Apr 28, 2019 Water use in electricity generation for water-energy nexus analyses:

• Electricity plant water withdrawals are estimatedusingacomprehensiveanaly-sis. • Estimates resemble reported levels on spatio-temporal scales of country/year. • The results support perspectives in larger scale water-energy nexus man- agement. • More open source, freely available and detailed data are however urged. With almost 40% of the global population suffering from water scarcity, the need to manage water resources is evidently urgent. While water and energy systems are intrinsically linked, the availability of comprehensive, integrated data sets across the domains of water and energy is generally lacking. As a result, estimated indicators representing volumes of water usage per unit of electricity or fuel produced are often required to analyse the water-energy nexus. In this paper, an “ ensemble ” of indicators is assembled representing water usage spanning different electricity-generation technologies based on previously published works in an attempt to depict the level or lack of detail in current large-scale energy-sector water-usage data. Based on these, the degree in which using such estimates is suitable for reproducing electricity-production water-usage at coarser spatio-temporal scales is assessed. The performance of the ensemble median/min/max as a predictor of water use is evaluated for the period from 1980 to 2015 using additional information about the constituents of the European energy system. Comparing with the reported values for 1980 – 2015, the median provides a skillful re- production of historical yearly water use for the EU (EU28) as a whole. A further analysis for 2015 indicates that reasonable agreement is also seen at the country level. Thus, the results suggest that an “ ensemble-based ap- proach ” has the potential to provide sturdy estimates of yearly water use by energy systems for analyses

