Assessment of low-cost, non-electrically powered chlorination devices for gravity-driven membrane water kiosks in eastern Uganda

: Recontarnination during transport and storage is a common challenge of water supply in low-income settings, especially if water is collected manually. Chlorination is a strategy to reduce recontamination. We assessed seven law-cost, non-electrically powered chlorination devices in gravity-driven membrane filtration (GDM) kiosks in eastern Uganda: one floater, two in-line dosers, three end-line dosers (tap-attached), and one manual dispenser. The evaluation criteria were dosing consistency, user-friendliness, ease of maintenance, local supply chain, and cost. Achieving an adequate chlorine dosage (~2 mg/Lat the tap and:::: 0.2 mg/L after 24 h of storage in a container) was challenging. The T-chlorinator was the most promising option for GDM kiosks: it achieved correct dosage (CD, 1.5- 2.5 mg/L) with a probability of 90 per cent, was easy to use and maintain, economical, and can be made from locally available materials. The other in-line option, the chlorine-dosing bucket (40 per cent CD) still needs design. improvements. The end-line options AkvoTur (67 per cent CD) and AquatabsFlo® (57 per cent CD) are easy to install and operate at the tap, but can be easily damaged in the GDM set-up. The Venturi doser (52 per cent CD) did not perform satisfactorily ,.vi.th flow rates> 6 L/min. The chlorine dispenser (52 per cent CD) was robust and user-friendly, but can only be recommended if users comply ,.vi.th chlorinating the water themselves. Establishing a sustainable supply chain for chlorine products was challenging. Where solid chl01ine tablets were locally rarely available, the costs of liquid chlorine options were high (27-162 per cent of the water price).

SUSTAINABLE DEVELOPMENT GOAL (SDG) 6 calls for universal and equitable access to safe and affordable drinking water fo r all by 2030. Water kiosks contribute to SDG 6 by providing safe water at the community level in rural areas. Water is considered as safe if it meets the microbial guideline of the World Health Organization (WHO), which requires that E. coli must not be detected in any 100 ml sample (WHO, 2017). Even though water might be safe at a communal water source, microbial water quality often deteriorates during transport and storage after collection (Wright et al ., 2004;Harris et al., 2013;Opryszko et al., 2013;Meierhofer et al., 2019). Typically, 20 L jerry cans are used to transport and store drinking water in rural areas. Biofilms that grow inside these plastic containers and poor water, sanitation and hygiene (WASH) conditions can lead to recontamination Oagals et al., 2003;Mellor et al., 2013;Opryszko et al., 2013). As a result, the water no longer meets the WHO microbial guideline at the point of consumption.
Chlorination is a water treatment strategy that provides residual disinfection, reducing recontamination risks during transport and storage. WHO recommends a concentration of0.2-0.5 mg/L free residual chlorine (PRC) at the point of consumption (WHO, 2017). Consistent provision and consumption of treated water is important to maintain good health (Hunter, 2009;Brown and Clasen, 2012;Enger et al., 2013). However, users' compliance with chlorinating water at the household level often is inadequate and establishing the necessary level of compliance has been found to be difficult (McLaughlin et al., 2009;Levy et al., 2014). The installation of a chlorinator at the point of collection would circum vent the need to establish user compliance and can increase the proportion of chlorinated water available for consumption. Therefore, the objective of our study was to assess low-cost, non-electrically powered chlorinators at the point of collection.

Gravity-driven membrane filtration water kiosks
Together with local partners, five gravity-driven membrane (GDM) kiosks were constructed in low-income areas in eastern Uganda, with ultrafiltration membranes driven by gravity filter surface water pumped up with solar pumps from Lake Victoria (total organic carbon: 8 mg/L, pH: 6, turbidity: usually 10-25 NTU, peaks up to 150 NTU ) . After filtration, the water at the tap meets the WHO microbial guideline value, but recontamination after transport and storage at the household level has been observed . A scheme of the GDM kiosk is presented in Figure   Adapted from Orner et al. (2017) Copyright practicalaclionpublishing. com

