication is a part of a series of theoretical works based on the aqueous chemistry of iron corrosion and aiming at easing research on water treatment using Fe⁰ 47–51. The expansive nature of Fe⁰ oxidative dissolution followed by precipitation of Fe^II/Fe^III species (oxides and hydroxides) at pH > 4.5 52, 53 is properly considered. Much of the impetus for this work has come from the work of Noubactep et al. 12, who have proposed (i) Fe⁰ as a universal agent for save drinking water provision, and (ii) to amend BSF with a reactive zone (RZ) made up of Fe⁰ and inert materials (e.g. sand). Subsequent works have rationalized the mixture of Fe⁰ with (i) inert materials 47, 54, and (ii) reactive but nonexpansive materials 51 for long-term water treatment. A key question that remains is how to avoid that dissolved iron is present in the treated water? An analysis of published data from Khan et al. 37 to Ingram et al. 46 suggests that a fine sand layer must be placed downstream from the RZ containing Fe⁰. Based on this observation several design options are discussed. For the sake of clarity a conventional BSF will be first presented.

Biosand filter (BSF)

The BSF was developed at the University of Calgary in the early 1990s by Manz 27, 28. The BSF is a downscaling of the slow sand filtration (SSF) technology 55 for intermittent water filtration at household level. The BSF has been reported to be efficient at removing chemicals (e.g. iron, manganese and sulphur), pathogens (e.g. bacteria, viruses) and turbidity from low turbid water 27–30. Unlike a ‘simple’ slow sand filter, the BSF has the ability to perform multiple functions as a single unit 4. In a BSF, settlement, straining and filtration act in synergy to remove biological, chemical and physical contamination/pollution. Ideally, safe drinking water is produced.

The most important process in a BSF occurs in a biological layer (biofilm) called the Shmutzdecke 4, 12, 33, 34, 56. The Shmutzdecke develops in the presence of atmospheric oxygen on the surface of the uppermost layer of sand that is responsible for the removal of microorganisms. Ideally, the fine sand layer downwards from the Shmutzdecke is almost anoxic (O₂ free) as O₂ is consumed by microbial activity within the Shmutzdecke. It should be anticipated here that water exiting such an anoxic system is ideal for an Fe⁰ filter (RZ) as volumetric expansion is mostly limited to the formation of Fe₃O₄ with an expansion coefficient (η) of 2.08 52, 53. Under oxic conditions more volumetric expansive iron oxides and hydroxides formed (η ≤ 6.4). As a rule, the larger the η value, the more rapid the loss of hydraulic conductibility (permeability loss) of the RZ.

Design of a BFS

The engineering principles of household BSF is described in several recent communications 2, 4, 27–30, 32, 57–59. All household BSF share five basic design components 57: (i) a good source of sand and gravel, (ii) a bucket filled with 40–75 cm of fine sand (sand bed), (iii) a layer of static standing water (supernatant water), (iv) a maturation time of 14 to 21 days for formation of the Shmutzdecke and (v) a flow control system. Ideally, water should flow freely from the filter (not tap).

Intensive research is still targeted at designing smaller, lighter (portable), less expensive but efficient BSF, e.g. Project DFID 201727 (www.dfid.gov.uk) 4. Alternative BSF designs for better efficiency in aqueous contaminant removal have been tested 2, 4, 45, 46, 58, 59. Tested design parameters include: (i) sand type, sand size and sand depth 31, 32, 60, (ii) maturation time or time to formation of the Shmutzdecke, (iii) elevation head, (iv) standing water depth, (v) filter pause time or daily water throughput, as well as (vi) amendment of conventional BSF with reactive layers containing metallic iron (Fe⁰). The present communication is focused on the design of Fe⁰-amended BSF.

Each of the six named operational parameters enumerated above has been reported to significantly impact the performance of the BSF for water treatment 2, 4, 33, 45, 59. In particular, despite amendment with Fe⁰, resulted BSF are still tested for their capacity for microbial attenuation and the removal mechanism controversially discussed 33, 61, 62. This controversial discussion is not in accordance with the contemporary knowledge on the mechanism of contaminant removal in Fe⁰ beds 14, 15, 63–69 as iron corrosion products should be regarded as ‘collectors’ in BSF 70.