Water and energy systems are inextricably interdependent. The water sector is a major consumer of energy for purposes such as water treatment, pumping and desalination. Similarly, water is essential for cooling power plants, electricity generation and bio-fuel production, as well as in the extraction, mining, processing, refining and disposal of fossil-fuel residues. 44% of total global water withdrawals are used for energy production, a dominant share of which is cooling water in thermoelectric electricity generation (Collins et al., 2009). Energy and water are both limited resources that are essential for the fundamental services, including food production, required by a rapidly growing global population that is projected to reach 9.7 billion in 2050 (United Nations, 2017). As a result, it is increasingly critical to manage the nexus between energy and water properly (Kurian, 2017) in the broader context of dependent socio-economic sectors, including the wider water-energy-food nexus (Griggs et al., 2013;Howells and Rogner, 2014;United Nations General Assembly, 2015). Moreover, proper water-energy management is especially crucial in light of the fact that electricity and fuel production relies on an estimated 90% of non-sustainable water sources (WWAP, 2014), as well as the increasing demands for water, energy and food driven by, among others, the growth in population and economies (Hoekstra et al., 2012).
Over the last decade or so, analyzing issues within or related to the water-energy nexus has become increasingly important for both the scientific and policy-making communities (Dai et al., 2018;Miralles-Wilhelm, 2016). Likewise, the capacity to assess water and energy interlinkages at an increasingly higher resolution has also improved accordingly. Analyses of the water-energy nexus span a broad range of spatial levels, from the local (e.g. plant or city) (Chen and Chen, 2016) to the regional or national (Kibaroglu and Gürsoy, 2015;Mayor et al., 2015). Meanwhile temporalities range from multi-decadal (including climate change) (Mekonnen et al., 2016;Voisin et al., 2013) down to days or hours (or even lower) for operational applications (Castronuovo and Lopes, 2004).
A comprehensive review of methods and tools for macroassessments of the water-energy nexus has recently been carried out by (Dai et al., 2018). From this analysis it is evident that, while a wide range of new methods and frameworks for comprehensively assessing interactions between water, energy and other elements have been developed, in general the availability of tools for nexus analyses that are at the same time integrative and multi-level is still poor (Daher and Mohtar, 2015;Howells et al., 2013). Instead, methodologies used for analyzing the water-energy nexus tend to be characterized by specific levels and data requirements (Liu et al., 2017), ranging from purely qualitative assessments to highly data-intensive model-based approaches (Granit et al., 2013). The review also found that none of the studies and methods considered provide a 'singular framework' for performing nexus studies.
The challenges of data availability at relevant spatio-temporal levels for analyzing the water-energy nexus, for example, on water use by energy systems and vice versa, is well documented (Chini and Stillwell, 2017;IRENA, 2015;Larsen et al., 2016). While in general water and energy systems can be considered to be well-monitored and managed (developing countries excluded), the availability of integrated data sets covering both domains is often severely limited at the relevant levels of aggregation in relation to nexus calculations, that is, beyond the site-specific level. Further, such data may be incomplete and inconsistent due, for example, to differences in the inherent conditions for the collection of data on water use by the energy sector between countries and regions, which can constrain the applicability and comparability of estimated water uses. For example, records from the US, while otherwise of good quality, have significant gaps concerning water-intensive energy technologies like nuclear (Macknick et al., 2012). Conversely, dependencies between water and energy systems, that is, water consumption or withdrawals related to specific energy technologies, may be expressed in terms of representative volumes of water use per unit (e.g. L/MWh) of electricity or fuel produced (Basheer and Elagib, 2018;Gleick, 1994;Inhaber, 2004;Macknick et al., 2012). This approach introduces a significant source of uncertainty arising from the (lack of) accuracy, but it also enables quantitative nexus calculations to be made at different levels and is frequently used by integrated assessment models.
In light of the poor data on water usage within the energy sector, as highlighted above, this paper addresses the extent to which reported estimates of water usage in electricity production provide an accurate 'bridge' when modelling the interdependencies between water and energy systems. Thus, many initiatives, like the Platform for Regional Integrated Modelling and Analysis (PRIMA) (Kraucunas et al., 2015), as well as the ETSAP-TIAM community (Føyn et al., 2011), aim at developing flexible multi-scale tools for analyzing the water-energy nexus in order to satisfy users' increasing demands by linking existing model components with new ones that use such an approach. In this context, the present study may be seen as an attempt to identify and validate a suitable set of parameters. To estimate water usage, multiple literature estimates of water withdrawal and consumption rates for electricity production technologies are collected in conjunction with the distribution of individual power plants and their corresponding technologies in order to calculate the country-level EU28 yearly water usages for 1980-2015, followed by a validation against reported numbers (Eurostat, 2018). The analysis is relevant because it highlights the best possible estimates of water usage within electricity production at coarser scales using freely available sources, albeit at coarse spatiotemporal resolutions (country/yearly). Thus, despite a certain resemblance between estimates and reported values, the paper also aims to show that the currently available data on energy-sector water usage is very inadequate, not least, in their detail and availability. Despite the current focus on providing open-access environmental data of increasing quality, data on the water-energy nexus are still limited in their availability. One aim of this paper is therefore to convey this information to users mainly in the academic community but also to politicians in light of the current tendency towards more open and available data.

Data and methodology
In this study, withdrawals of water are defined as the total amount of water that is extracted or diverted from its groundwater or surface water source and used during electricity-generation operations (as opposed to, e.g., including the construction phase), including the return flow. Thus, the cooling water addressed in this paper is freshwater only. Water consumption is similarly defined as the net balance, including only evaporated and transpired water, as well as water stored in crops and/or other products. Both terms (withdrawals and consumption) are jointly referred to as 'water usage'. Using this definition water consumption becomes a subset of water withdrawals.
The term 'median', as used in this study, is the most commonly used term in the recent literature (Davies et al., 2013;Macknick et al., 2012). It can therefore to some extent be regarded as the standard metric. However, some literature uses the term 'average' (Mielke et al., 2010;NETL, 2010), whereas other, typically older literature sources simply give a representative value (Gleick, 1994;Inhaber, 2004). Furthermore, many studies build upon each other by re-issuing the findings of older studies. However, for any literature sources where the median estimate is based the middle value in between the reported minimum and maximum spans, this may introduce a certain bias towards underestimation, as argued by (Macknick et al., 2012). The span of the entire range is addressed by employing the minimum and maximum estimates. Correspondingly, the estimated mean and min/max ranges cannot be asserted to be 'robust' from a strictly statistical perspective (e.g., as quantified by t-procedures).