Description
The floater floats in the clean w ater storage tank. Slowly dissolving 90% trichlorois ocyanuric acid (TCCA) tablets (0 = 3 inches) are placed into the device and dissolve in the w ater. Opening or closing the sl its at the bottom of the floater, as w ell as changing the number of tablets and the number of floaters, allow for adjusting the dosage.
The CDB is an air-tig ht closed bucket (30 L) with a bypass pipe that was installed in-line between the clean water tan k and t he kiosk's tap. A floater (see above), containing TCCA tablets (0 = 3"), is placed inside the bucket. Valves at the inflow of the bucket and the bypass reg ulate the proportion of w ater passing through the CDB versus the bypass pipe and thereby regulate dosing. The higher the proportion of water passing through the bucket, the higher the chlorine concentration. To improve mixing behaviour, the inlet to the bucket is at the bottom and the outlet at the water surface. This is attained by an elbow with a~ 10 cm pipe facing upw ards attached to the outlet.
The T-ch lorinator consists of a cylinder with small holes that is placed into a T-fitting. It w as installed in-line between the clean w ater tank and the kiosk's tap. The cyl inder contains TCCA tablets (0 = 1 "), w hich are eroded by the flow of w ater. To adjust the dosage the number of tablets can be altered. Alternatively, different cylinders wi th different amounts and sizes of holes can be manufactured and can easily be exchanged.

Description
The Venturi doser uses the Venturi effect for chlorine dosage. Water passes from a pipe wi th a la rge diameter to a pipe w ith a smal ler diameter. Th is leads to an increase in flow velocity, whil e the stat ic fluid pressure decreases. The pressure difference at the constriction causes liquid chlorin e to be ' sucked in'. The Venturi doser needs to be operated endof-pipe and, therefore, wa s installed rig ht before t he tap. A 1.2% NaOCI solution was used to refill the device. A needle valve that res t ricts the flow regu lates th e dos ing.
Water flowing through the Aquatabsflo® dissolves sodium dichloroi socyanurate (NaDCC) tabl ets placed inside the cartridge. The devi ce w as installed after the tap with a bayonet catch for easy removal. Tablets are not so ld separately; th us, the entire device is replaced if the ch lorine is used up. The concentration can be adjusted w ith a screw, restricting the outflow, and thereby increasing the contact time w ith the chlorine tablets.
Sim il ar to the Aquatabs Fl o®, th e device is installed after the tap. At the bottom of a container ( ca. 1 L, not airtight), a PVC plate is fixed, on which a PVC pipe can be placed. The pipe has two 2-mm slits. Water enters through one sli t, erodes the TCCA tabl ets (0 = 1 ") inside and leaves through the o pposite slit. To adjust the dosage, the PVC pipe can be turned. The more the slits face in the direction of the f low, the higher the concentration.
The chlorine dispenser is a con tainer with a ball valve. It w as the only ' non-pass ive' option evaluated and w as located next to th e w ater kiosk bu ilding. The case p rotects the chlorine from heat and, therefore, slows down decay. Customers place their 20 L jerry can fil led with water below the dispenser and tu rn the ball valve to add chlorine to the jerry can. One turn re leases 3 ml of a 1.2% NaOCI.
A continuous and constant flow rate is optimal for water-powered and diffusion chlorinators (Skinner, 2001). GDM systems are therefore challenging to these kinds of chlorinators because the flow rate is variable (3-24 L/min) and the pressure is low (2-30 kPa). The flow rate is determined by the water level in the storage tank and the height of the tank, and the pressure by the height difference between the level in the storage tank and the tap.

Adequate dosing
Adequate dosing requires balancing several criteria. Branz (2017) advises to 1) meet the chlorine demand of the water; 2) maintain sufficient free residual chlorine (PRC) concentrations during transport and storage; and 3) avoid exceeding international maximum guideline values and user taste and odour objections. In t he study area, the average chlorine demand was l .S mg/ L over 24 h of storage in an uncleaned jerry can (n = 46, st andard deviation (SD) = 0.3 mg/L). Wilhelm et al. (2018) and international guidelines (WHO, 2017; CDC, 2020) recommend ;::a: 0.2 mg/L PRC after 24 h of transport and storage. WHO (2017) recommends a maximum of 5 mg/L FRC. Taste acceptability limits vary. In Bangladesh, t he acceptability limit was found to be 'well below 2 mg/L' (Crider et al., 2018), while in Zambia, water with 2 mg/ L FRC was rated 'too strong and bitter', and in Ethiopia, 2 mg/L was 'noticeable' (Lantagne,200S). In contrast, water tastings with households (n = 197, n = sample number) in the study area revealed a high acceptance of water chlorinated with 2 mg/ L FRC.
Balandng the three criteria, we targeted a FRC concentration of 2 mg/L at the tap. This is slightly higher than Wilhelm et al. (2018) recommend (1.SS mg/L) for improved or low turbidity sources for 24 h protection. The higher target concentration is due to the relatively high chlorine demand measured in the study area. Reasons fo r the high chlorine demand include biofilms attached to the storage container's walls and deposits Oagals et al., 2003;Mellor et al., 2013;Opryszko et al ., 2013), as well as the absence of a lid on most storage containers (Lantagne, 200S;Ali et al., 2015). Additionally, the high content of organic compounds (total organic carbon = 8 mg/L) in the lake water could increase chlorine demand, similarly as observed in piped networks (Lu et al., 1999;Yee et al., 2006).