Operating mode of a BSF

The supernatant water layer provides a head of water that is sufficient to drive the water through the filter bed, whilst creating a retention period of several hours for the water 2, 55, 71. A BSF is layered with an underdrain system which provides an unobstructed passage for treated water from the filter bed and supports the sand bed. The outlet flow control maintains submergence of the medium during operation to minimize potential air-binding problems.

Water percolates slowly through the porous sand medium. Thereby, inert particles and microorganisms are removed from the aqueous phase. An algal mat forms on the surface of the sand bed and this is termed Schmutzdecke (a biolayer or biofilm of living organisms). After several months of operation, the surface of a BSF becomes clogged due to the deposition of suspended solids. At this time, cleaning of the filter bed is required. The BSF is cleaned by removing the top 2–3 cm of the sand bed including the Schmutzdecke layer. After a scraping of the sand bed, a re-sanding is necessary with the accompanying maturation time of the formation of a new Schmutzdecke.

Limitations of a conventional BSF

Discounting any design limitations, the BSF is not destined to treat chemical contaminants in general and inorganic contaminants in particular. In fact, the affinity of sand for metal adsorptive removal is very low 72, 73. Accordingly, whenever chemical contamination is suspected, alternative to conventional BSF should be sought. One such an alternative is the amendment of conventional BSF by a reactive layer containing reactive Fe⁰. The suitability of Fe⁰/sand filters for water treatment arises from the fact that iron corrosion products act as collectors 70, 74 for all classes of contaminants 61, 65, 75.

The suitability of Fe⁰-amended BSF for the developing world arises from the fact that the quality of available water is rarely assessed. A perfect illustration is the arsenic crisis in South East Asia where people have consumed As-polluted waters for decades 41, 76. On the other hand, success of conventional BSF and other water treatment technology are not currently validated by biological and/or chemical analysis but by the decrease of the frequency of water born diseases 4, 8, 13, 21, 44. For all these reasons, the need of a technology able at removing all classes of contaminants for water is obvious.

Fe⁰-amended BSF on an historical perspective

The present section will consider only stand-alone Fe⁰ filters for save drinking water provision.

First generation Fe⁰-amended BSF: Removal of chemicals

The efficiency of iron-oxide-coated sand for contaminant removal is well-documented and has been used for water treatment in household filters for decades 20, 42, 77–79. Conventional BSF have been proven efficient to remove dissolved iron from the aqueous phase. Moreover, BSF has been proven more efficient in removing inorganic contaminants (e.g. As) when the inflowing water was rich in dissolved Fe. Accordingly, Khan et al. 37 amended conventional 3-Kolshi filters with a layer of Fe⁰ to achieve better As removal by a BSF-like filter. It is very important to notice that Fe⁰ is added by Khan et al. 37 to produce iron oxides for As removal in filters. Accordingly, the Fe⁰-amended 3-Kolshi filter can be regarded as first generation Fe⁰-amended BSF.

The first generation Fe⁰-amended BSF was proven very efficient but not sustainable because of too rapid loss of hydraulic conductivity 40, 41, 76. As demonstrated in previous works 12, 47–51, 54, ‘diluting’ Fe⁰ with (inert and) nonexpansive materials is a pre-requisite for sustainability. In other words, the first generation Fe⁰-amended BSF was wrongly designed.

Second generation Fe⁰-amended BSF: Removal of pathogens

The second generation Fe⁰-amended BSF was born with the work of You et al. 80. These authors demonstrated that adding Fe⁰ to a sand filter, efficiently removed viruses. The removal was ascribed to pathogen adsorption onto iron (hydr)oxides formed via (anaerobic) iron corrosion. Virus removal by Fe⁰ was rapid and most of the removed viruses were irreversibly bound. The irreversible bounding of viruses by Fe⁰ corresponds to co-precipitation as presented 2 years later by Noubactep 63–69. Strictly, the filter of You et al. 80 is not necessarily a BSF but an Fe⁰ filter for pathogen removal. In fact, the formation a Schmutzdecke layer is not explicitly intended. Rather, Fe⁰ itself is the pathogen removing agent. However, recent research based on You et al. 80 are testing Fe⁰-amended BSF as alternative to conventional BSF 33, 45, 46, 62, 71.

