). Associations between climate variables and water quality in low-and middle-income countries: A scoping review

Understanding how climate change will affect water quality and therefore, health, is critical for building resilient water services in low-and middle-income countries (LMICs) where the effect of climate change will be felt most acutely. Evidence of the effect of climate variables such as temperate and rainfall on water quality can generate insights into the likely impact of future climate change. While the seasonal effects on water quality are known, and there is strong qualitative evidence that climate change will impact water quality, there are no reviews that synthesise quantitative evidence from LMICs on links between climate variables and water quality. We mapped the available evidence on a range of climate exposures and water quality outcomes and identified 98 peer-reviewed studies. This included observational studies on the impact of temperature and rainfall events (which may cause short-term changes in contaminant concentrations), and modelling studies on the long-term impacts of sea level rise. Evidence on links between antecedent rainfall and microbiological contamination of water supplies is strong and relatively evenly distributed geographically, but largely focused on faecal indicator bacteria and on untreated shallow groundwater sources of drinking water. The literature on climate effects on geogenic contaminants was sparse. There is substantial research on the links between water temperature and cyanobacteria blooms in surface waters, although most studies were from two countries and did not examine potential effects on water treatment. Similarly, studies modelling the impact of sea level rise on groundwater salinity, mostly from south-Asia and the Middle East, did not discuss challenges for drinking water supplies. We identified key future research priorities based on this review. These include: more studies on specific pathogens (including opportunistic pathogens) in water supplies and their relationships with climate variables; more studies that assess likely relationships between climate variables and water treatment processes; studies into the relationships between climate variables and geogenic contaminants, including risks from heavy metals released as glacier retreat; and, research into the impacts of wildfires on water quality in LMICs given the current dearth of studies but recognised importance.


Introduction
Climate change poses serious threats to drinking water and health (McMichael et al. 2006;Howard et al. 2010).Long-term changes in temperature and rainfall patterns, and sea level rise represent shifts in stresses that can affect water resource availability and quality (Jiménez-Cisneros et al., 2014;IPCC, 2021).Floods, droughts, and windstorms pose short-term threats in the form of infrastructure damage and loss of services (Howard et al. 2016).Globally, outbreaks of infectious diseases following an extreme weather event have been commonly attributed to contaminated drinking water supplies and other deficits in water supply and sanitation (Alderman et al. 2012;Cann et al. 2013).
The World Health Organization (WHO) estimates an additional 48,000 deaths from diarrhoea will be caused by climate effects between 2030 and 2050 (WHO, 2014), with 60% of diarrhoea deaths in low-and middle-income countries currently attributed to inadequate water, sanitation and hygiene (WASH) (Prüss-Ustün et al. 2019).Understanding how slow-and rapid-onset climate events, coupled with land use change and population growth, affect drinking water is critical towards building resilience in the water sector and mitigating health impacts of climate change (Howard and Bartram, 2010;WHO, 2017a).
Previous reviews have synthesised qualitative evidence on the potential links between climate change and water quality (Delpla et al. 2009;Howard et al. 2016;Hunter, 2003;Khan et al. 2015;Levy et al. 2016;Paerl et al. 2020;Whitehead et al. 2009).Increased runoff and stormwater overflow from heavy rainfall over saturated catchments can transport pollutants to surface waters, and cause an increase in pathogens, turbidity, nutrients, and organic matter concentrations (Khan et al. 2015;Schijven and de Roda Husman, 2005).Surface waters receiving wastewater discharges may face additional challenges with increased intensity of rainfall (Jalliffier-Verne et al. 2017), especially in areas (Werner et al., 2012) with inadequate wastewater treatment.
Pollutant concentrations in surface waters may decline due to decreased overland flow during extended dry spells or drought, although the presence of point sources of pollution in the catchment can counter this effect (Mosley, 2015).Decreased reservoir volumes, coupled with warmer temperatures and longer residence times, can increase the risk of cyanobacterial blooms (Paerl et al. 2020).Decreased river flow and groundwater recharge in coastal areas can intensify seawater intrusion into aquifers, already vulnerable because of sea level rise (-Jiménez-Cisneros et al., 2014;Vallejos et al. 2015).
The effect of warmer ambient temperature on pathogen survival in source waters will depend on residence time, depth, and wind conditions, and the temperature sensitivity of the pathogen (Schijven and de Roda Husman, 2005).Some pathogens such as Vibrio cholerae may thrive better in a warmer environment (Vezzulli et al. 2013).Higher temperatures have been correlated with lower concentrations of enteroviruses, Cryptosporidium and Giardia in surface waters in Europe, but flowrate and residence time were better predictors of pathogen concentration, indicating a combined effect of rainfall and temperature (Schijven and de Roda Husman, 2005).
Despite the comprehensive body of literature qualitatively summarising the links between climate change and water quality, there are no reviews on the associations between water quality and quantitative measurements of climate variables focusing on low-and middle-income countries (LMICs).Water supplies in LMICs typically rely on simpler technologies, are not well-resourced and may not have adequate water treatment.With a projected increase in flooding in south and south-east Asia, and drought in large parts of central Africa, increased pressure on water supplies is expected (Jiménez-Cisneros et al., 2014;MacDonald et al. 2009).The increased vulnerability of LMICs to climate change (World Bank, 2010) demands a closer analysis of the potential impacts of climate change on water quality.
The effects of climate change on water quality are likely to be similar to those already observed in response to changes in temperature and rainfall (Jiménez-Cisneros et al., 2014).There is strong quantitative evidence of seasonal changes in water quality in LMIC settings, but studies examining seasonal changes in water quality do not capture the impact of individual rainfall events, which may cause short-term contaminant spikes and skew seasonal averages (Kostyla et al. 2015).
As climate change is expected to lead to increased occurrence of extreme rain events, understanding relationships between individual events and water quality will be essential in understanding the consequences of climate change.Furthermore, the impact of slow-onset events such as sea level rise or drought on water quality in LMICs has not been reviewed.
To address this gap, a scoping review of the available literature from LMICs was undertaken investigating the relationships between observed water quality and climate variables, and modelled changes in water quality based on climate change projections.A scoping review was chosen because we wanted to explore a breadth of climate variables (exposures) and water quality threats (outcomes).Scoping reviews are useful tools to map existing evidence and identify research gaps, and to summarize studies with heterogeneous study design (Armstrong et al. 2011;Arksey and O'Malley, 2005).
The conceptual basis for the scoping review groups climate effects into 3 categories and shows how they cascade to water quality threats (Fig. 1).We recognize that these relationships depend on the interventions around infrastructure and management that can mitigate the consequences of climate change for water quality.