Different chlorine products
Solid and liquid chlorine products were used. The solid tablet form of chlorine is easier to handle and store as opposed to liquid chlorine, which has a shorter shelf life (Clasen and Edmondson, 2006). Solid chlorine was used in the form of slowly dissolving 90 per cent TCCA tablets and NaDCC tablets. Three-inch TCCA tablets (Henkel Polymer Ltd, US$0.0l /m 3 for 2 mg/ L; exchange rate: US$ l = UGX 3,800) were available in Kampala, Uganda, at the time of the study. One-inch TCCA tablets were imported from Switzerland for the study. After the study, our local partner was able to purchase them in airobi (S0.02/m 3 for 2 mg/L). aDCC (Medentech, $0.22/ m 3 fo r 2 mg/L) tablets are available in Uganda (direct communication by Medentech). Liquid chlorine was purchased at the local market as 1.2 per cent sodium hypochlorite solution (NaOCl, $0.S2-0.96/m 3 for 2 mg/ L) or self-produced as NaOCl solution (0.5-0.6 per cent), using a Mini-WATN M electrochlorinator (Antenna Foundation, $2S0). For liquid chlorine, the WATA system is economically attractive (~S0.15-0.59/m 3 for 2 mg/L, including salt cost and device cost; lifespan of device: 1-5 years; water production: 600-1,200 m 3 /year) and circumvents the need for a chlorine supply chain. evertheless, the local kiosk operators expressed difficulties in handling and operating the device.

Evaluation criteria
For each criterion, we developed a rating system and 'good', 'moderate', and 'poor 1 were the possible ratings.
Dosing consistency. Dosing consistency was defined as the probability of the device achieving a 'correct dosage' (CD) of 1.5-2.5 mg/L PRC at the kiosk's tap, balancing the minimum requirements of;::: 0.2 mg/L after 24 h of storage and taste acceptability level (~2 mg/ L). The following ratings were applied: Good ;::: 90 per cent, moderate;::: 50 per cent, and poor< 50 per cent.
User-friendliness. If the operators reported that the device was easy to handle and did not express any complaints regarding its basic use (refilling chlorine products or adjusting chlorine dosage), the device received a good rating. If they reported difficulties or expressed a complaint regarding basic use (e.g. not easily accessible or the burden was beyond basic use), the rating attributed was moderate. If they were unable to use the device, the rating attributed was poor.
Ease of maintenance. If the device did not need to be repaired during the tests in the field, the rating attributed was good. If minor repairs had to be done that could be solved by the operator, the rating was moderate. If repairs could not be done by the operator or if any of the devices ceased to function, the rating attributed was poor.
Local supply chain. The following rating was applied regarding the local availability of the device and its spare parts: availability on the local market (LM) of Busia in eastern Uganda or in the neighbouring local market across the Kenyan border (rating = good), in the countrys capital Kampala (= moderate) or it needs to be imported (= poor). Local availability was evaluated separately for chlorine products and hardware, including spare parts.
Cost. Calculations include the cost of the chlorine to achieve 2 mg/ L PRC and the cost of the device, but no labour costs. We considered different device lifespans (1-5 years) and capacities (600--1,200 m 3 /year). The current water price is UGX SO/jerry can ($0.013 = $0.66/m 3 ) . Assuming 900 m 3 sold per year, a revenue of $592 would be generated. After deducting salary costs and maintenance expenses, 16 per cent of the yearly income ($93) would remain as profit. Under these assumptions, the cost of chlorinating the water cannot exceed $0.l/m 3 (rating = good). A rise in the water price to UGX 100/ jerry can (= $1.32), as is common in other parts of Uganda, would enable a price of $0.59/m 3 for chlorination (= moderate). For higher chlorination costs, the rating attributed was poor.