Third generation Fe⁰-amended BSF: Removal of chemicals and pathogens

The third generation Fe⁰-amended BSF was born with the work of Ngai et al. 38. The filter designed by Ngai et al. 38, 39 combines the concept of a BSF with the innovation of a diffuser basin containing reactive Fe⁰. In this design, pathogens are removed mostly by physical straining provided by the fine sand layer (sand bed) 23. Chemical contamination (e.g. As) is removed by adsorption onto iron oxides and hydroxides generated from rusted Fe⁰.

A fundamental mistake was recently discovered in this design 12. In fact, molecular O₂ necessary for the formation of the Schmutzdecke is quantitatively consumed by the over-laying Fe⁰ layer. In other words, a reactive over-laying Fe⁰ makes the BSF inoperative for pathogen removal as no Schmutzdecke is formed. Thus, the obtained filter is at best a first generation Fe⁰ filter and strictly not an Fe⁰-amended BSF because the ‘Fe⁰ unit’ and the ‘BSF unit’ are two different components of the filter. On the other hand a pure Fe⁰ layer (100% Fe⁰) at the entrance of the filter will rapidly clog due to the formation of voluminous corrosion products (η > 2.1) 52, 53.

Fourth generation Fe⁰-amended BSF

The fourth generation Fe⁰-amended BSF was born with the work of Gottinger 43. It is characterized by an RZ containing a mixture of Fe⁰ and an inert material (e.g. sand). Fe⁰ is purposefully mixed with sand to prevent clogging 12. It should be noticed that mixing Fe⁰ and sand is current in the research on Fe⁰ for groundwater remediation 73, 81–85. The present communication is limited to household and small community Fe⁰ filters. However, the results are up-scalable to larger Fe⁰/H₂O systems.

The filter designed by Leupin and Hug 86, 87 fulfilled these criteria but is classed here as the chronology is not an absolute factor. Furthermore, this filter was designed mainly for As removal. Gottinger 43 was the first researcher to mix Fe⁰ and sand on a volumetric basis and test several Fe⁰/sand ratios (e.g. 50:50 and 40:60 and 0:100) in triplicate columns. Rangsivek and Jekel 88–90 tested 10:90, 20:80 and 30:70 Fe/pumice volumetric mixtures for metal removal but the tests were not systematic. They mostly worked with the 10:90 mixture without specifying the rational 88, 90. Independent theoretical calculations 47, 54 have shown that the threshold value of the volumetric proportion of Fe⁰ for which the column RZ is clogged at Fe⁰ depletion is 52%. The calculations were based on the assumption of ideally packed spherical Fe⁰ particles for which an initial porosity of 36% is relevant. Natural sand is never perfectly spherical. Kubare and Haarhoff 32 reported an average porosity of 45% for sand beds. However, the theoretical threshold value for Fe⁰ depletion of 52% (corresponding to 36% porosity) will be considered in discussing the design of Fe⁰-amended BSF. The meaning of this value is that any filter containing more than 52% Fe⁰ (volumetric proportion of the solid phase) will clog before Fe⁰ completely depletes. The excess Fe⁰ amount relative to 52% should be regarded are pure material wastage with the additional curse of shortening the filter service life 54.

Designing Fe⁰-amended BSF

Strategic planning

The presentation above showed that an Fe⁰-amended BSF should be sequenced as follows: (i) supernatant water (e.g. 5 cm), (ii) fine sand (e.g. 40 to 50 cm), (iii) Fe⁰/sand (x cm), (iv) fine sand (y cm), (v) sand (e.g. 5 cm) and (vi) gravel (e.g. 5 cm). The thickness of individual layers is from Lea 57. In the design of Jenkins et al. 71 sand and gravel are replaced by gravel and rock, respectively. The major feature for these layers is the difference in particle size. The above sequence suggests that the depths of the Fe⁰/sand layer and that of the underlying fine sand are yet to be determined (x and y values). However, the paramount question is which Fe⁰ material should be used?

