Markus Andrews

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

1.
Hanjra, M. A., Ferede, T. & Gutta, D. G. Reducing poverty in sub-Saharan Africa through investments in water and other priorities. Agric. Water Manag. 96, 1062–1070 (2009).
Article  Google Scholar 
2.
Moris, J. R. Irrigation Development in Africa: Lessons of Experience (Routledge, 1990).

3.
Aw, D. & Diemer, G. Making a Large Irrigation Scheme Work: A Case Study from Mali (World Bank, 2005).

4.
Bertoncin, M., Pase, A., Quatrida, D. & Turrini, S. At the junction between state, nature and capital: irrigation mega-projects in Sudan. Geoforum 106, 24–37 (2019).
Article  Google Scholar 

5.
Adams, W. M. Wasting the Rain: Rivers, People and Planning in Africa (Earthscan, 1992).

6.
Chambers, R. & Moris J. R. (eds) Mwea: An Irrigated Rice Settlement in Kenya (Weltforum Verlag, 1973).

7.
Awojobi, O. & Jenkins, G. P. Were the hydro dams financed by the World Bank from 1976 to 2005 worthwhile? Energy Policy 86, 222–232 (2015).
Article  Google Scholar 

8.
Bazin, F., Hathie, I., Skinner, J. & Koundouno, J. Irrigation, Food Security and Poverty—Lessons from Three Large Dams in West Africa (IIED and IUCN, 2017).

9.
World Commission on Dams Dams and Development: A New Framework for Decision-Making (Earthscan, 2000).

10.
Thomas, D. H. L. & Adams, W. M. Adapting to dams: agrarian change downstream of the Tiga dam, northern Nigeria. World Dev. 27, 919–935 (1999).
Article  Google Scholar 

11.
Balasubramanian, V., Sie, M., Hijmans, R. J. & Otsuka, K. Increasing rice production in sub-Saharan Africa: challenges and opportunities. Adv. Agron. 94, 55–133 (2007).
CAS  Article  Google Scholar 

12.
Alam, M. Problems and potential of irrigated agriculture in sub-Saharan Africa. J. Irrig. Drain. Eng. 117, 155–172 (1991).
Article  Google Scholar 

13.
Mold, A. Will it all end in tears? Infrastructure spending and African development in historical perspective. J. Int. Dev. 24, 237–254 (2012).
Article  Google Scholar 

14.
Veldwisch, G. J., Bolding, A. & Wester, P. Sand in the engine: the travails of an irrigated rice scheme in Bwanje Valley, Malawi. J. Dev. Stud. 45, 197–226 (2009).
Article  Google Scholar 

15.
Carney, J. Converting the wetlands, engendering the environment: the intersection of gender with agrarian change in the Gambia. Econ. Geogr. 69, 329–348 (1993).
Article  Google Scholar 

16.
Zarfl, C., Lumsdon, A. E., Berlekamp, J., Tydecks, L. & Tockner, K. A global boom in hydropower dam construction. Aquat. Sci. 77, 161–170 (2015).
Article  Google Scholar 

17.
Woodhouse, P. et al. African farmer-led irrigation development: re-framing agricultural policy and investment? J. Peasant Stud. 44, 213–233 (2017).
Article  Google Scholar 

18.
Dakar Declaration on Irrigation: Building Resilience and Accelerate Growth in Sahel and West Africa by Boosting Irrigated Agriculture (ICID, 2013).

19.
Merrey, D. J. & Sally, H. Another well-intentioned bad investment in irrigation: the Millennium Challenge Corporation’s ‘compact’ with the Republic of Niger. Water Altern. 10, 195–203 (2017).
Google Scholar 

20.
Rufin, P. et al. Global-scale patterns and determinants of cropping frequency in irrigation dam command areas. Glob. Environ. Change 50, 110–122 (2018).
Article  Google Scholar 

21.
Blanc, E. & Strobl, E. Is small better? A comparison of the effect of large and small dams on cropland productivity in South Africa. World Bank Econ. Rev. 28, 545–576 (2014).
Article  Google Scholar 

22.
Flyvbjerg, B. Policy and planning for large-infrastructure projects: problems, causes, cures. Environ. Plan. B 34, 578–597 (2007).
Article  Google Scholar 

23.
Ansar, A., Flyvbjerg, B., Budzier, A. & Lunn, D. Should we build more large dams? The actual costs of hydropower megaproject development. Energy Policy 69, 43–56 (2014).
Article  Google Scholar 

24.
Ika, L. A., Diallo, A. & Thuillier, D. Critical success factors for World Bank projects: an empirical investigation. Int. J. Proj. Manag. 30, 105–116 (2012).
Article  Google Scholar 

25.
Duponchel, M., Chauvet, L. & Collier, P. What Explains Aid Project Success in Post-Conflict Situations? Policy Research Working Paper Series 5418 (World Bank, 2010).