Data collection and analysis
The chlorinators were consecutively installed as shown in Figure 1 and dosing was adjusted to ~2 mg/ L. Four of the seven devices were extensively tested (COB, T-chlorinator, Venturi, and AkvoTur). PRC concentrations were measured at least once per week over 1. 5-2 months. On every data collection day, a sample was taken every 5-15 min over 2-6 hours to capture the fluctuation of flow rates.
The remaining three devices were tested less extensively. For the float er and the chlorine dispenser, sam ples were taken over 3-5 days during one week, and for the AquatabsPlo®, sam ples were taken over 0.5 h on two days. Flow rates were not recorded . PRC was measured using a LaMotte DC1500-CL colorimeter and diethylphenylenediamine (DPD) tablets (LaMotte DPD l TesTabs). Samples were t aken at the tap of the GDM kiosks. The tap was opened, flushed for 10 seconds and a 10 ml vial was rin sed three times. Then, the sam ple was taken and im mediately analysed. Samples from the chlorine dispenser were taken fro m the jerry can after the water was chlorinated and the jerry can was shaken.
To gain insight on user-friendliness, ease of maintenance, local supply chain, and costs, we 1) conducted qualitative interviews with the kiosk operators and staff from our local partner organization; 2) searched local markets for devices and consumables; and 3) evaluated observations of the research staff involved in the fieldwork . In addition, the operators were trained by our local partner organization to use the devices. The operators and GO staff were informed about the goal, purpose, and methodology of the study and asked for informed consent. The study protocol was reviewed and approved by the Ethics Committee of fakerere University on 18 July 2018, the Uganda ational Council for Science and Technology, and the Ethical Committ ee of Eawag, the Swiss Federal Institute of Aquatic Science and Technology on 6 June 2018.
Data was analysed using Excel and R-Studio. To test for normal distribution, histograms and one-sample Kolmogorov-Smirnov tests (two-sided) were used. PRC measurements were normally distributed (p > 0.05). The Pearson correlation coefficient (two-tailed) and linear regression models were used to explore the correlation between PRC concentrations and flow rate.

Floater
The floater is commonly used to chlorinate swim ming pools. In a technical manual, Luff (2001) mentioned the use of this device to chlorinate drinking water. However, the authors are not aware of any study that has assessed the floater's performance in chlorinating drinking water.
Our data indicates that the floater did not achieve consistent dosage (x = 1.5 mg/L (x = mean), SD= 0.9 mg/L, min = 0.1 mg/L, max = 3.1 mg/L, CD = 37 per cent, n = 15). During the analysis, the storage tank was constantly less than a quarter full. As the floater is placed in the storage tank, it is difficult to refill the tablets and to adjust the dosing. A positive factor, however, is that the floater is robust, available in Kam pala and can be operated at low cost ( ~$0.02/ m 3 ) . Yet, in the context of the GDM set-up, it is problematic to chlorinate the water from the inside of the storage tank as chlorinated water could flow back from th e storage tank to the membrane tank and damage the membranes. Further studies are needed concerning th e use of this device. Intermittent or variable flow also influences the dosing reliability because the contact time with the water varies.

Chlorine-dosing bucket (COB)
The CDB was designed after observing challenges with the floater in the GDM set-up. The floater was integrated into an in-line device developed at Eawag. No similar use of a floater is known to the authors.
The CDB did not achieve consistent dosage (x =1 .7, SD= 0.9 mg/L, min = 0.3 mg/L, max = 3.5 mg/L, CD = 40 per cent, n = 56, flow rate = 2.7-7.5 L/min). In a simple linear regression model, the flow rate and PRC concentrations were significantly correlated (R = -0.75, p = 0.003). To reduce the flow dependency, the proportion of water passing the bucket could be reduced or the size of the bucket increased. This is predicted by the steady state solution of the mass balance equation (Equation 1) and should be further researched. Some positive aspects are that the CDB is user-friendly, robust, can be made from low-cost locally available materials, and operated at low cost (::;; $0. l /m 3 ) .

T-ch/o rinator
The T-chlorinator assessed in this study is an adaption of the model looked at by Omer et al. (2017). They used the chlorinator in a gravity-fed piped distribution system . Compared to our study, the flow rate was substantially higher (60 L/min) and the target concentration substantially lower (0.27 mg/L).
In our assessment, the T-chlorinator achieved the best dosing consistency of all devices (x = 2.0 mg/ L, SD = 0.3 m g/ L, m in = 1.1 mg/ L, max = 3.0 mg/ L, CD = 90 per cent, n = 64, flow rate = 3-8 L/min). No significant correlation between the flow rate and PRC concentration was found (R = -0.23, p = 0.070), which is consistent with Orner et al. (2017). The device is userfriendly and robust. The m aterials to bu ild the device are inexpensive (~$15) and available in Uganda, except for the required 1-inch chlorine tablets that need to be imported from airobi, Kenya. The operating costs (::;; $0.05 / m 3 ) are relatively low.