Characterizing the intrinsic reactivity of Fe⁰ materials 43, 91–97 is a key issue for designing Fe⁰-amended BSF. The ideal material should be able to efficiently provide clean water for at least 12 months (1 year). Therefore, all tested materials should be characterized for their intrinsic chemical reactivity 93, 94, 97 and their sphericity in order to enable results comparability. Next to the intrinsic reactivity, used devices should be characterized by 32, 60: (i) their dimensions, e.g. internal diameter and depth of reactive layers for cylindrical columns, (ii) the particle size and the surface state of used materials (e.g. Fe⁰, gravel, pumice, sand), (iii) the mass of Fe⁰ and (iv) the volumetric proportion of Fe⁰ in the reactive layer. The next section will discuss the thickness of the reactive layer.

Thickness of the Fe⁰/sand layer

A cylindrical bed is considered; H is the height and D is the internal diameter. The cylinder contains an RZ with the height H_rz (x value from Section 4.1) and the volume V_rz. H_rz is necessarily lesser than H (x < H or H_rz < H). Beds are supposed to be filled by spherical granular materials. The compactness (or packing density) C (−) is defined as the ratio of the volume of the particles to the total packing volume (V_rz). Considering the granular material as composed of mono-dispersed spheres subjected to soft vibrations, the compactness C is generally considered to be equal to 0.64 for a random close packing 51. It is assumed that the particles are nonporous.

The initial porosity Φ₀ (−) of the RZ and the thickness H_rz of the RZ are, respectively, then given by:

• ((1))

• ((2))

The filling of the bed porosity by iron corrosion products determines the filter service life and has been extensively discussed in previous works 51, 54.

To completely define the x value (H_rz), V_rz has to be correlated to the mass of Fe⁰. V_rz is the volume occupied ideally by Fe⁰ and sand of similar grain size and shape (Eq. (3)). For simplification V_Fe will be expressed as a fraction of V_rz (Eq. (4)).

• ((3))

• ((3a))

• ((4))

• ((4a))

• ((5))

where α_i is the volumetric fraction of each phase, τ (≤0.52) is the volumetric proportion of Fe⁰ relative to the solid phase (Fe⁰ + sand) in the RZ 54. Per definition, α_pore = Φ = 1 − C and C = 1 − Φ. Thus Eq. (5) reads:

• ((5a))

The mass of Fe⁰ necessary to fill the volume V_Fe is given by Eq. (6):

• ((6))

where ρ_Fe is the specific weight of Fe (7800 kg/m³).

Combining Eqs. (2), (5a) and (6) gives the following relationship between H_rz (x value) and m_Fe (Eq. (7)):

• ((7))

Equation (7) is the equation of the RZ. Three major issues should be considered: (i) τ ≤ 0.52 (52% Fe⁰ v/v), (ii) H_rz ≥ 5 cm, and (iii) H_rz < H. The threshold value H_rz = 5 cm is considered the minimum height for the realization of a homogeneous well-mixed Fe⁰/sand RZ 98. Ideally, H_rz is only a fraction of H (H_rz ≪ H) for example, Noubactep et al. 12 considered that H_rz should fulfil the condition H_rz ≤ 0.1H.

The mass of sand to be used is given by Eq. (8).

• ((8))

where ρ_sand is the specific weight of sand (2650 kg/m³).

Illustration

To discuss the applicability of the established equations, the BSF of Lea 57 with a diameter of 30 cm a height of 90 cm is used. The bed volume of this filter is 63.6 L. Accordingly, if the reactive layer must occupy at most 1:10 of the bed, its volume should not be larger than 6.4 L.

For comparison, at τ = 0.52, 1 kg of Fe⁰ is mixed to 0.16 kg of sand and the mixture occupies a volume (V_rz) of 0.2 L. When mixed to 8.32 kg of sand, the same mass of Fe⁰ (1 kg) corresponds to a τ value of 0.02 (about 6% w/w) and the mixture occupies a volume V_rz = 5.01 L.

Given that water treatment by Fe⁰ is a deep bed filtration, the present example illustrates the necessity of using lower values of τ with the additional advantage of reducing the clogging probability. The discussion in the next section will be mostly based on the threshold τ value of 0.52.

Discussion

This work attempts to sustain research on Fe⁰-amended BSF by optimizing filter design in the perspective to rationalize experimental conditions and enable/ease results comparison. Accordingly, the most important issue regards sizing a filtration bed. Both for laboratory and field works, one of the first tasks is to decide which column(s) to use. In some cases, available columns are simply used. But even in these cases, they should be properly filled to achieve reliable results. Sizing a filter bed will be discussed here in two different perspectives: (i) selecting the appropriate bed size, and (ii) selecting the appropriate thickness of the reactive layer (H_rz).

