Markus Andrews

In Lake Hallwil, the contribution of oxic methanogenesis to overall diffusive CH4 emissions has been estimated to be 90%6 or 63–83%5, but we show here that NOMC ~ 17%.
In the mass balance of the SML extending from 0 to 5 m water depth5,6, Günthel et al.5 used an average sediment flux of Fsed = 1.75 mmol m−2 day−1, averaging flux estimates of Donis et al.6 from two sediment cores, one collected at 3 m and the other at 7 m water depth. The δ13C of the CH4 in the pore water of these two cores differ substantially6, indicating differences in production and oxidation of CH4 between the sediments in the SML and at 7 m water depth. The estimate of Fsed in the SML should therefore be based on the core collected at 3 m water depth. Using the approach of Donis et al.6, the correct Fsed derived from the data of this core is Fsed = 2.8 mmol m−2 day−1 (Peeters et al.9, see Supplementary Note 2.1 for details).
Günthel et al.5 and Donis et al.6 apparently have erroneously used gas transfer coefficients instead of proper CH4 fluxes to calculate emissions. This conclusion is demonstrated by the perfect agreement between the values published erroneously as CH4 fluxes, Fsurf, by Günthel et al.5 and the values of the gas transfer coefficients of CH4 at 20 °C, kCH4, calculated by us (Table 1). The values published by Donis et al.6 as CH4 fluxes are very similar to these kCH4 and therefore also do not represent CH4 fluxes but gas transfer coefficients (for details, see Supplementary Note 2.2).
Table 1 Average CH4 surface fluxes from Lake Hallwil and gas transfer velocities.
Full size table

The gas transfer coefficient of CH4 must be multiplied by the difference between the surface concentration (0.3 mmol m−3, ref. 6) and the atmospheric equilibrium concentration of CH4 (CH4,equ = 0.003 mmol m−3 at 20 °C9), i.e. by ~0.3 mmol m−3, to obtain Fsurf. Fsurf is therefore ~3.3 times smaller than the values of the gas transfer coefficients erroneously taken by Günthel et al.5 and Donis et al.6 as CH4 fluxes (Table 1 and details in Supplementary Note 2.2).
Donis et al.6 and Günthel et al.5 used values obtained from measurements with floating chambers to calculate emissions, but these values claimed to represent Fsurf appear to be in fact values for transfer coefficients, suggesting the same mistake as in the case of the wind models. Donis et al.6 stated: “Average flux (April–August 2016) is equal to 0.8 ± 0.2 mmol m−2 d−1 from MacIntyre relationship for positive buoyancy and to 0.6 ± 0.3 mmol m−2 d−1 from chamber measurements. The latter, not significantly different from the wind-based relationship, was used for the mass balance”. Günthel et al.5, co-authored by D. Donis, claim that the “MacIntyre relationship for positive buoyancy”10 provides an average value of 0.7 for Fsurf, but in fact 0.7 is the average value for kCH4 in unit m day−1 (0.7 m d−1, see Table 1) and Fsurf for this model is 3.3 times smaller (0.21 mmol m−2 d−1, see Table 1). The value by Donis et al.6 for the MacIntyre relationship10 is even slightly larger than 0.7 and therefore clearly incompatible with Fsurf but is rather a gas transfer coefficient as is obvious in the case of Günthel et al.5. The good agreement between the value for the gas transfer coefficient obtained from the MacIntyre model for positive buoyancy flux10 and the values from the chamber measurements suggests that the values from the chamber measurements are not gas fluxes but also gas transfer coefficients (see Supplementary Note 2.2 for more details).
Donis et al.6 derived from their chamber measurements the wind-based model “Hallwil relationship” specifically for Lake Hallwil. The establishment of this Hallwil relationship required that Donis et al.6 calculated gas transfer coefficients from their chamber measurements. In their Supplementary Fig. 4, Donis et al.6 show that the values from their chamber measurements agree well with those from the Hallwil relationship (Supplementary Fig. 2 and Supplementary Note 2.2). Note, however, that the values for the Hallwil relationship are in fact gas transfer coefficients and not Fsurf, supporting that also the values from the chamber measurements represent gas transfer coefficients and not Fsurf (Supplementary Fig. 2 and Supplementary Note 2.2 for more details). This conclusion implies that the values from the chamber measurements by Donis et al.6 must be multiplied by ~0.3 mmol m−3 to give proper CH4 fluxes, which are then ~3.3 times smaller than the CH4 fluxes used in the mass balances of refs. 5,6.
Because there are only four chamber measurements available for 2016 and one of them was exceptionally low (see ref. 6 and Supplementary Note 2.2), the Hallwil relationship is considered here to provide the most reliable estimate of the average k600 in Lake Hallwil and therefore applied to calculate the average surface CH4 flux for April to August 2016, i.e., Fsurf = 0.24 mmol m−2 d−1 (see Table 1 and Supplementary Note 2.2). The reliability of the Hallwil relationship was confirmed by Günthel et al.5 and by Hartmann et al.11 comparing different estimates of surface fluxes in the South Basin of Lake Stechlin.
With Fsed = 2.8 mmol m2 day−1 and Fsurf = 0.24 mmol m2 day−1, NOM = 416 mol day−1 and the contribution of NOM to total emissions is NOMC = 17% (Supplementary Table 1 in Supplementary Note 2.3 includes also additional estimates of NOMC). The low value of NOMC suggests that most of CH4 in the SML originates from the sediments, which is consistent with the δ13C isotopic composition of CH4 in Lake Hallwil9. The uppermost CH4 in the sediment core from the SML is characterized by δ13C about –59‰, which corresponds very closely to the δ13C of the CH4 in the open water of the SML ranging from −62‰ to −58‰ (Figs. 4 and 5 both in ref. 6). Thus the δ13C values suggest that the CH4 from the uppermost pore water in the sediment of the SML is the source of the CH4 in the open water and do not indicate a reduction of the δ13C expected in case of substantial CH4 production. More