Methods
The scoping review was designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (Moher et al. 2009).

Search strategy and inclusion criteria
Peer-reviewed literature was searched in three databases -PubMed, Scopus, and Web of Science -in May 2020.Search terms included a combination of keywords related to climate variables and climate change and drinking water quality (Supplementary information, Table S1), and were chosen to reflect the potential climate change A. Nijhawan and G. Howard impacts depicted in Fig. 1.No restrictions were placed on date of publication.Studies from high-income countries (based on the World Bank classification), and those without titles, abstracts, or full-text available in English were excluded.
Studies were included in the review if they: investigated the link between water quality and specified values of temperature or rainfall amounts, measured within a specified period of time prior to sampling; or modelled changes in water quality based on long-term climate change projections.Reference lists of included papers were manually scanned and papers that met the inclusion criteria were included in the review.
Studies were excluded if they reported only seasonal variation of water quality without quantitative measurements of temperature or rainfall; if they investigated freshwater not used for drinking; or investigated marine or brackish waters.

Quality of reporting
Studies included in the synthesis were assessed for the quality of reporting using the checklists in Supplementary Information (Table S2).The checklists were developed based on the Collaboration for Environmental Evidence Critical Appraisal Tool Version 0.1 (Konno et al. 2020) and NIH National Heart, Lung and Blood Institute's (Bethesda, Maryland, USA) Quality Assessment Tool for Observational Cohort and Cross-sectional Studies.

Data extraction
Relevant data on study location, type of study, climate variable (rainfall, temperature) or climate event (windstorm, coastal storm surge, sea-level rise), non-climate variables and drinking water quality parameters was extracted for each study that met the inclusion criteria.The findings from each paper were summarised.Given the heterogeneity of the studies and the measures reported, we did not attempt metaanalysis of the data.

Search results, study characteristics and quality of reporting
The database search generated 21,516 results (Fig. 2).After removing duplicates, both authors screened titles and abstracts, and 143 studies were considered eligible for full-text review.The bibliographies of studies were searched during the full-text review and an additional 57 studies were added manually.In all, 98 studies (Table S3) were included in the qualitative synthesis and 102 were excluded as they did not meet our inclusion criteria.
For the purpose of the qualitative synthesis, the papers were categorised according to the water quality threat studied.Nineteen studies examined the links between microbiological quality, specifically evidence of faecal contamination, in source waters and rainfall.Of these, three studies also considered the impact of ambient temperature.Fiftyfour studies reported observed or modelled associations between cyanobacterial in surface water used for drinking-water supply and climate variables.Twenty-one studies examined the impact of climate variables and change on the salinity of coastal aquifers used as drinking water sources.Salinity was expressed as electrical conductivity (EC), total dissolved solids (TDS), or chloride (Cl − ) concentrations.Four studies explored variability in geogenic contaminant (arsenic and fluoride) levels with rainfall.
The percentage of studies that fulfilled each criterion of the quality assessment are depicted in the Supplementary Information (Fig. S1 a and  b); 81% of studies adequately reported methods that would allow for replication of the study, while fewer than 20% of studies reported random sampling.The rest of the studies collected samples at points of interest, such as intakes of water treatment plants, or at existing monitoring stations.Three-quarters of the studies collected samples over a period of at least 12 months or during every season that exists in the study location.We also included a criterion on whether the authors reported statistical measures of association; 25% of the observational studies did not perform statistical tests and reported the relationship between exposure and outcome based on visual interpretation of plotted results.While 81% of studies measured potential covariates (e.g., nonclimate variables that could influence water quality), only 65% of studies included them in the statistical analysis.
Nearly 60% of the modelling studies chose exposure variables (temperature, rainfall or sea level rise) based on historical trends or climate projections from general circulation models (GCM) to predict changes in water quality.Over 70% of studies used measured data to calibrate their model, and a similar number included covariates in the model.

Microbiological quality
The identified studies reported the microbiological water quality in spring sources, well water and piped water supplies, using faecal indicator organisms and in some cases pathogens.The studies were predominantly from sub-Saharan Africa and south-Asia.Despite the heterogeneity in location and source type, a significant (p < 0.05) positive association was reported between antecedent rainfall and faecal indicator organisms and pathogens (presence and concentration) in the majority of the studies, but there was mixed evidence on the association between temperature and contamination.