Rationale selection of the column diameter

Equation (7) gives the thickness of the RZ H_rz as a function of m_Fe, D, τ and C. From these four parameters, C can be considered a constant. Accordingly, H_rz is a function of the used mass of Fe⁰ (m_Fe), the volumetric proportion of Fe⁰ in the solid phase (τ) and the internal diameter of the cylinder (D). Thus, Eq. (7) can allow the calculation of the diameter of the column to be used to achieve a certain H_rz at a given τ value.

The results of such calculations are summarized in Tab. 1 for τ = 0.52, C = 0.64 and the following values of m_Fe: 1.0, 2.5, 5.0, 10.0 and 25.0 kg. From Tab. 1, it is obvious that the sole mass of Fe⁰ that is applicable for lab experiments (D = 3 cm) is 1 kg. The corresponding H_rz value is 55 cm which represents 61% of the filter height (90 cm). Rigorously, the resulting system is not a BSF but rather an SONO-like filter. Nevertheless, the following sequence could be tested: 20 cm sand bed, 55 cm RZ (x = 55 cm) and 15 cm sand bed (y = 15 cm). Whether this design is satisfactory or not should be tested in laboratory investigation. In the case y = 15 cm sand bed is not sufficient to capture dissolved Fe from the RZ, the possibility of adding a second sand bed should be tested (Fig. 1). On the other hand, for H_rz > 0.33H the perspective of sandwiching an Fe⁰/filter between two BSF must be tested (Fig. 2). The first BSF removes dissolved O₂ among others and the second BSF removes dissolved Fe.

Table 1. Variation of the height of the Fe⁰/sand reactive layer (H_rz) as a function of the internal diameter (3.0 ≤ D (cm) ≤ 100) of used columns for a τ value of 0.52. The corresponding masses of sand and Fe⁰ are given. Given that τ = 0.52 is the threshold value for sustainable Fe⁰ beds, these calculations show clearly that for laboratory studies (D ≤ 12 cm), a maximum of 2.5 kg sand should be used to keep the reactive layer lesser than 10 cm. On the other hand, if a field column has an internal diameter of 50 cm 25 kg Fe⁰ will build a 5 cm reactive layer.

D (cm)	Fe^0 mass
	1.0 kg	2.5 kg	5.0 kg	10.0 kg	25.0 kg
3	55.0	136.2	272.4	544.8	1362.0
6	13.6	34.1	68.1	136.2	340.5
12	3.4	8.5	17.0	34.1	85.1
30	0.54	1.36	2.72	5.45	13.62
50	0.20	0.50	1.00	2.00	5.00
75	0.09	0.22	0.44	0.87	2.18
100	0.05	0.12	0.24	0.49	1.21

Figure 1. Schematic diagram of an Fe⁰-amended BSF. The first column is a conventional BSF. The thickness of the RZ (x value) and the aptitude of the under-laying sand bed (y value) to free water from the RZ from iron will be tested for each material. For larger x values or very reactive Fe⁰, a third column could be essential.

Figure 2. Schematic diagram of a three compartments Fe⁰-amended BSF. The first and the third columns are conventional BSF. The thickness of column 2 (RZ) depends on the intrinsic reactivity of used Fe⁰ and can be optimized using Eqs. (7) and (9). The dimension of column 3 (second BSF) can also be optimized using Eqs. (8) and (10).

Figure 3 summarizes the results of the variation of H_rz as a function of D (τ = 0.52). It is clearly shown that the thickness of the reactive layer increases with decreasing diameter. It is shown (Fig. 3a) that H_rz values of up to >1300 cm (13 m) are obtained for larger m_Fe values. However, relevant values must fulfil the condition H_rz ≤ H (H = 90 cm). Figure 3b is limited to H_rz = 0.67H (60 cm) for a better visualization. Table 1 shows that D = 50 cm is the largest value which may enable a 5 cm H_rz while using 25 kg Fe⁰. This situation is likely to occur in a field small-community-scale water plant. However, because Fe⁰ beds are deep bed filtration systems, it is advantageous to use a lower τ value (τ ≤ 0.52) to achieve a thicker H_rz with the same Fe⁰ mass.