Data
The data used in this study can be grouped into three main categories depending on their nature and how they are used. I) The first category covers data on the water withdrawal and consumption rates of electricity production as a function of the energy source and cooling Table 1 Literature sources for the estimates of water usage per unit of energy produced, an indication of the available information per relevant energy source, and the number of total occurrences in the literature of the median/minimum/maximum levels for withdrawals and consumption. The literature is sorted chronologically within the categories of peer-reviewed and grey literature respectively. 'CSP' denotes concentrated solar power, and PV denotes photovoltaic solar technology.

Energy
source/Reference (Gleick, 1994) ( Coal technology in question. The sources for these data are listed in Table 1. II) The second category covers data on individual electricitygeneration plants which are used to estimate country-specific electricity generation per energy source, cooling technology and water source (fresh/saline). These data derive from (Enipedia, 2018;Global Energy Observatory, 2018;Shift Project Data Portal, 2018). These databases aim to include up-to-date information on individual power plants, although the present study necessitated substantial further online research to confirm whether or not, for example, plants were active or the type of cooling technology. III) The third category covers actual reported surface and groundwater withdrawals (both freshwater) used for cooling electricity generation plants, data being derived from (Eurostat, 2018). Annual data for the EU are available from 1970 to 2015. However, for earlier years there are an increasing number of data gaps on reported freshwater withdrawals. 1980 is therefore used as a starting point. For data gaps which are still present after 1980,linear interpolation has been used to fill the gaps, since this is the most conservative method and since a method aiming to reproduce patterns from neighbouring countries was considered too arbitrary for the short periods which needed gaps to be filled (a few years).

Estimated water use by energy source and technology
The estimated water usage related to specific energy technologies found, for example, in the scientific literature exhibit great local variations across natural, geographical, technological, hydro-climatic conditions and differences in definitions. Table 1 gives a list of previous studies within the categories of both peer-reviewed and 'grey' (report-style) literature as identified by the authors, who have estimated water usage by different energy technologies. While the list is comprehensive, it is by no means complete. Figs. 1 and 2 give all individual withdrawal and consumption factor estimates (median, min and max levels) respectively from the listed studies (markers) as well as their average value (bars) and Appendices A and B give the corresponding median values in table form. In general, the levels were found to be reasonably similar across literature sources, particularly for the median values, which also form the basis of the present study. The definitions in depicting water usage vary across literature, which is often not elaborated but instead visible from the numbers themselves or from the categories used. As an example, some sources group water usages across technologies, making the ranges appear exceptionally broad. All such cases are omitted in order to reflect only actual combinations of energy source and cooling technology. Another example includes pond cooling where the levels vary substantially which is believed to be related to definitions (using either the total withdrawn amount or only make-up water). In the present study, this is accounted for by dividing these definitions into two separate shares. It is further highlighted that data points in Figs. 1 and 2 may appear directly on top of each other having the same or corresponding values.