Microbial community analysis
An overall microbial community analysis is presented as PCA plots and a dendrogram in Fig. 1. The cluster analysis showed good agreement between sample replicates, which clustered most closely. WWTP influent (i.e., untreated sewage) collected in the post-monsoon season clustered most closely with water samples from S4, S5 and S6 collected at same time, while WWTP influent from the monsoon season clustered with water samples from S6 collected in the same season. The WWTP effluent from both monsoon and post-monsoon season clustered together. The PCA plot with data from all the sampling times (Fig. 1b) generally showed a separation of downstream and WWTP influent water samples from the upstream and WWTP effluent samples along principal component 1, with only a few exceptions. Genera mostly found in the human gut microbiome15,26 like Streptococcus, Trichococcus, Lactobacillus, Enterococcus, Prevotella and Arcobacter, were highly prevalent in downstream and WWTP influent water samples, which separated these samples from the upstream water samples in the PCA. Among the three factors analysed (i.e., location, sampling time and water sample types), locations and sampling time had a significant effect on the similarity of the samples in the ANOSIM, although with relatively low R values (ANOSIM; Location: R = 0.29, p value = 0.001 and Sampling time: R = 0.16, p value = 0.01). ANOSIM further indicated no statistically significant differences between the microbial communities in water from locations S4, S5 and S6 and the wastewater influent (ANOSIM; (1) S4 and Inf: R = 0.0309, p value = 0.357: (2) S5 and Inf: R = 0.0617, p value = 0.369 and (3) S6 and Inf: R = 0.0123, p value = 0.3690).
Fig. 1: Cluster and PCA analysis at rank genus for 16S rRNA gene sequencing reads.

a Cluster analysis [all seasons], PCA plot b all seasons, c monsoon [June 2019] and d post monsoon [August 2019]. Arrows in the PCA plots indicate the ten variables with the highest loadings (vector lengths) in the PC1 and 2 space.

Full size image

An interesting picture emerged when a separate PCA (Fig. 1c, d) and cluster analysis (Supplementary Fig. 1) was conducted for water samples from the monsoon and the post-monsoon season. In both seasons, there were substantial, but seasonally distinct, contributions of genera found in the human gut microbiome to the variance among water samples along principal component 1: in the monsoon season Arcobacter, Aeromonas, Streptococcus and Prevotella had significant PC1 loadings; in the post-monsoon season Enterococcus, Acinetobacter, Streptococcus and Trichococcus had significant PC1 loadings. Separation of wastewater treatment plant effluent (WWTP Effluent) samples along PC1 away from the WWTP influent (WWTP Influent) samples in both sampling events signified the benefits of wastewater treatment, because human gut-associated genera became less predominant in treated wastewater microbiomes, as expected27. Accordingly, there was a clear separation of the most upstream water samples from the most downstream water samples along PC1 in both events, with the downstream water samples becoming more similar to WWTP Influent (Fig. 1c, d). Evidently, as the Bagmati River flowed into more densely populated areas, the characteristics of its water microbiome changed from a composition more similar to treated, to a composition more similar to untreated urban sewage, but the composition of the urban sewage was variable for the monsoon and post-monsoon season.
Abundance of human gut and putative pathogenic bacteria in the water microbiomes
A more detailed breakdown of the microbial community composition in the Bagmati River for the monsoon and post-monsoon season is reported in Table 1, and Supplementary Tables 1 and 2, which compare the total percentage relative abundance of putative human gut28 and pathogenic29 bacteria at genus and species level for different sampling sites in the Bagmati River, and the WWTP influent and effluent (Refer to Supplementary Tables 3–5 for more detailed lists of bacteria). Based on our previous findings24, species identities are not always reliable due to the limited read accuracy of the MinION sequencing reads, but the overall trends are nonetheless indicative of changes in microbial composition. For all sampling events, the water collected at the most upstream site S1 and S2 showed the lowest relative abundance for both human gut and putative pathogenic bacteria, whereas the highest relative abundance was observed in the water collected at the most downstream sites S4–S6 (Table 1, Supplementary Tables 1 and 2). The microbial water quality of water samples collected at site S1 can be considered as baseline data, as this watershed is distant from the densely populated Kathmandu Valley and has the minimal influence of human and urbanisation activities. Figure 2, and additional figures in Supplementary Information (Supplementary Fig. 2–9) show how the abundance of human gut and putative pathogenic genera changed in space and time along the Bagmati River. As the river flowed downstream, the abundance of some of these groups of bacteria increased, and the most drastic and significant increase was observed at the sites S4, S5 and S6 downstream of the Pashupatinath Temple as compared to site S1 (Two-sample t test, p value  More