Figure 3. Variation of the height of the reactive layer (H_rz) as function of the diameter (D ≤ 100 cm) of used filter: (a) all values, and (b) H_rz ≤ 60 cm. The lines are not fitting functions; they simply connect points to facilitate visualization.

Rationale for the thickness of the reactive zone

The thickness of the RZ is necessarily correlated to the intrinsic reactivity of used Fe⁰. It is intuitive to assume that for each Fe⁰ material, a range of τ values (τ ≤ 0.52) may exist for which filter operation is optimal. A survey of the experimental conditions of available works suggests that they are highly qualitative. For example, Tellen et al. 45 introduced a 5 cm RZ containing 0.5 kg Fe⁰ in 44 cm sand bed without specifying the thickness of the sand layer under-laying the RZ.

The design of Tellen et al. 45 was slightly modified by Pachocka 33. She introduced a 3.2 cm RZ containing an Fe⁰ volumetric proportion of 15% (τ ≤ 0.15). The Fe⁰ mass is not specified but calculations using Eq. (7) while considering the geometry and the dimensions of her device showed that about 2.2 kg Fe⁰ was used. Discounting the fact that building a homogenous 3.2 cm Fe⁰/sand layer is difficult, it is certain that such a thin layer is not suitable for a deep bed filtration process. Accordingly, guidelines are urgently needed to assist the laborious work of amending conventional BSF with reactive Fe⁰.

The suitability of (τ, H_rz) values for several Fe⁰ materials and material amounts should be tested in laboratory studies before a trend is identified for generalization and scaling up to household and small-community Fe⁰-amended BFS.

For further illustration, calculations are made using Eq. (7) for D = 30 cm and various Fe⁰ masses. The results are summarized in Fig. 4 and Tab. 2. Values in Tab. 2 are obtained for τ = 0.52. Figure 4 shows clearly, that H_rz decreases with increasing τ values. For example, 25 kg of Fe⁰ occupy a 13.6 cm at τ = 0.52, 26.2 cm at τ = 0.27 and 101.2 cm at τ = 0.07. While limiting the scale to H_rz = 0.67H (H_rz = 60 cm), it is shown that 5.0, 10.0 and 25.0 kg are not suitable for τ < 7.0% (or τ < 0.07). These examples show clearly that there is an infinite number of possible Fe⁰-amended BSF. A systematic approach is therefore essential to identify and characterize useful combinations. The next section will discuss the design of a BSF containing 3 kg of a reactive Fe⁰.

Figure 4. Variation of the height of the reactive layer (H_rz) as function of the initial Fe⁰ volumetric proportion for different used masses of Fe⁰. The lines are not fitting functions; they simply connect points to facilitate visualization.

Table 2. Variation of the height of the Fe⁰/sand reactive layer (H_rz) as a function of the mass of Fe⁰ for a τ value of 0.52. The corresponding mass of sand is given. Calculations are made for a cylindrical bed with a internal diameter of 30 cm. Given that τ = 0.52 is the threshold value for sustainable Fe⁰ beds, such a bed should not contain less than 10 kg of Fe⁰ to fulfil the condition H_rz ≤ 5 cm. The corresponding V_rz value is 3.85 L.

Fe^0	(kg)	0.5	1.0	2.5	5.0	7.5	10.0	20.0
Sand	(kg)	0.157	0.314	0.785	1.568	2.352	3.136	6.272
Hrz	(cm)	0.27	0.54	1.36	2.72	4.09	5.45	10.90
Vrz	(L)	0.19	0.39	0.96	1.93	2.89	3.85	7.71

Constructing a filter with 3 kg Fe⁰

A conventional concrete intermittent BSF is used. An approach is suggested to purposefully amend it with an RZ containing 3 kg Fe⁰. First, Eq. (7) is modified to account for the filter geometry. The cross-section (S) is a square with a length (L) of 30 cm (0.3 m). The volume of the filter is V = S H = L²H. The modified Eq. (7) reads as (Eq. (9)):

• ((9))

The results (Tab. 3) show that H_rz takes values from 1.28 to 33.39 cm when the volumetric proportion of Fe varies from 0.52 to 0.02%. The corresponding admixed mass of sand varies between 0.94 and 49.94 kg. Taking 50 cm as the maximal thickness of fine sand (sand bed) the resulting beds represent 2.6–66.8% of the sand bed volume.