Antecedent rainfall
Groundwater quality in shallow wells was reported to respond rapidly to individual rainfall events.Significant correlation between thermotolerant coliform (TTC) count and amount of rainfall within hours of sampling was reported for wells in Uganda (Kulabako et al. 2007) whereas in Vanuatu, the detection of E. coli above 10 MPN 1 /100 mL was statistically correlated with 24 h rainfall for wells equipped with a Nira AF-85 handpump, but not for other hand pump types or motorised pumps (Foster et al. 2019).Barrell and Rowland (1979) also reported an increase in faecal coliform (FC; another term for thermotolerant coliforms) and faecal streptococci (FS) counts in wells with the onset of rains, in the village of Keneba in The Gambia, although no statistical analysis was reported.In rural Guatemala, the concentration of E. coli in shallow wells was reportedly 3.8 times higher at sites where rainfall had occurred 24 h before sample collection (Eisenhauer et al. 2016), although this association was not statistically significant.
Heavy rainfall (defined as rainfall in the 90 th percentile of reference data) prior to sample collection led to deterioration in water quality of shallow wells.The detection of Cryptosporidium in wells in India was positively correlated with heavy rainfall 2 days prior to sampling (Daniels et al. 2016), while the detection of E. coli in wells in Bangladesh was positively corelated to heavy rainfall occurring in the 7, 15 and days prior to sampling (Wu et al. 2016).The concentration of E. coli in the wells was significantly associated with the number of heavy rainfall days in the 3-day period prior to sampling only (Wu et al. 2016).Guo et al. (2019) identified 2 rainfall-related variables as predictors of E.coli contamination in wells in Tanzania: the number of heavy rainfall days (days with rainfall > 10 mm) in the 14 days prior to sampling, and standardized precipitation index 2 (SPI).SPI had a stronger effect on E. coli count in wells with handpumps, compared to wells with motorized pumps.On the other hand, heavy rainfall within 14 days of sample collection was a better predictor of E. coli count in wells with motorized pumps.
Longer-term rainfall averages also affected well water quality.The 1 MPN: most probable number 2 SPI measures how much wetter or drier the month is compared to normal conditions for that time of year A. Nijhawan and G. Howard median well water TTC levels (13 CFU/100 mL)3 were an order of magnitude higher than TTC levels (2 CFU/100 mL) during months with above 40 mm rainfall, compared to months with less than 40mm of rainfall in northern Mozambique (Cronin et al. 2006).Nogueira et al. (2003) also reported an increase in FC counts of well water with an increase in monthly rainfall amount, although no statistical measures of association were reported.Lapworth et al. (2020) found a statistically significant correlation between TTC levels in wells during the dry season and annual average rainfall in Ethiopia, Malawi and Uganda.Studies reported mixed results on the effect of rainfall on water quality in deep wells.Cryptosporidium or Giardia detection in deep wells in India were not associated with short-term or seasonal cumulative rainfall patterns (Daniels et al. 2016).Engström et al. (2015) reported a similar lack of correlation between contamination (TTC) in deep boreholes and short-term rainfall, in South Sudan.However, there was a significant association between contamination and long-term rainfall depth (5-day and monthly cumulative).
Antecedent rainfall was linked to faecal contamination in protected springs in Uganda (Howard et al. 2003;Taylor et al. 2009b).Correlation between median TTC count and amount of rainfall in the 24 h and 48 h before sample collection was attributed to the rapid recharge of protected springs (Howard et al. 2003).Taylor et al. (2009b) detected more than 10 3 CFU/100 mL TTC in spring water within one hour of rainfall, when the daily rainfall exceeded 5 mm/day.More intense rainfall (exceeding 20 mm/day) was associated with even higher contamination (> 10 4 CFU/100 mL) (Taylor et al. 2009b).Levels of FS were also strongly correlated with 24 h and 48 h rainfall amounts (Howard et al. 2003).Spikes in faecal contamination in springs within an hour of commencement of rainfall were also reported by Buckerfield et al. (2019) in southwest China, and water quality remained impaired 1 to 4 days after rainfall.While Nogueira et al. (2003) found an increase in FC counts of spring water with an increase in monthly rainfall amount, no statistical measures of association were reported.
Heavy rainfalls at the end of drought have been linked to faecal contamination of groundwater in Ethiopia during the El Nino 2015/16 drought (MacDonald et al. 2019).The three types of the sources examinedsprings, boreholes with handpumps, and hand-dug wellshad negligible contamination during the dry spell but were contaminated with TTC during the first rainfall marking the end of the drought.The water quality improved after the first flush and did not deteriorate during the heavy rains later in the year.
Short-term rainfall effects were reported for piped-water systems in multiple studies.The presence of faecal contamination indicated by a positive result using the H 2 S rapid test (Manja et al. 1982) in public standpipes connected to mechanized boreholes and an elevated storage tank in rural India, was significantly correlated with a heavy rainfall event (rainfall in the 80 th percentile of observed values in the study period) 1 to 7 days before sample collection, but not with heavy rain 8 to 14 days prior (Mertens et al. 2019).Similarly, E. coli levels in tap samples in an intermittent water supply in India were statistically higher when rain had occurred within 24h of sample collection (Kumpel and Nelson, 2013).In Tanzania, both SPI and heavy rainfall (rainfall > 10mm) within 14 days of sample collection led to higher E. coli counts in samples collected from public taps.Samples from water piped to the house was not strongly affected by either climate variable (Guo et al. 2019).
Cumulative rainfall was also significantly associated with the number of contaminated tap samples reported, with both weekly and monthly rainfall amounts affecting water quality.Bastaraud et al. (2020) reported on the results of a 32-year monitoring program of the urban piped water supply in Antananarivo, Madagascar.The number of contaminated samples were correlated with the amount of weekly rainfall.Sulfite-reducing clostridia (SRC) and intestinal enterococci (IE) were detected after four and five weeks of cumulative rainfall.While no correlations were found between cumulative rainfall and E. coli, the relatively small number of positive samples could have affected the performance of the statistical model.In Brazil, the percentage of samples positive for faecal coliforms (FC) increased with monthly rainfall amount for both chlorinated and unchlorinated tap samples, although no statistical measures of correlation were reported (Nogueira et al. 2003).J-Nkanga (1980) reported the combined effects of rainfall and hazards in the distribution system on piped water quality in Benin City, Nigeria.An anti-microbial resistant strain of E. coli artificially introduced into the sewers was detected in tap water samples from households near the septic drainage field, confirming sewage contamination in the water supply.The concentration of the detected strain varied with rainfall amounts in the 7-day prior to sampling, but no statistical measures of association were reported (J-Nkanga, 1980).