The ongoing COVID-19 pandemic had its first confirmed case in the United Kingdom in late January 2020, but transmission increased rapidly leading to the government imposing a lockdown on the whole population, banning all “non-essential” travel and contact with people outside one’s home on 23 March 2020.1 Globally, the lockdown has caused households to change their typical consumption behaviours drastically across a variety of major categories, resulting in an initial sharp increase in spending, especially in essentials and food items.2 Studies dedicated to the impact of COVID-19 on water consumption focused on aggregate demand and general demand peaks. For example, in Germany, a significant shift in aggregate demand peak was observed from 07:10 pre-lockdown to 09:40 during lockdown.3 In a Waterwise4 report, certain regions in the UK saw a 35% increase in peak daily consumption during the lockdown. In Brazil, analysis of data from 26 days pre-lockdown and 26 during lockdown has revealed an 11% increase in household water consumption attributable to the lockdown.5 Although this rise can be generally attributed to an increase in diurnal consumption owing to people remaining at home, increase in preventive behaviours such as hand-washing1 also became contributory factors.
Household water demand in England and Wales is already at an all-time high, constituting 55% of the 32 Cubic Gigametres per year (Gm3/yr) total UK household water consumption footprint, with southeast England having the highest per capita consumption (PCC) and being already declared as severely water-stressed.6,7,8 The impacts of the extended time people stayed at home under the lockdown and the ensuing changes in behaviour arising from this have been an increase in household water demand, exacerbating existing pressure on network water supply.
Water utility companies are increasingly searching for ways to understand the full nature of household water use, how to improve network demand forecasting and achieve effective water efficiency interventions. By presenting a data-driven detailed characterisation of household clusters, including their unique patterns, we have demonstrated how the understanding of the impact of these unique patterns of behaviour on network demand can help in the design of demand forecasting and intervention that targets households on the basis of their shared cluster characteristics. Many demand strategies have relied on existing socioeconomic (SE) and sociodemographic (SD) household variables (e.g., ACORN9) and self-reported behaviours through surveys and water use diaries.10,11 Our work not only significantly enhances the precision of forecasting and intervention when enriched with SE and SD variables, but also provides a scalable framework for the inclusion of ordinary-metered and unmeasured households that share SE/SD characteristics peculiar to particular clusters.
We analysed the weekly water consumption data, at an hourly resolution, for January to May 2020 of 11,528 smart-metered households. We then classified the households according to their temporal patterns of consumption, highlighting their unique characteristics and their respective shares of relative and absolute consumption before and during the COVID-19 lockdown.
All households in the study are from a single water provider, collected across two geographical areas 50 miles apart consisting of 24 District Metering Areas (DMAs). As the aim of this study was to quantify the impact of the Covid-19 lockdown on aggregate water demand while highlighting household clusters’ underpinning temporal demand patterns, only anonymised smart meter data was utilised. Data on SD/SE or occupancy variables of the participating households were not available.
Overall temporal water consumption patterns
The analysis revealed an average consumption of 3256 cubic metres per day (m3/d) for the 11,528 households across the network for the period before the COVID-19 lockdown, equating to a per household consumption (PHC) of 284 litres per day (l/h/d), as per the UK average.12 Consumption remained even between the first week of January (J1) and the first week of February (F1) averaging 350 m3/d (291 l/h/d), followed by a 20% decline in February week 2–3 (F2–F3), before returning to average values in February week 4 (F4) to March week 3 (M3) as in Fig. 1b. A sharp increase was recorded in March week 4 (M4), the week of the COVID-19 lockdown, to 3756 m3/d (326 l/h/d), a rise of 10% on the previous week, reaching 4747 m3/d (411 l/h/d) in May week 4 (MY4), some 46% above pre-lockdown average. The cause of the 20% drop in consumption in the second week of February 2020 remains unknown. The water utility did report the loss of four days of data in that period owing to equipment power outage. The absence of any other plausible cause is suggestive that this may have resulted from “Storm Ciara”,13 which brought heavy rain and very strong winds to the region on 9 February 2020, causing widespread power issues.
Fig. 1: Households’ consumption patterns and trends before and during the COVID-19 lockdown in the UK.