Reconstruction of water use by energy plants in Europe
To evaluate the performance of water-usage levels derived from the literature review, the water used by European energy systems is estimated. To do this, information on the share of different (a) energy sources and (b) cooling technologies (national), (c) the source of the water body used in nexus calculations (i.e. sea-or freshwater), principal cooling technologies related to energy source and country, and the electricity generated for each of the EU28 countries is collected. Only open-source data were used (see below). Based on this information, the amount of electricity plant cooling water is estimated. Hereafter this estimate of electricity plant cooling water will be referred to as EW ('estimated withdrawals'). Finally, the results are compared with the actual reported freshwater withdrawals available through Eurostat.  Hereafter, this dataset will be referred to as RW ('reported withdrawals'). As a part of the same literature review, the electricity plant cooling water consumption share (i.e. the net share excluding the return flow as defined above) is also estimated. Hereafter, this estimate is referred to as EC ('estimated consumption'). However, EC could not be validated against the reported levels, as data on reported consumption quantities could not be found (hence, the abbreviation 'RC' is not used here).
Estimates of electricity plant cooling water are divided into two separate analyses.
EW1980-2015: This analysis focuses on the temporal evolution of water use in electricity plants. It employs country-specific estimates of energy sources and cooling technologies, whereas data on the freshwater/sea-water shares are largely dependent on the plantspecific information, which is not generally available historically. The corresponding reported withdrawals are denoted 'RW1980-2015'.
EW2015: This analysis is based on the latest year for all data sources (2015). Unlike EW1980-2015, this assessment includes a comprehensive analysis of the level of a single electricity plant after (Enipedia, 2018;Global Energy Observatory, 2018), which allows country-specific water and energy sources to be taken into account, as well as cooling technologies (also feeding back into EW1980-2015. The corresponding reported withdrawals are denoted 'RW2015'. As for EW, EC1980-2015 and EC2015 (estimated consumption) were calculated based on the same literature, methodology and years.
Steps in the analysis: (a) Water usage. As described in Section 2.2, this step in the analysis assessed the water withdrawal and consumption levels (volumes of water per energy unite.g. L/MWh) for each energy source and sub-technology within each of these. See Table 1 Figs. 1 and 2). (d) Sea−/freshwater use. At this point, the analysis provides an estimate of the total water usage for electricity generation within EU28. To address freshwater usage only, as described above, the databases of (Enipedia, 2018;Global Energy Observatory, 2018) were assessed for a single plant within EU28. This was done to determine the cooling water source and cooling technology. In a few select cases this information was readily available, whereas for most plants it proved necessary to inspect the plant location visually using satellite imagery. This extra inspection revealed further shortcomings in the databases that substantially affects the results (in relative terms), and these were therefore corrected accordingly. For some countries and technologies, a significant number of electricity plants exist, necessitating a substantial work load if all were to be represented. In these few cases, a threshold was used such that only the largest electricity plants constituting a minimum of 75% of the country's electricity generation were assessed. The remaining plants were then assumed to have the same distribution of technology and water sources, although this assumption might not hold fully (see Discussion). If the two databases differed, an online search was used to assess the (current) power-plant configuration. Plants with an estuary, fjord or river outlet location (most often UK and Benelux country plants) were assumed to use seawater, its usability for other sectors (e.g. irrigation) being poor or nonexistent. The resulting freshwater percentages are shown in Table 2.
Below, the key energy source and country-specific methodology issues are addressed.