Temperature
Three studies reported on the effect of temperature on levels of faecal indicator bacteria.Detection of E. coli in wells in rural Bangladesh was more likely with higher mean temperatures in the 7-and 15-day period prior to sampling.Concentration of E. coli was positively correlated with the number of hot days (days with temperature in the 90 th percentile of reference data) in the 7-and 15-day period before sampling but negatively with the average temperature in the 3-day period prior to sampling (Wu et al. 2016).In Tanzania, the only temperature variable to be correlated with E. coli counts in piped water and wells was daily minimum temperature (Guo et al. 2019).The percentage of samples positive for faecal coliforms (FC) in a piped water supply in Brazil increased with increase in water temperature for both chlorinated and unchlorinated samples from piped water supply, protected springs, and private wells, although no statistical measures of correlation were reported (Nogueira et al. 2003).

Observational studies
The most frequently reported proxy for cyanobacteria biomass was chlorophyll-a concentration.Several studies reported positive correlations between ambient or water temperature and chlorophyll levels during a cyanobacteria bloom, although this association varied with the temperature range over the study period, thermal stratification and nutrient levels in the water column, and the dominant cyanobacteria genus (Berger et al. 2006;Deutsch and Alameddine, 2019;Hennemann and Petrucio, 2016;Hoang et al. 2018;Li et al. 2018a;Ni et al. 2012;Touati et al. 2019;Wang et al. 2013;Yang et al. 2019;Zhang et al. 2016).Chlorophyll-a levels were also directly correlated with cumulative rainfall 15 days prior to sampling (Li et al. 2018b) and negatively correlated with reservoir volume under drought conditions (Rocha et al. 2018).
Several studies reported a positive correlation between temperature and Microcystis biomass (Duong et al. 2013;Li et al. 2017;Liu et al. 2011;Touati et al. 2019;Xu et al. 2010;Zhao et al. 2019).Deng et al. (2014) and Guo et al. (2017) observed that temperature was the critical factor controlling Microcystis density (cells/mL) once the nutrient thresholds were exceeded.Anabaena biomass was correlated only with photosynthetically active radiation, but not water temperature (Li et al. 2018a;Wu et al. 2006;Zhang et al. 2016).C. raciborskii was the dominant Cylindrospermopsis species detected in studies, but its biomass was also not associated with temperature in several locations (Bouvy et al. 1999(Bouvy et al. , 2000;;Lei et al. 2014).
Microcystin (MC) toxins, predominantly MC-RR, MC-YR and MC-LR (Major et al. 2018) were isolated from Microcystis species and some Anabaena species (Otten and Paerl, 2011), while strains of C. raciborskii were associated with the toxin cylindrospermopsin (CYN) (Lei et al. 2014).The association between MC concentrations and temperature was inconsistent across studies (Amé et al. 2003;Barros et al. 2019;Dao et al. 2016;Duong et al. 2013;Li et al. 2017;Mankiewicz-Boczek et al. 2015;Nasri et al. 2007;Ni et al. 2012;Wu et al. 2006;Yu et al. 2014) and depended on the dominant cyanobacterial species present.High MC levels were generally associated with toxic M. aeruginosa and M. flos-aquae species (Chen et al. 2009;Mohamed et al. 2015;Otten and Paerl, 2011).Wang et al. (2002) found that the relationship between MC concentration and temperature depended on the type of toxin present, with only MC-YR significantly correlated with temperature.Similar mixed results were also reported across studies on association between temperature and CYN levels (Barros et al. 2019;Lei et al. 2014).
The impact of rainfall on cyanobacteria blooms was closely linked to the amount of thermal destratification induced by the event and prevalent temperature conditions.Autumn rainfall in a deep reservoir in China was associated with an absence of blooms the following summer because input sediment load with a higher density than existing bottom layers in the reservoir led to de-stratification of the water column (Li et al. 2015).In contrast, Guo et al. (2018) reported that the sequence of heavy rains, which transported large amounts of nutrients into the river, followed by hot and dry weather, causing reduced river inflow, triggered a large bloom in Qianting river in China.Bloom duration in Lake Taihu increased with decrease in precipitation and wind speed between 1987 and 2009 but was not significantly correlated with annual averages of temperature or precipitation (Zhang et al. 2012).
Half of the cyanobacterial blooms in Lake Taihu between 2007 and 2015 occurred within 4 days following extreme rainfall events with strong winds Yang et al. (2016).Typhoons had a similar impact on cyanobacterial blooms.After an initial disruption immediately following the typhoon, blooms reappeared with higher biomass (Zhu et al. 2014) and covered a larger area (Chen et al., 2020) than before the typhoon but gradually reduced several days after it had passed.
Few studies examined the impact of cyanobacteria on drinking water supplies.Toxin levels (MC-LR and MC-RR) in Nile River source water at the treatment plant intake in Damietta city, Egypt were positively correlated with temperature, but the temperature dependence of toxin removal efficiency of water treatment steps was not explored (Mohamed et al. 2015).Toxin levels in domestic reservoirs receiving water from the treatment plant exceeded the WHO guideline value (1 µg/L MC-LR) and were also correlated with temperature (Mohamed et al. 2016).Nasri et al. (2007) found no temperature dependence of toxin levels at the raw water intake in Cheffia Dam, (Algeria) but did not explore associations between efficiency of toxin removal and temperature.