a Differences in per household consumption (PHC) for January–May 2019 and 2020. b Weekly average 24-hour consumption for all households–January week 1 (J1) to May week 4 (MY4)—showing normal consumption trend, anomaly due to data loss and increase in consumption during lockdown period. c Hourly consumption patterns, showing households’ average proportion of hourly consumption to their daily average. d Households’ hourly mean and standard deviation consumption in litres. e Boxplots illustrating the comparison between pre-lockdown and lockdown total consumption in cubic metres (m3). Value at the top of whisker is the maximum consumption; bottom of whisker is the minimum consumption; top bound of the box is the upper quartile value; bottom bound of the box is the lower quartile value; the line in the centre of the box is the median and the x in the centre of the box is the mean ((bar x)). f Weekly cluster consumption trend showing how much each of the four clusters consumes per week in m3. The error bars indicate standard deviation (σ). g Weekly number of households per cluster. The error bars indicate standard deviation (σ).

Full size image

Comparison between this study and similar data from the previous year, January–May 2019, for the same households revealed similar patterns of consumption and cluster behaviours. However, analysis of the data revealed a respective rise in PHC across the network of 13%, 22%, and 29% in March, April and May 2020 (Fig. 1a).
To examine the temporal (hourly) consumption patterns, four quartiles (Q1–Q4) were assigned to the values between the lowest and highest consumption range, revealing a consistent 24-hour pattern throughout the period irrespective of the volume of consumption (Fig. 2a–d). Households Q1 represents 1–2% (per hour) of daily consumption and occurs invariably between 00:00 and 06:00. Q2, representing 3–4% of daily consumption, occurs principally between 14:00 and 15:00 and Q3, 5–6%, occurring at different times, particularly 12:00–13:00 and 21:00. The Q4 (peak) occurs at 9:00–11:00 and 19:00–20:00. The daily mean network water demand was 27% higher during lockdown than pre-lockdown, median 43% higher and Q4 26% higher. Figure 1e presents a comparison of consumption before and during the lockdown.
Fig. 2: Clusters’ hourly consumption patterns and comparison of clusters’ share of consumption before and during the COVID-19 lockdown in the UK.

a Cumulative pattern and percentage of hourly consumption for households in the “Evening Peak (EP)” cluster. Consumption is in (m3). b Cumulative pattern and percentage of hourly consumption for households in the “Late Morning Peak Peak (LM)” cluster. Consumption is in (m3). c Cumulative pattern and percentage of hourly consumption for households in the “Early Morning Peak (EM)” cluster. Consumption is in (m3). d Cummulative pattern and percentage of hourly consumption for households in the “Multiple Peak (MP)” cluster. Consumption is in (m3). e Average daily consumption per cluster pre-lockdown. f Average daily consumption per cluster during the lockdown. e, f Value at the top of the whisker is the Maximum consumption; bottom of the whisker is the minimum consumption; top bound of the box is the upper quartile value; bottom bound of the box is the lower quartile value; the line in the centre of the box is the median and the x in the centre of the box is the mean ((bar x)). g Clusters’ share of total hourly consumption pre-lockdown. h Clusters’ share of total hourly consumption during the lockdown.