Table 3. Data for the amendment of a conventional BSF with a reactive layer containing 3 kg Fe⁰ mixed with several masses of sand (m_sand) to yield various proportion of Fe⁰ (P_Fe). H_rz is the resulting thickness of the reactive layer, V_rz its volume and P_rz its volumetric ratio relative to a 50 cm reactive layer.

τ (−)	Hrz (cm)	Vrz (L)	msand (kg)	PFe (%)		Prz (%)
				(w/w)	(v/v)
0.52	1.28	1.16	0.94	76.13	52.0	2.6
0.47	1.42	1.28	1.15	72.30	47.0	2.8
0.42	1.59	1.43	1.41	68.07	42.0	3.2
0.37	1.80	1.62	1.74	63.35	37.0	3.6
0.32	2.09	1.88	2.17	58.07	32.0	4.2
0.27	2.47	2.23	2.76	52.12	27.0	4.9
0.22	3.04	2.73	3.61	45.36	22.0	6.1
0.17	3.93	3.54	4.98	37.61	17.0	7.9
0.12	5.56	5.01	7.47	28.64	12.0	11.1
0.07	9.54	8.59	13.54	18.14	7.0	19.1
0.02	33.39	30.05	49.94	5.67	2.0	66.8

Considering the practical constrain that H_rz ≤ 5 cm 98, only τ values <0.25 are applicable. That is, 3.0 kg Fe⁰ should be mixed to at least 6.46 kg sand to obtain an Fe⁰-amended BSF. The efficiency of selected possible filters (τ ≤ 0.25) should be tested. It is important to notice in this regard that the first permeable reactive barrier (demonstration pilot scale in Borden, Ontario, Canada) contained only 8% volumetric ratio of Fe⁰ (τ = 0.08) and has been properly working for more than 5 years 81, 99. Accordingly, smaller τ values should be tested in parallel experiments. In this effort the suitability of the thickness of the under-laying sand bed should be tested (y value Section 4.1, Fig. 1).

For P_rz > 10% (τ ≤ 0.12 – Tab. 3) the alternative of a three-column-system should be used (Fig. 2). In the perspective of an Fe⁰/sand column in sandwich between two conventional BSF (Section 4.3.1) the possibility of using thin columns to save Fe⁰ must be tested. For example, while using 1 kg of Fe⁰, halving the internal diameter from 9.0 to 4.5 cm increases the H_rz value from 6.1 to 24.2 (τ = 0.52, Fig. 3). This is a four times thicker layer for deep bed filtration in comparison to only 5.6 cm H_rz in a conventional BSF (L = 30 cm, Tab. 3). In other words, using thinner columns enhanced efficiency and save Fe⁰.

The mathematical relation for the variation of H_rz values at constant volume (V_rz1 = V_rz2) is easy to establish and is given as follows:

• ((10))

For D₂ = D₁/2, H_rz2 = 4H_rz1. This corresponds to the results obtained while working with masses.

Ways to efficient Fe⁰-amended BSF

The presented concept of Fe⁰-amended BSF is based on the profound understanding of the complex chemical and physical processes involved in the 20-years-old Fe⁰ remediation technology 63–69, 81, 99–104. To mimic the subsurface Fe⁰ bed as closely as possible, anoxic conditions must be created before the Fe⁰/sand layer (RZ). This condition is satisfied the best by a conventional BSF. On the other hand, to avoid dissolved Fe in the effluent, water from the RZ must migrate through a thick fine sand bed. Ideally this is another conventional BSF (Fig. 2). Accordingly, the simplest way to efficient Fe⁰-amended BSF seems to go through an experimental design with an RZ (a small column, Eq. (10)) sandwiched between two conventional BSF. After such conclusive principle experiments, the next advantageous step could be to test three compartments in the same column (Fig. 1). In this manner, the important requirement of compact systems for less skilled populations (including illiterates) is properly addressed 19. Progressively, it is conceivable to miniaturize the system down to Brita-type filters for a limited volume of tap water. However the focus of this communication is on household and small-community Fe⁰-amended BSF.