Modelling studies
In models developed by Zhu et al. (2018), temperature was a good predictor of cyanobacteria biomass in Lake Erhai, used to supply drinking water to Dali City in China.Longyang (2019) modelled effects of increasing temperatures on blooms in a deep lake in southeast China and found that an increase in autumn and winter temperatures had a bigger impact on chlorophyll-a concentrations than summer temperatures, since summer conditions are already favourable for cyanobacteria growth.Climate change under RCP 8.5 was projected to impact cyanobacteria blooms through an increase in surface water temperature and extended thermal stratification in the summer, especially in the long-term (2070-2099) (Tang et al. 2015).Compared to the baseline year (2005), the blooms were also projected to cover a larger area in the lake and last longer.
Simulations by Shan et al. (2020) predicted that an increase in water temperature would lower the nutrient (nitrogen and phosphorus) thresholds required to limit blooms and maintain the MC concentration below the WHO guideline of 1 µg/L.These thresholds varied across lakes in China.Similar trends were presented by Huo et al. (2019) for lakes across China under RCP 2.6, 4.5, 6.0 and 8.5 scenarios.In the worst-case scenario under RCP 8.5, a 4.8% and 37% reduction in thresholds for total nitrogen and total phosphorous respectively, would be required to prevent bloom formation (Huo et al. 2019).

Observational studies
The observational studies on saltwater intrusion in coastal and island groundwaters were predominantly from south Asia.Akter et al. (2020) found a weak statistical correlation between groundwater salinity and land surface temperature (LST), and moderate correlation between salinity and potential evapotranspiration (PET) in coastal Bangladesh.Thilagavathi et al. (2017) reported no association between annual average rainfall and groundwater electrical conductivity (EC) in Pondicherry, India, based on seasonal data collected over a ten-year period from 2006 to 2016.However, no statistical measures of correlation were reported and no consistent trends in long term average annual rainfall were detected at the study setting.Sreekesh et al. (2018) found evidence of coastal erosion and sea level rise along the coast of Kerala, India; annual rate of erosion was correlated with increase in EC and total dissolved solids (TDS) of groundwater.In a study across Comoros Islands, Kenya and Tanzania, Comte et al. (2016) observed a correlation between TDS and annual rainfall only in volcanic aquifers, indicating the effect of geology on the response of water quality to rainfall.Chattopadhyay and Singh (2013) analysed groundwater quality in the Lakshadweep Islands in the Western Indian Ocean, during above and below average (± 30%) annual rainfall conditions.Wells near the coast had the highest TDS (1400 to 2300 mg/L), indicating slight to moderate salinity during the drier year.In comparison, the highest TDS reported during years with above average rainfall in the same wells was 900 mg/L.Wells towards the centre of the island had <600 mg/L TDS in both years, suggesting higher vulnerability of groundwater along the rim of the island.Gingerich et al. (2017) studied the impact of a coastal storm in the Republic of Marshall Islands during a period of regional sea level rise linked to the El Niño Southern Oscillation (ENSO).Seawater flooding caused by the storm swell increased chloride concentrations in well waters from 53-195 mg/L to 10,080-20,880 mg/L (seawater Cl − concentration was 19,600 mg/L).These concentrations were reduced to 1800-3990 mg/L after ten days; however, it took 22 months for the concentrations to reduce to pre-flood levels, aided by artificial recharge with rainwater that began seven months after the flood.