Full size image

According to our findings, households’ proportion of total hourly water demand depends upon the clusters they belong to (Fig. 2g, h), although the ratio of their hourly consumption to their daily demand is largely consistent irrespective of the time of year or volumes consumed (Fig. 2a–d).
Household cluster characterisation before and after lockdown
The results reveal four distinct clusters of household water consumers characterised by unique diurnal and night-time consumption patterns. The clusters are named Evening Peak (EP), Late Morning (LM), Early morning (EM), and multiple peak (MP).
EP
Households in EP typically use ~6% of their daily consumption between 07:00 and 08:00 but their most significant consumption occurs between 19:00 and 20:00, which invariably constitutes ~10% of their daily demand in just 1 hour. Their Q2 constitutes ~4% of their relative daily consumption per hour and occurs between 09:00 and 16:00, with Q1, occurring between 00:00 and 05:00, representing ~1–2% (Fig. 2a). During the pre-lockdown weeks, this cluster constituted ~30% of the households across the network and has been responsible for over 50% (~76 m3/hr) of the relative consumption between 19:00 and 22:00 (Fig. 2g) and ~33% (1065 m3/d) of the total daily consumption, with a mean ((bar x)) of 39 m3/hr, standard deviation (σ) of 25 m3/hr and maximum (max) of ~92 m3/hr (Fig. 2e).
During the lockdown, the percentage of households in EP dropped to a network average of 25% and their dominance of consumption between 19:00 and 22:00 decreased to an average 45%, being ~135 m3/hr (see Fig. 2h) along with their share of the total daily consumption, which also decreased to 26% (~1087 m3/d). Lockdown hourly mean, standard deviation and maximum for EP were, respectively, (bar x) 63 m3/hr, σ 44 m3/hr and max 165 m3/hr (Fig. 2f).
LM
LM describes households whose peak (Q4) occurs typically at ~10:00, representing ~11–12% of their relative daily water consumption in just 1 hour, with their next significant water use activities (~5% of daily consumption/hour) occurring at 19:00. Q2 for this cluster constitutes ~4% of their relative daily consumption per hour and occurs between 14:00 and 17:00, with a Q1 being identical to EP and EM (Fig. 2b). On average, this cluster has the highest relative consumption between 10:00 and 12:00, constituting 38% (~63 m3/hr) pre-lockdown (Fig. 2g.), was represented by ~30% of households and had a 25% (808 m3/d) share of the total daily consumption, with (bar x) 28 m3/hr, σ 19 m3/hr and max 76 m3/hr (Fig. 2e).
The percentage of households in LM increased to an average of 37% across the network during the lockdown weeks but their consumption between 10:00 and 12:00 remained at an average of 38%–~74 m3/hr (see Fig. 2h). Their share of the total daily consumption increased to ~31% (~1281 m3/d). Lockdown being respectively (bar x) = 51 m3/hr, σ = 36 m3/hr and max = of 134 m3/hr (Fig. 2f).
EM
Households in EM have the fewest instances of peaks which constitute ~12–13% in 24 hours and occurs between 07:00 and 08:00. Q3 for this cluster, ~7% of their relative daily consumption, occurs at 19:00, Q2 between 10:00 and 17:00, constituting ~3–4% of their relative daily consumption per hour, and Q1 identical to EP and LM (Fig. 2c). On average, this cluster, made up of 26% of household, was responsible for 40% (~59 m3/hr) of pre-lockdown consumption occurring between 07:00 and 08:00 (Fig. 2g.) and 22% (723 m3/d) of the total daily consumption, with (bar x) of 26 m3/hr, σ of 17 m3/hr and max of 73 m3/hr (Fig. 2e).
EM experienced the sharpest decrease in the number of households during the lockdown period—an average of 12% across the network, resulting in a significant drop in their share of relative consumption between 07:00 and 08:00 to from 40% to 20%–~38 m3/hr (see Fig. 2h). Their share of the total daily consumption also fell to 12% (433 m3/d). Lockdown being, respectively, (bar x) 17 m3/hr, σ 10 m3/hr and max of 40 m3/hr (Fig. 2f).
MP
MP has the highest instances of Q4s within 24 hours (about seven instances of 6–7% of their relative daily consumption). They also have multiple instances of Q3s and Q2s at 5% and 4% of relative daily consumption, respectively. Their Q1, like the other clusters, resides between 00:00 and 06:00, constituting ~2–3% of relative daily consumption (Fig. 2d). During the pre-lockdown weeks, this cluster represented 14% of the households across the network. MP dominates consumption between 00:00 05:00—at an average of 32% (~8 m3/hr) (Fig. 2g) and about 20% (661 m3/d) of the total daily consumption, with (bar x) 19 m3/hr, σ 8 m3/hr and max 29 m3/hr (Fig. 2e).
MP experienced the most significant increase in the number of households during the lockdown period—a 93% increase between M3 and M4 maintaining an average of 26% of all households during the lockdown period. This has resulted in an increase in their share of hourly consumption between 00:00 and 07:00 to an average of 45%–~24 m3/hr; between 12:00 and 17:00 to an average of 39%–~98 m3/hr and 23:00 to 36%–~60 m3/hr (see Fig. 2h). Their share of the total daily consumption also rose to ~32% (~1326 m3/d). Lockdown being, respectively, (bar x) 67 m3/hr, σ 34 m3/hr and max of 110 m3/hr (Fig. 2f).
In another study,14 segmentation was based on heterogeneous micro-component consumption patterns and behaviour regularities and temporal characteristics. This work, unlike our study, performed a disaggregation of sub-minute smart meter data into end-use events, subsequently clustering households based on their end-use similarities. One difference of this study to the one here reported, however, is the consumer household sample size. In our study, some 11,528 households were assessed, and currently, sub-minute smart meters are unavailable across such a large region. Our segmentation was derived from normalised hourly smart meter data, being based on temporal patterns of consumption. The silhouette coefficient value, when ‘t-distributed stochastic neighbour embedding’ (t-SNE)15 was used for dimensionality reduction (as opposed to PCA), improved slightly from 3.9 to 4.1 for n_cluster = 4. However, this improvement only marginally enhanced the k-means results (by  More