26.
Flyvbjerg, B. Survival of the unfittest: why the worst infrastructure gets built—and what we can do about it. Oxf. Rev. Econ. Policy 25, 344–367 (2009).
Article  Google Scholar 

27.
Scott, J. C. Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed (Yale Univ. Press, 1998).

28.
Li, T. M. Beyond ‘the state’ and failed schemes. Am. Anthropol. 107, 383–394 (2005).
Article  Google Scholar 

29.
de Bont, C., Komakech, H. C. & Veldwisch, G. J. Neither modern nor traditional: farmer-led irrigation development in Kilimanjaro region, Tanzania. World Dev. 116, 15–27 (2019).
Article  Google Scholar 

30.
Li, T. M. The Will to Improve: Governmentality, Development, and the Practice of Politics (Duke Univ. Press, 2007).

31.
Project Performance Audit Report—Chad Lake Chad Polders Project (Credit 592-CD) Report No. 6751 (World Bank, 1987).

32.
Chad: Appraisal of Sategui-Deressia Irrigation Project Report No. 145a-CD (World Bank, 1974).

33.
Adams, W. M. Large scale irrigation in northern Nigeria: performance and ideology. Trans. Inst. Br. Geogr. 16, 287–300 (1991).
Article  Google Scholar 

34.
Biswas, A. K. Irrigation in Africa. Land Use Policy 3, 269–285 (1986).
Article  Google Scholar 

35.
Ansar, A., Flyvbjerg, B., Budzier, A. & Lunn, D. in The Oxford Handbook of Megaproject Management (ed. Flyvbjerg, B.) Ch. 4 (Oxford Univ. Press, 2017).

36.
Adams, W. M. The downstream impacts of dam construction: a case study from Nigeria. Trans Inst. Br. Geogr. 10, 292–302 (1985).
Article  Google Scholar 

37.
Hirschman, A. O. Development Projects Observed (Brookings, 1967).

38.
de Bont, C. The continuous quest for control by African irrigation planners in the face of farmer-led irrigation development: the case of the Lower Moshi area, Tanzania (1935–2017). Water Altern. 11, 893–915 (2018).
Google Scholar 

39.
Bertoncin, M. & Pase, A. Interpreting mega-development projects as territorial traps: the case of irrigation schemes on the shores of Lake Chad (Borno State, Nigeria). Geogr. Helv. 72, 243–254 (2017).
Article  Google Scholar 

40.
Schumacher, E. F. Small Is Beautiful: A Study of Economics as If People Mattered (Vintage, 1973).

41.
Jones, B. Desiccation and the West African colonies. Geogr. J. 91, 401–423 (1938).
Article  Google Scholar 

42.
Adams, W. M. How beautiful is small? Scale, control and success in Kenyan irrigation. World Dev. 18, 1309–1323 (1990).
Article  Google Scholar 

43.
Ahlers, R., Brandimarte, L., Kleemans, I. & Sadat, S. H. Ambitious development on fragile foundations: criticalities of current large dam construction in Afghanistan. Geoforum 54, 49–58 (2014).
Article  Google Scholar 

44.
Green, N., Sovacool, B. K. & Hancock, K. Grand designs: assessing the African energy security implications of the Grand Inga dam. Afr. Stud. Rev. 58, 133–158 (2015).
Article  Google Scholar 

45.
Mbara, C. J., Gadain, H. M. & Muthusi, F. M. Status of Medium to Large Irrigation Schemes in Southern Somalia Technical Report No. W-05 (FAO-SWALIM, 2007).

46.
Flyvbjerg, B. Policy and planning for large-infrastructure projects: problems, causes, cures. Environ. Plan. B 34, 578–597 (2007).
Article  Google Scholar 

47.
Burney, J., Woltering, L., Burke, M., Naylor, R. & Pasternak, D. Solar-powered drip irrigation enhances food security in the Sudano–Sahel. Proc. Natl Acad. Sci. USA 107, 1848–1853 (2010).
CAS  Article  Google Scholar 

48.
Higginbottom, T. P., Symeonakis, E., Meyer, H. & van der Linden, S. Mapping fractional woody cover in semi-arid savannahs using multi-seasonal composites from Landsat data. ISPRS J. Photogramm. Remote Sens. 139, 88–102 (2018).
Article  Google Scholar 

49.
Müller, H., Rufin, P., Griffiths, P., Siqueira, A. J. & Hostert, P. Mining dense Landsat time series for separating cropland and pasture in a heterogeneous Brazilian savanna landscape. Remote Sens. Environ. 156, 490–499 (2015).
Article  Google Scholar 

50.
Pal, M. Random forest classifier for remote sensing classification. Int. J. Remote Sens. 26, 217–222 (2005).
Article  Google Scholar 

51.
Gorelick, N. et al. Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 202, 18–27 (2017).
Article  Google Scholar 

52.
Abatzoglou, J. T., Dobrowski, S. Z., Parks, S. A. & Hegewisch, K. C. Terraclimate, a high-resolution global dataset of monthly climate and climatic water balance from 1958–2015. Sci. Data 5, 170191 (2018).
Article  Google Scholar 

53.
Weiss, D. J. et al. A global map of travel time to cities to assess inequalities in accessibility in 2015. Nature 553, 333–336 (2018).
CAS  Article  Google Scholar 

54.
Lloyd, C. T., Sorichetta, A. & Tatem, A. J. High resolution global gridded data for use in population studies. Sci. Data 4, 170001 (2017).
Article  Google Scholar 

55.
Kaufmann, D., Kraay, A. & Mastruzzi, M. The worldwide governance indicators: methodology and analytical issues. Hague J. Rule Law 3, 220–246 (2011).
Article  Google Scholar 

56.
Walsh, R. P. D. & Lawler, D. M. Rainfall seasonality: description, spatial patterns and change through time. Weather 36, 201–208 (1981).
Article  Google Scholar 

57.
Wood, S. N. Generalized Additive Models: an Introduction with R 2nd edn (Chapman and Hall/CRC, 2017).

58.
Wood, S. N. Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. J. R. Stat. Soc. B 73, 3–36 (2011).
Article  Google Scholar  More