Ve nturi doser
The Venturi doser was developed by MSR, PATH, and Stanford University.
In our assessment, the dosing was less consistent for a flow rate of 6-20 L/min (x = 2.1 mg/I, SD = 0.7 mg/L, min = 0.2 mg/L, max = 3.3 mg/L, CD = 52 per cent, n = 58). For flow rates below 6 L/min, which sometimes occur at GDM kiosks, the WHO guideline value of 5 mg/L (WHO, 2017) was exceeded. o significant correlation between the flow rate and PRC (R = -0.3, p = 0.121) was observed. The operators were able to handle the Venturi doser. Even though the device is made of fragile parts, they are all protected in a robust metal box. o repairs were required. The device is still a prototype, is not available locally, and is relatively expensive ($150). Liquid chlorine is available locally (~$0.82/m 3 ) or can be selfproduced (~S0.17-0.59/m 3 ), but at relatively high costs. Pickering et al. (2019) assessed the AquatabsFlo® to chlorinate water at the point of collection and reported an average PRC concentration of 0.37 mg/L (SD= 0.32, target concentration: 0.2-0.5 mg/ L, n = 2,335).

AquatabsF/o®
We found the dosing consistency to be moderate (x = 2.2 mg/L, SD = 0.6 mg/L, min = 1. 7 mg/ L, max = 3.6 mg/L, CD = 57 per cent, n = 8). However, the findings are based on only eight sam ples. Failure of the device to chlorinate (PRC :,; 0.2 mg/L) was observed four times prior to data collection, and two times during data collection. Removing, shaking, and reinstalling the device restored PRC concentrations to ~2 mg/ L. The two failed measu rements were not included in the analysis. Ad justing the dosage with t he provided screw was difficult. Furthermore, the device had to be removed from the kiosk's tap when the kiosk closed at night to prevent vandalism and frequent removal can damage the plastic bayonet catch . In other set-ups, the cartridge can b e permanently installed before the storage tank, where it is better protected (Pickering et al., 2019). The device, including chlorin e tablets, is available in Kampala. When the tablet s are used up, the whole device has to be replaced. This leads to an increase in costs ($0.22/ m 3 for 2 mg/L).

AkvoTur
The AkvoTur is similar to the AquatabsFlo®, but with a better mechanism to adjust dosage and the option to refill chlorine tablets. It was designed at Eawag.
Th e dosing consistency of the Ak-voTur was moderate (x = 2. 1 mg/L, SD = 0.5 mg/L, min = 0.8 mg/L, max = 3.6 mg/L, CD= 67 per cent, n = 78, flow rate = 2.5-9 L/min). T o significant correlation between the flow rate and the PRC concentration was observed (R = 0.09, p = 0.436). The device cannot handle flow rates above 12 L/min, as this would cause overflow. The handling is similar to the AquatabsFlo®, but overdosing is possible if the cylinder containing the tablets is not positioned correctly. Like the AquatabsFlo®, the AkvoTur had to be removed daily from the kiosk's tap to prevent vandalism . The plastic threading suffers from the daily removal; therefore, th e design needs improvements. Materials to produce the Ak-voTur are available locally at low cost (~$7). As for the T-chlorinator, the 1-inch tablets need to be imported from Nairobi, Kenya.

Chlorine dispenser
Over 27,000 chlorine dispensers have been installed in Kenya, Uganda, and Malawi (Ahuja, 2017). Yates et al. (2015) studied dispenser programmes in four emergency situations. The programmes always employed a local promoter for community training. Confirmed dispenser use (PRC :::: 0.2 mg/L in stored household water) ranged from 5 to 87 per cent.  The dispenser releases 3 ml of NaOCl per turn of the ball valve. In our assessment, one turn provided a mean 0. 9 mg/L in a 20 L jerry can (SD = 0.2 mg/L, min = 0.5 mg/ L, max = 1.3 mg/L, n = 27), and two turns 1.9 mg/L (SD = 0.7 mg/L, min 0.6 mg/L, max = 3.9 mg/L, CD = 52 per cent, n = 17). The device is easy to use, robust, and available in Kam pala at -$20. The liquid chlorine is available locally at -$0.96/m 3 or can be self-produced (-$0. 17--0.59/m 3 ) . As users have to chlorinate the water themselves, it is likely that the water would not always be chlorinated before consum ption. This was confirmed by observations of the operator, as well as by Yates et al. (2015). The dependence upon user compliance is a significant disadvantage of the chlorine dispenser. Table 2 summarizes the assessment of the diffe rent chlorination devices.