1.
UNICEF and World Health Organization. Progress on household drinking water, sanitation and hygiene 2000-2017; Special focus on inequalities. (Joint Monitoring Programme, New York, 2019).
2.
IPCC. Climate Change 2014: Impacts, Adaptations, and Vulnerability; Part A: Global and Sectoral Aspects; Summary for Policymakers. 1-32, https://doi.org/10.1017/CBO9781107415379.003. (Cambridge University Press, Cambridge, UK and New York, 2014).

3.
Hoque, S. F. & Hope, R. The water diary method – proof-of-concept and policy implications for monitoring water use behaviour in rural Kenya. Water Policy 20, 725–743 (2018).
Article  Google Scholar 

4.
Thomson, P. et al. Rainfall and groundwater use in rural Kenya. Sci. Total Environ. 649, 722–730 (2018).
Article  Google Scholar 

5.
Thomas, E. et al. Quantifying increased groundwater demand from prolonged drought in the East African Rift Valley. Sci. Total Environ. 666, 1265–1272 (2019).
CAS  Article  Google Scholar 

6.
Foster, T. & Hope, R. A multi-decadal and social-ecological systems analysis of community waterpoint payment behaviours in rural Kenya. J. Rural Stud. 47, 85–96 (2016).
Article  Google Scholar 

7.
Foster, T. & Hope, R. Evaluating waterpoint sustainability and access implications of revenue collection approaches in rural Kenya. Water Resour. Res. 53, 1473–1490 (2017).
Article  Google Scholar 

8.
Global WASH Cluster, Sanitaiton and Water for All, UNICEF, and ICRC. COVID-19 and WASH: Mitigating the socio-economic impacts on the Water, Sanitation and Hygiene (WASH) Sector. (2020).

9.
Shannon, R., Erhardt, D. & Kolker, J. Considerations for Financial Facilities to Support Water Utilities in the COVID-19 Crisis (World Bank, Washington, DC, 2020).

10.
Diffenbaugh, N. S., Giorgi, F., Raymond, L. & Bi, X. Indicators of 21st century socioclimatic exposure. Proc. Natl Acad. Sci. USA 104, 20195–20198 (2007).
CAS  Article  Google Scholar 

11.
Choularton, R. J. & Krishnamurthy, P. K. How accurate is food security early warning? Evaluation of FEWS NET accuracy in Ethiopia. Food Security 11, 333–344 (2019).
Article  Google Scholar 

12.
Boluwade, A. Remote sensed-based rainfall estimations over the East and West Africa regions for disaster risk management. ISPRS J. Photogramm. Remote Sens. 167, 305–320 (2020).
Article  Google Scholar 

13.
Manandhar, A., Greeff, H., Thomson, P., Hope, R. & Clifton, D. A. Shallow aquifer monitoring using handpump vibration data. J. Hydrol. X 8, 100057 (2020).
Article  Google Scholar 

14.
Thomas, E. et al. Reducing drought emergencies in the Horn of Africa. Sci. Total Environ. 727, 138772 (2020).
CAS  Article  Google Scholar 

15.
World Health Organization. Coronavirus disease 2019 (COVID-19) Situation Report 42. (WHO, Geneva, 2020).

16.
World Health Organization. Coronavirus disease 2019 (COVID-19) Situation Report 71. (WHO, Geneva, 2020).

17.
McNicholl, D. et al. Performance-Based Funding for Reliable Rural Water Services in Africa. (Uptime consortium, 2019).