Modelling studies
Modelling studies predicted saltwater intrusion (caused by an inland shift of the freshwater-saltwater transition zone) and land surface inundation in coastal aquifers and shrinking of island freshwater lens (Sathish and Elango, 2019;Sefelnasr and Sherif, 2014).Aquifers with higher hydraulic conductivity in coastal Argentina (Carretero et al. 2013), and smaller hydraulic gradients in the Nile Delta aquifer (Sherif and Singh, 1999) were more susceptible to saltwater intrusion than aquifers with lower hydraulic conductivity or larger hydraulic gradient.
Increased groundwater extraction was predicted to increase saltwater intrusion caused by sea level rise in coastal Argentina (Carretero et al. 2013).Simulations by Abd-Elhamid et al. (2016a, 2016b), Abdelaty et al. (2014) and Mabrouk et al. (2018) for different parts of the Nile Delta aquifer in Egypt detected comparable or greater landward shifts from increased groundwater pumping than from sea level rise.The freshwater lens in Kish Island in Iran was projected to shrink by less than 10% under three sea level rise scenarios -0, 1 and 4m.This effect became more pronounced with decreased groundwater recharge but could potentially be offset through increasing recharge by 20% (Mahmoodzadeh et al. 2014).Similarly, reducing groundwater extraction to 50% of current rates was projected to offset the loss of freshwater resources in the Nile Delta aquifer caused by 0.1 m sea level rise (Abd-Elhamid et al. 2016a) and 0.5m sea level rise, but not 1m sea level rise (Sefelnasr and Sherif, 2014).
The simulated groundwater salinity of wells in coastal Karnataka in India in the year 2034 considering 1 mm/year sea level rise, 50% decrease in recharge and 200% increase in extraction was similar to the salinity increase caused by just the changes in recharge and extraction, relative to baseline conditions in 2014 (Lathashri and Mahesha, 2016).Sea level rise of 1.2m in the Bay of Bengal was predicted to increase salinity at the intake of a water treatment plant in the coastal city of Chittagong, Bangladesh from the current level of < 1 parts per thousand (ppt) to between 2 and 2.5 ppt (Akhter et al. 2012).
In the Laccadive Islands off the western coast of India, sea level rise of 0.05 and 0.1m was projected to shrink the freshwater lens in the centre of the island by 28% and 60%, respectively (Bobba et al. 2000).The lens thickness on the outer rim of the island was projected to decrease to zero, implying mixing of freshwater and seawater.This is similar to the observed findings of Chattopadhyay and Singh (2013) for the same group of islands.Okello et al. (2015) modelled the freshwater lens on Lamu Island in Kenya in the 2090s with estimates of local sea level rise derived from the global mean projections of 0.43m and 0.48m for A1b and A2 emissions scenarios (Nakicenovic et al. 2000), respectively.The volume of the freshwater lens was projected to increase by 60% under A1b scenario.In contrast, A2 conditions were expected to shrink the volume by 78%.This difference was attributed to the increased potential evapotranspiration, caused by the higher average temperatures under A2 as compared to A1b.While monthly precipitation was projected to increase under both scenarios, the higher temperatures under A2b had a drastic impact on the simulated groundwater quantity and quality.Rodríguez et al. (2004) reported an inverse relationship between arsenic concentration and monthly mean precipitation amount in Zimapán valley in Mexico, while Tathagata and Rolee (2011) reported an inverse relationship between arsenic concentrations and mean seasonal rainfall amounts in West Bengal.No statistical analysis was reported in either study.Contrary to these findings, Gonçalves et al. (2007) found a positive association between monthly rainfall amount and arsenic concentrations in Ouro Preto, Brazil, although no statistical measures were reported.

Geogenic contaminants
Only one study was identified that explored the relationship between rainfall and fluoride concentrations in groundwater.Davraz et al. (2008) indicated that the fluoride content of spring water in Turkey decreased during months with rainfall but did not report statistical measures of association.