Provided that a relevant reactive Fe⁰ is characterized, selected, and used, the effectiveness of Fe⁰-amended BSF does not need to be demonstrated, except some technology verification in the field (monitoring). Beside proper system design, the two sole tasks are: (i) selecting and processing the appropriate Fe⁰ materials (including composites), and (ii) avoiding the use of Fe⁰-amended BSF for waters of pH < 4.5 12. In fact, the Fe⁰ remediation technology is based on the anodic dissolution of iron in neutral and close to neutral aqueous systems. In this pH range, primary iron dissolution is followed by a continuous build up and transformation of a corrosion product layer in the vicinity of Fe⁰ 105–107.

There is strong evidence that the Fe⁰-amended BSF will be efficient. For example, Westerhoff and James 108 performed field continuous-flow experiments at τ = 0.25 (P_Fe = 50% w/w) for almost one year for efficient removal of nitrate. During this period, up to 1500 bed volumes of water were treated. Assuming that two bed volumes of a BSF correspond to the daily need of a rural family, the 1500 bed volumes corresponds to the water demand for 2 years. However, the τ value (0.25) of Westerhoff and James 108 was not the result of any systematic preliminary work. Considering that a system at τ = 0.08 was efficient for more than 5 years under anoxic conditions 81, 99, it is likely that the system of Westerhoff and James 108 was not optimal with regard to long-term permeability.

Concluding remarks

Save drinking water provision using Fe⁰-amended BSF is already proven an efficient technology 33, 37, 41, 45, 46, 80. The estimated huge number of people (884 million) still living without access to improved drinking water 109 is an urgent appeal to the scientific community for (i) a rapid improvement, and (ii) a generalized implementation of this efficient technology. This technology primarily has a single serious limitation: available water must have a pH ≥ 4.5. Fortunately, natural water with pH ≤ 4.5 is unusual. Therefore, Fe⁰-amended BSF should be regarded as a universal technology that can be used at household and small community levels, but in general for decentralized safe drinking water provision.

Using a mathematical modelling, the present work has presented several tools to optimize the efficiency of conventional BSF by adding an RZ containing a layer of Fe⁰ admixed to a nonexpansive material (e.g. MnO₂, pumice, sand) 47, 51, 54. A systematic procedure is introduced to sustain further research on Fe⁰-amended BSF. In particular the RZ must have a minimal thickness of 5 cm (H_rz ≤ 5 cm), and the volumetric proportion of Fe⁰ in the RZ must not be larger than 52% (τ ≤ 0.52). As concerning the sequential design of the Fe⁰-amended BSF, it is essential that dissolved O₂ is quantitatively removed from water before it reaches the RZ. On the other hand, a find sand layer must be placed downwards from the RZ and its thickness should be sufficient for scavenging dissolved Fe from the RZ. Like for other filters 4, 40 layers of other treatment materials (e.g. activated carbon, wooden charcoal, Zeolite) 110, 111 can be purposefully added before and/or after the RZ to optimize the treatment operation. It is essential to notice in this regard that efficient and low-cost traditional methods for aqueous iron removal in packed columns have been described in the literature 111–115.

Further research at several fronts is needed to develop approaches for the proper design of the Fe⁰-amended BSF. Relevant research fields include the intrinsic reactivity of Fe⁰, the size, surface state and geometry of Fe⁰ and nonreactive particles, the size of the filter or the column, the configuration the Fe⁰/admixture layer, the thickness of individual layers, the nature of the admixing agent, the relative particle size of materials in the reactive layer (uniformity), the size of the resulting devices and the water flow velocity. Further mathematical and numerical modelling should be applied to design and validate experimental results. Pilot scale installations are needed to validate the practicality of experimental and modelling results.

Acknowledgements

Thoughtful comments provided by Emmanuel Chimi (University of Douala, Cameroon) on the draft manuscript are gratefully acknowledged. Thanks are due to Sabine Caré (Université Paris-Est/France) for fruitful collaboration on the evolution of the porosity of iron filtration systems. The manuscript was improved by the insightful comments of anonymous reviewers from CLEAN – Soil, Air, Water.

The authors have declared no conflict of interest.