18.
Colman, A. W., Graham, R. J. & Davey, M. K. Direct and indirect seasonal rainfall forecasts for East Africa using global dynamical models. Int. J. Climatol. 40, 1132–1148 (2020).
Article  Google Scholar 

19.
McNicholl, D. et al. Results-Based Contracts for Rural Water Services. (Uptime consortium, 2020).

20.
Funk, C. C. et al. A quasi-global precipitation time series for drought monitoring (USGS, Reston, VA, 2014). More

Total water consumption in the USA in 2015 was 1218 billion litres per day, of which thermoelectric power used 503 billion litres, irrigation used 446 billion litres and 147 billion litres per day went to supply 87% of the US population with potable water13.
Data centres consume water across two main categories: indirectly through electricity generation (traditionally thermoelectric power) and directly through cooling. In 2014, a total of 626 billion litres of water use was attributable to US data centres4. This is a small proportion in the context of such high national figures, however, data centres compete with other users for access to local resources. A medium-sized data centre (15 megawatts (MW)) uses as much water as three average-sized hospitals, or more than two 18-hole golf courses14. Some progress has been made with using recycled and non-potable water, but from the limited figures available15 some data centre operators are drawing more than half of their water from potable sources (Fig. 2). This has been the source of considerable controversy in areas of water stress and highlights the importance of understanding how data centres use water.
Fig. 2: Water source by year for Digital Realty, a large global data centre operator.

Consumption from potable water was 64% (2017), 65% (2018) and 57% (2019)15.

Full size image

This section considers these two categories of data centre water consumption.
Water use in electricity generation
Water requirements are measured based on withdrawal or consumption. Consumption refers to water lost (usually through evaporation), whereas water withdrawal refers to water taken from a source such as natural surface water, underground water, reclaimed water or treated potable water, and then later returned to the source16.
Power plants generate heat using fossil fuels such as coal and gas, or nuclear fission, to convert water into steam which rotates a turbine, thereby generating electricity. Water is a key part of this process, which involves pre-treating the source water to remove corroding contaminants, and post-treatment to remove brines. Once heated into steam to rotate the turbine, water is lost through evaporation, discharged as effluent or recirculated; sometimes all three16.
The US average water intensity for electricity generation for 2015 was 2.18 litres per kilowatt hour (L/kWh)17, but fuel and generator technology type have a major impact on cooling water requirements. For example, a dry air cooling system for a natural gas combined cycle generator consumes and withdraws 0.00–0.02 L/kWh, whereas a wet cooling (open recirculating) system for a coal steam turbine consumes 0.53 L/kWh and withdraws 132.5 L/kWh. Efficiency varies significantly, with consumption ranging from 0.00 to 4.4 L/kWh and withdrawal ranging from 0.31 to 533.7 L/kWh depending on the system characteristics16.
Hydropower systems also use large volumes of water despite being considered a cleaner source of electricity. Water evaporation from open reservoirs is a major source of losses, particularly in dry regions and where water is not pumped back into the reservoir or passed onto downstream users. The US national average water consumption for hydropower is 16.8 L/kWh compared to 1.25 L/kWh for thermoelectricity17.
With the majority of generation still from fossil fuels18, the transition to renewables is important for both carbon and water intensity. Only solar and wind energy do not involve water in generation, yet both still consume water in the manufacturing and construction processes9. Estimates suggest that by 2030, moving to wind and solar energy could reduce water withdrawals by 50% in the UK, 25% in the USA, Germany and Australia and 10% in India19.
In the data centre sector, Google and Microsoft are leading the shift to renewables. Between 2010 and 2018, the number of servers increased 6 times, network traffic increased 10 times and storage capacity increased by 25 times, yet energy consumption has only grown by 6%6. A major contributor to this has been the migration to cloud computing, as of 2020 estimated to be a $236 billion market20 and responsible for managing 40% of servers4.
Due to their size, the cloud providers have been able to invest in highly efficient operations. Although often criticised as a metric of efficiency21, an indicator of this can be seen through low power usage effectiveness (PUE) ratios. PUE is a measure of how much of energy input is used by the ICT equipment as opposed to the data centre infrastructure such as cooling22, defined as follows:

$${rm{PUE}}=frac{{rm{Data}} {rm{Centre}} {rm{Total}} {rm{Energy}} {rm{Consumption}}}{{rm{ICT}} {rm{Equipment}} {rm{Energy}} {rm{Consumption}}}$$
(1)