Summary of evidence
Rainfall was the most frequently reported predictor of microbiological contamination.The majority of studies identified were of small, point sources of water supply generally from groundwater.These supplies tend to use simple technologies, and often lack adequate sanitary protection and water treatment, making them more susceptible to contamination.The reviewed studies showed that such supplies respond rapidly to heavy rainfall which can cause elevated contamination through rapid recharge of shallow groundwater or contaminants being washed into shallow sources with faulty headworks (Daniels et al. 2016;Howard et al. 2003;MacDonald et al. 2019;Taylor et al. 2009b).Heavy rainfalls following drought are likely to cause 'first flush' contaminant spikes in rural areas, as waste accumulates near sources during the prolonged dry period (MacDonald et al. 2019).Differences in climate predictors of faecal contamination between urban and rural settings pointed to variation in patterns of faecal loading (Buckerfield et al. 2019;Guo et al. 2019), although further investigation of these effects is needed.
Deterioration in water quality was also closely linked to population density (Howard et al. 2003); presence of animals at the study site (Daniels et al. 2016); groundwater level and ponding at the source (Kulabako et al. 2007); and percentage of developed area around sources (Wu et al. 2016).The associations with topography, local hydrogeology, and land use (Daniels et al. 2016;Engström et al. 2015;Wu et al. 2016) have important implications for improving resilience of water and sanitation services to climate change.
These factors, combined with lack of disinfection, would suggest such water supplies will be the most vulnerable under climate change.It is quite likely that the nearly 800 million people living in LMICs who lack access to even a basic water supply will get access only to such shared, community-managed water supplies as a first step in improving access to water supply.The evidence of this review suggests that greater attention must be paid to disinfecting water from such point sources, most likely through household water treatment.It would be preferable, however, to focus on moving people to higher levels of water supply service that include disinfection, (WHO, 2017b) and preferably to continuous, safely managed piped water at home (Howard, 2021).
Evidence from the reviewed studies indicates that even chlorinated water supplies may be vulnerable to contamination, with public taps responding to rainfall and temperature more strongly than water piped to the house, especially when hazards are present near the distribution network (Guo et al. 2019;Nogueira et al. 2003).J-Nkanga (1980) note that damaged pipes passing through septic tank drain fields led to high levels of contamination in piped water supplies during heavy rainfall.This threat is reinforced from other studies that note that the risk of disease from contaminated water in a large utility in Uganda was primarily driven by contamination in distribution systems (Howard et al. 2006b).Intermittent supplies, which serve about 1 billion people in LMICs are also susceptible to seasonal deterioration in water quality (Bivins et al. 2017;Etchie et al. 2014;Kumpel and Nelson, 2013).The implication of these findings is that management of piped water supplies in LMICs will need to substantially improve if they are to cope with the additional threats from climate change.
The studies on cyanobacterial blooms in water supply reservoirs predominantly focused on temperature as the exposure variable, and its effect on the dominant cyanobacteria genus, nutrient levels and temperature range observed during the study.Cyanobacteria blooms can be expected to occur more frequently under a warming climate because their growth is optimized at higher temperatures, while the growth of other primary producers levels off or declines as temperatures exceed 25 • C (Paerl and Paul, 2012).Warmer conditions also intensify thermal stratification in lakes and reservoirs, which gives a further advantage to buoyant cyanobacteria species (Paerl and Paul, 2012).
Temperature effects on toxin production varied between studies and seemed to depend on the specific genus and type of toxin present (Barros et al. 2019;Wang et al. 2002), but not on the climate zone of the lake or reservoir (Dao et al. 2016;Mankiewicz-Boczek et al. 2015;Nasri et al. 2007).Heavy rainfall and windstorms led to a temporary release of nutrients from resuspended sediments in shallow lakes and transported blooms through the lake, causing blooms to reappear after an initial disruption during the event (Chen et al., 2020;Yang et al. 2016;Zhu et al. 2014).The effect on bloom formation seemed to depend on the weather conditions immediately after the event, and the amount of thermal destratification caused by it (Guo et al. 2018, Li et al. 2015, Li et al. 2018b).
Decreased rainfall under future climate change may reduce nutrient input from non-point pollution.On the other hand, more intense rainfall and windstorms can increase nutrient transport and sediment resuspension but also cause mixing and thermal destratification (Chorus and Welker, 2021).Therefore, the effect of climate change on cyanobacteria blooms, and their toxicity, may vary between locations, depending on nutrient availability, local climate patterns and type of bloom-forming strains present.Studies projected stricter control over nitrogen and phosphorus input into surface waters to minimise bloom formation (Guo et al. 2017;Huo et al. 2019;Shan et al. 2020), especially where warmer winters are expected (Longyang, 2019).
Annual rainfall was the only climate variable significantly associated with coastal aquifer salinity in the observational studies (Chattopadhyay and Singh, 2013;Comte et al. 2016;Thilagavathi et al. 2017).However, the lack of long-term climate or water quality data limited the strength of evidence presented in these studies.Sea level rise (Akhter et al. 2012;Carratero et al. 2013;Sherif and Singh, 1999), long-term changes in precipitation (Lathashri and Mahesha, 2016;Razack et al. 2019;Zhao et al. 2016), and evapotranspiration (Okello et al. 2015) were projected to cause a gradual increase in salinity of coastal groundwater and a decrease in the volume of freshwater.
Increase in sea level and changes in aquifer recharge can lower the hydraulic head of the aquifer relative to seawater and seaward groundwater flux, causing the freshwater-saltwater interface to shift inland ( (Werner et al., 2012)).However, as multiple studies reported, the overexploitation of groundwater is projected to have a comparable or greater effect on saltwater intrusion than sea level rise and maintaining current levels of groundwater recharge and pumping will be essential to mitigate these effects.
Literature on geogenic contamination and climate variables that met our inclusion criteria was scarce and none of the studies included reported statistical measures of association.This precludes any inferences on the likely effects of climate change.Multiple studies have reported temporal variations in arsenic concentrations with season and groundwater levels, pointing to possible links with recharge in some locations, but the reported trends are not uniform (Cheng et al. 2005;Savarimuthu et al. 2006).

Limitations of the review
There are a number of limitations with this scoping review.Some papers with titles and full text in Spanish and Portuguese were rejected over language which may have led to an under-representation of studies form South and central America.Furthermore, no studies in Chinese or Russian were identified in the search, which may have excluded work from China and countries in the former Soviet Union, respectively.
The search was limited to studies in peer-reviewed journals and did not include technical reports, or articles published in magazines of engineering societies or other professional organizations.This may have excluded reports on effects of temperature and rainfall on water quality from utilities and engineering firms.
There was over-representation of certain geographical areas in the included studies.The studies on cyanobacteria were overwhelmingly from two countries (Brazil and China).Despite reports of cyanobacteria blooms from India and 21 countries across Africa (D'Silva et al. 2012;Ndlela et al. 2016), few studies reported quantitative measurements of temperature or rainfall and therefore, did not meet our inclusion criteria.