Limitations
Due to how the data collection in t he field was done, the num ber of samples taken for PRC m easurements of the different devices vary. While the sample sizes for Aqu atabsFlo® (n = 8), the chlorine dispenser (n = 27), and t he floater (n = 15) are limited, they are much larger for the other devices (n = 56-78). As the sam ple size increases, t he sam pling distribu tion clusters more tightly around the mean and, consequent ly, the standard deviations shrink. Therefore, our results might overest imat e the standard deviation of the devices with small sam ple sizes.

Conclusion and recommendations
Seven low-cost, non-electrically powered chlorinators were assessed at GDM water kiosks in eastern Uganda. Dosing consistency was challenging. Only the T-chlorinator h ad a probability ;::c: 90 per cent to achieve the target concentration of 1.5-2.5 mg/L PRC at th e tap. The probabilities of reaching the target concentration of the AkvoTur, the Venturi doser, the AquatabsFlo®, and the chlorine dispenser were 52-67 per cent, and of the floater and the CDB <40 per cent.
All the devices assessed were user-friendly and easy to maintain, except for the floater, which is difficult to refill, and the tap-attached options Ak-voTur and AquatabsFlo®, which need to be removed daily in the GDM set-up to prevent vandalism . Because of the daily removal of these devices, the risk of damage during handling is increased. In other set-ups, these devices can be installed before the clean water tank, and are, therefore, better protected (Pickering et al., 2019). The Venturi doser is also attached to the tap, but it is robust and can be permanently installed. The findings of the Venturi doser, however, indicate inconsistent dosing when flow rates drop below 6 L/min.
All the devices or parts needed to construct them, as well as chlorine products, are available in Uganda or the border town in Kenya that is 1 km away, except for the Venturi doser and the I -inch TCCA ch lorine tablets. In the calculated scenarios (600-1,200 m 3 chlorinated water/year over 1-5 years), the chlorination costs of the tablet-based options were :,; $0.1 /m 3 and could be covered by the revenues of the kiosk. The only exception was t he AquatabsPlo® ($0.22/m 3 ). The cost s of the liquid chlorine-based options were between $0.18 and $ 1.07 /m 3 and could not be covered by the revenues of the kiosk given the current water price ($0.66/m 3 ) . Adding the cost for liquid chlorination to the water price, therefore, would have a significant impact on the business model.
Overall, the T-chlorinator perfo rmed best in our assessment for the GDM kiosks, which are characterized by low pressure (2-30 kPa), a variable and low flow rate (3-24 L/min), and no possibility to install the t ap-att ached options in a safe place. However, t he required I-inch TCCA tablets need to be imported from airobi, Kenya . The chlorine dispenser is the only option where users have to chlorinate the water manually themselves. Establishing a consistent practice to do so on the part of users has been found to be difficult (Yates et al., 2015), but is critical fo r health improvements (Hunter, 2009;Brown and Clasen, 2012;Enger et al., 2013). Therefore, the chlorine dispenser can only be recommended if user com pliance were high.
Even though the GDM kiosks in this study have a specific design, general recommendations can be drawn from our experience.
1. If the chlorine demand of the water is high (> 1.3 mg/Lover 24 h), it may be difficult to find a chlorinator that consistently (:e: 90 per cent) guarantees :e: 0.2 mg/L PRC after 24 h of storage and at the same time does not interfere with taste acceptability (~2 mg/L). In this case, measures to reduce the chlorine demand are recommended; for example, safe storage containers (Roberts et al., 2001;Reed et al., 2011;Mellor et al., 2013;Gartner et al., 2020) or periodically cleaning and disinfecting the containers (Steele et al., 2008;Meierhofer et al., 2019;Gartner et al., 2020). 2. The installation of in-or end-line chlorination options offers a potential solution to circumvent the challenge of establishing consistent and correct manual chlorine dosing by the user. However, low-pressure systems and variable flow rates pose challenges to the dosing consistency of many in-and end-line chlorinators. Further developing these kinds of devices to im prove their performance is required. 3. The establishment of a reliable and affordable supply chain for the chlorine products suitable for drinking water treatment may be challenging. Yet, it is a key factor for the sustainable operation of chlorination devices.