PUE is relevant to understanding indirect water consumption because it indicates how efficient a particular facility is at its primary purpose: operating ICT equipment. This includes servers, networking and storage devices. An ideal PUE of 1.0 would mean 100% of the energy going to power useful services running on the ICT equipment rather than wasted on cooling, lighting and power distribution. Water is consumed indirectly through the power generation, so more efficient use of that power means more efficient use of water.
Traditional data centres have reported PUEs reducing from 2.23 in 2010 to 1.93 in 20206. In contrast, the largest “hyperscale” cloud providers report PUEs ranging from 1.25 to 1.18. Some report even better performance, such as Google with a Q2 2020 fleet wide trailing 12-month PUE of 1.1023.
As data centre efficiency reaches such levels, further gains are more difficult. This has already started to show up in plateauing PUE numbers24, which means the expected increase in future usage may soon be unable to be offset by efficiency improvements25. As more equipment is deployed, and more data centres are needed to house that equipment, energy demand will increase. If that energy is not sourced from renewables, indirect water consumption will increase.
Power generation source is therefore a key element in understanding data centre water consumption, with PUE an indicator of how efficiently that power is used, but it is just the first category. Direct water use is also important—all that equipment needs cooling, which in some older facilities can consume up to 30% of total data centre energy demand26,27,28.
Water use in data centre cooling
ICT equipment generates heat and so most devices must have a mechanism to manage their temperature. Drawing cool air over hot metal transfers heat energy to that air, which is then pushed out into the environment. This works because the computer temperature is usually higher than the surrounding air.
The same process occurs in data centres, just at a larger scale. ICT equipment is located within a room or hall, heat is ejected from the equipment via an exhaust and that air is then extracted, cooled and recirculated. Data centre rooms are designed to operate within temperature ranges of 20–22 °C, with a lower bound of 12 °C29. As temperatures increase, equipment failure rates also increase, although not necessarily linearly30.
There are several different mechanisms for data centre cooling27,28, but the general approach involves chillers reducing air temperature by cooling water—typically to 7–10 °C31—which is then used as a heat transfer mechanism. Some data centres use cooling towers where external air travels across a wet media so the water evaporates. Fans expel the hot, wet air and the cooled water is recirculated32. Other data centres use adiabatic economisers where water sprayed directly into the air flow, or onto a heat exchange surface, cools the air entering the data centre33. With both techniques, the evaporation results in water loss. A small 1 MW data centre using one of these types of traditional cooling can use around 25.5 million litres of water per year32.
Cooling the water is the main source of energy consumption. Raising the chiller water temperature from the usual 7–10 °C to 18–20 °C can reduce expenses by 40% due to the reduced temperature difference between the water and the air. Costs depend on the seasonal ambient temperature of the data centre location. In cooler regions, less cooling is required and instead free air cooling can draw in cold air from the external environment31. This also means smaller chillers can be used, reducing capital expenditure by up to 30%31. Both Google34 and Microsoft35 have built data centres without chillers, but this is difficult in hot regions36. More

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript. More

1.
Matthews, G. V. T. The Ramsar Convention on Wetlands: Its History and Development (Ramsar Convention Bureau, 1993).
2.
Ramsar Convention on Wetlands Global Wetland Outlook: State of the World’s Wetlands and Their Services to People (Ramsar Convention Secretariat, 2018).

3.
Díaz, S. et al. Summary for Policymakers of the Global Assessment Report on Biodiversity and Ecosystem Services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES, 2019).

4.
Bowman, M. in Yearbook of International Co-operation on Environment and Development (eds Stokke, O. S. & Thommessen, Ø. B.) 61–68 (Earthscan, 2002).

5.
Morrison, T. H. et al. Nat. Sustain. 3, 947–955 (2020).
Article  Google Scholar 

6.
Jones, T. & Pritchard, D. Comprehensive Review and Analysis of Ramsar Advisory Mission (RAM) Reports (Ramsar Convention Secretariat, 2018).

7.
Bridgewater, P., Kim, R. E. & Bosselmann, K. Yearbook Int. Environ. Law 25, 61–78 (2014).
Article  Google Scholar 

8.
Millennium Ecosystem Assessment Ecosystems and Human Well-Being: Wetlands and Water – Synthesis (World Resources Institute, 2005).

9.
Green, A. J. et al. Front. Ecol. Environ. 15, 99–107 (2017).
Article  Google Scholar 

10.
Cumming, G. S. Anthropocene 13, 46–56 (2016).
Article  Google Scholar 

11.
Mauerhofer, V., Kim, R. E. & Stevens, C. Environ. Sci. Policy 51, 95–105 (2015).
Article  Google Scholar 

12.
Tanzi, A. et al. (eds) The UNECE Convention on the Protection and Use of Transboundary Watercourses and International Lakes (Brill Nijhoff, 2015)

13.
Loures, F. & Rieu-Clarke, A. (eds) The UN Watercourses Convention in Force (Routledge, 2013).

14.
Global Biodiversity Outlook 5 (Secretariat of the Convention on Biological Diversity, 2020).

15.
Heger, T. et al. BioScience 69, 888–899 (2019).
Article  Google Scholar 

16.
Mcleod, E. et al. Front. Ecol. Environ. 9, 552–560 (2011).
Article  Google Scholar 

17.
Hettiarachchi, M., Morrison, T. H. & McAlpine, C. Glob. Environ. Change 32, 57–66 (2015).
Article  Google Scholar  More