Recommendations for research
There is a need for further research that investigates the relationships between pathogens and climate variables and to produce more accurate estimates of public health risks.This may include improving quantitative microbial risk assessment models for water supplies in LMICs that can be linked to climate models.Given advances in analytical techniques and increasing capacity in LMICs there are greater opportunities for such studies to be undertaken.Relying on faecal indicator bacteria alone will likely under-estimate the potential impact of climate change on water quality, given that viral and protozoan pathogens in particular have very different survival rates in untreated water and resistance to water treatment processes (Betancourt and Rose 2004;Pedley et al. 2006).
There are also clear needs for further research on the relationships between piped water and climate variables in LMICs, given increasing rates of access to piped systems.This should include investigation of the likely impact of climate change on biofilm and opportunistic pathogens.Increasing ambient temperatures and source waters with high nutrient loads may well result in increasing issues with biofilms in LMICs (Hallam et al. 2001;Liu et al. 2016) and this would therefore seem to be an important area of future research.
No studies examining the implications for drinking water treatment met our inclusion criteria, despite the strong qualitative evidence of the potential impacts for water supply and need for adaptation (Delpla et al. 2009).Therefore, research into how climate change may affect water treatment and groundwater protection requirements is a further priority.
Further research into the links between temperature and toxin production, and the effect of temperature on nutrient thresholds is needed at the reservoir-level as these associations are often location-specific.Actions to prevent an increase in algal blooms will be important, in part because cyanobacteria have been associated with pathogens like Pseudomonas aeruginosa, Legionella pneumophila and Vibrio cholerae (Bomo et al. 2011;Chaturvedi et al. 2015;Taylor et al. 2009a).
In addition, algal blooms pose challenges for water treatment (Newcombe et al. 2021).No studies were identified on the implications of cyanobacterial blooms for water treatment processes.Blooms have been known to cause taste and odour problems in treated water, accumulate in sludge beds and release toxins, and increase potential for disinfection by-product (DBP) formation (Khan et al. 2015;Zamyadi et al. 2012Zamyadi et al. , 2013)).The removal of cyanotoxins and taste, odour and colour-producing compounds may require optimizing existing water treatment processes and additional processes such as granular activated carbon or membrane filtration, thus increasing treatment costs.The cost-effectiveness of additional water treatment over better catchment management and nutrient control in LMIC settings warrants more attention.The studies on saltwater intrusion did not directly consider the implications for drinking water supplies in affected areas, although all locations in the studies reported relying substantially on groundwater for domestic supply.Given the emerging evidence that increasing salinity in drinking water may be linked to health problems including hypertension and pre-eclampsia (Khan et al. 2011;Vineis et al. 2011), further work in this area is warranted.Unlike high-income countries, coastal communities in LMICs may not have access to expensive desalinization processes and managed aquifer recharge may be a more cost-effective adaptation option.This is gaining increasing attention in countries such as Bangladesh with increasingly saline groundwater and polluted surface water, but feasibility systems require careful assessment (Naus et al. 2021).Naser et al. (2021) found that people consuming MAR water had higher urinary sodium and higher blood pressure than those continuing to rely on surface water sources and concluded that MAR should not be promoted as a routine water supply option.Such risk substitutions have been identified before in Bangladesh in relation to arsenic (Howard, 2003), but the risks with microbiological contamination tend to substantially outweigh those associated with chemical contaminants (Howard et al. 2006a).The limited geographical range of the studies also suggest for greater attention, particularly in parts of coastal Africa that are affected by major windstorms.
Between 94 to 220 million people worldwide are potentially exposed to high arsenic concentrations in groundwater, of which over 94% live in Asia (Podgorski and Berg, 2020) and a similar number is estimated to rely on groundwater with elevated levels of fluoride, mostly in north Africa and Asia (Amini et al. 2008;Edmunds and Smedley, 2013).Given the global importance of groundwater for drinking supply, further research into potential effects of climate change is warranted.Furthermore, evidence from the Andes of water quality deterioration linked to glacier recession (Fortner et al. 2011;Guittard et al. 2020) suggests more research into the effect of climate change on chemical water quality in glaciated river basins is also needed.
The potential for wildfires is expected to increase in parts of Asia and Africa (Liu et al. 2010), but no studies from LMICs were identified on the potential impact on water supplies from forested catchments.Since hotter, dryer conditions are expected to increase in several LMICs, research is needed into the effects of wildfires on water quality, especially considering the effects of peak runoff events.Finally, we found few studies discussing the impact of multi-year drought on water quality.As several LMICs are projected to face greater rainfall variability and may suffer from longer dry spells including drought, this is an important area for future work.

Conclusions
There is a growing body of evidence on the threats of climate change to water supplies in LMICs.Changing rainfall and temperature patterns are likely to exacerbate existing challenges around safe water provision.Across source types and settings, climate threats to water quality were either closely influenced by, or in some cases, exceeded by the threats from population growth, poor sanitary protection, and poor catchment management.
The vulnerability of many simple point sources of water is of concern given the large numbers of people continuing to rely on such water supplies and the prospect that millions more may only be offered such water supplies in the medium-term.That these water supplies are already highly vulnerable to climatic variables suggest even greater attention is needed to provide safely managed water at people's houses, preferably via piped networks managed by regulated suppliers.
Given the evidence that climate change is likely to lead to more intense rainfall and dry periods (IPCC, 2021) these relationships suggest that without action, water quality will further deteriorate with climate change.This demonstrates the urgent need to consider climate factors and climate change in the planning and operation of water supplies.This will require improvements in long-term monitoring, water quality sampling immediately following climate events, and wider uptake of risk management tools like Water Safety Plans.This review also identified key areas of research that can strengthen our understanding of climate threats in low-and middle-income settings.

Fig. 1 .
Fig. 1.Conceptual framework of climate impacts on source water quality.

Fig. 2 .
Fig. 2. PRISMA flow diagram of the literature search and screening for the scoping review on associations between climate and drinking water quality in low-and middle-income countries.