Geothermal. The distribution of geothermal technologies within
Europe is based on (Bertani, 2015), as only a single Italian plant is listed within EU28 in (Global Energy Observatory, 2018), whereas Hungary, Ireland, Poland and UK also have geothermal electricity plants (Shift Project Data Portal, 2018). Solar-tidal-wave. To calculate the water usages of the 95.15 TWh produced by means of solar, tidal and wave generation, it is necessary to know the share produced by concentrated solar power (CSP), since this is the only one of these technologies to use water for cooling. Installed capacities in 2014 include 90,000 MW solar PV (SolarPower Europe, 2017), 2313 MW CSP (EurObserv'ER, 2017) and 5 MW of ocean energy (tidal and wave) (Magagna and Uihlein, 2015), corresponding to a solar CSP share of 2.5%. Here, the share of CSP technologies (parabolic trough, Fresnel or solar tower) is based on the list of CSP plants in (EurObserv'ER, 2017), and the specific water usages of these technologies are from (JISEA, 2015). whether or not wet/dry cooling is used is based on a manual assessment of CSP plants, concluding that all larger plants have a capacity in the order of 50 MW, and all use wet cooling (for plants where cooling type is listed -30 out of 45). These 45 50-MW CSP Table 2 Freshwater share (%) in electricity plant cooling per EU28 country and energy source in cases of the plant inspection differs from EU28 levels (as shown in left-hand column; see analysis step 'd' above). In the Baltic States a high level of mismatch between the databases was found, necessitating correction. One example is the absence of the gas−/oil-based Elektrenai power plant in Lithuania from the Enipedia database, even though it accounts for a dominant share of electricity generation in the country. Another example is the 84% use of coal declared for Estonian electricity generation at the Shift Project data portal, although no plants are listed either in Enipedia or by the Global Energy Observatory. Detailed data on the Czech Republic's electricity systems, including information on plants and cooling technology, were obtained from (Ansorge et al., 2016). For Greece, a lowered median water withdrawal rate of 2220 L/MWh was used compared to the 3238 L/MWh used for other countries (see references in Table 1 and values in Fig. 1 and Appendix A), as described in (Fernández-Blanco et al., 2017). For the Netherlands, the differences between RW and EW could be caused by differing definitions, since only 11 out of 32 gas plants (34% of generated electricity) are clearly located inland at river locations or by the sea, whereas for coal plants the corresponding level is 5 out of 8 (40% of generated electricity) (Global Energy Observatory, 2018). As the remaining parts are located close to the outlets of rivers or within estuaries, inconsistencies between the use of either the 'freshwater' or the 'sea' category could arise between databases. Similarly, the Borssele nuclear plant is located close to the Schelde River outlet but is listed as having freshwater cooling. Therefore, for the Netherlands, the estuary/downstream locations were defined as freshwater plants, since this was seen as the only plausible explanation for the high RW, although it is still not matched by EW using this definition. For Sweden, no coal power is listed by the Global Energy Observatory (Global Energy Observatory, 2018), whereas it is listed in Enipedia (Enipedia, 2018) and the Shift Project Data Portal (Shift Project Data Portal, 2018). This might be due to the interchangeable energy source for the Västerås power plant, the only coal-fired electricity plant in Sweden.

Results
In the following the results of reconstructing levels of the withdrawal and consumption of freshwater from the energy sector in EU28 are compared with reported withdrawal data from Eurostat   (Eurostat, 2018), freshwater shares per energy source and corresponding cooling technologies for these freshwater shares. The 'insert' plot corresponds to the main plot but uses a log-scale y-axis for improved reproduction of low-generation energy sources such CSP. (Eurostat, 2018). As discussed above power plants and their associated energy production (in TWh) in Europe, based on the available EU28 level information for the period from 1980 to 2015, are initially characterized, followed by estimates of corresponding water usages using the median, minimum and maximum factors from Figs. 1 and 2 (EW1980-2015and EC1980-2015. For 2015, for which more detailed information is available, an improved country-level analysis is carried out (EW2015 and EC2015).
Figs. 1 and 2 depict all the 896 median, min and max data points on electricity production water withdrawal and consumption which have been extracted in literature as a basis for this study (although a few are not visible within the selected Y-axis limits). See also Appendices A and B. A total of 56 technologies are listed based on eleven energy sources (gas is seperated between non-CCGT and CCGT for better visual overview). The span in levels from literature varies depending on energy source and technology but there is a tendency for the dominant combinations such as coal, gas and nuclear using tower and once-through cooling (see also Figs. 3 and 4) to have more consistent estimates. A tendency for the withdrawal estimates to be more similar than the corresponding consumption estimates is also seen. A reasonable level of agreement between RW and EW is seen (Fig. 3) using the median factors at the aggregated spatio-temporal levels used here. It is evident that conventional energy sources such as nuclear, coal and gas, which require excessive amounts of cooling water, clearly dominate the picture, whereas renewable (and less water-intensive) energy sources are essentially negligible. Results based on the more detailed country-level plant data from 2015 (indicated by a star), on the other hand, seems to slightly underestimate actual water use by the total European energy system. Furthermore, it is evident that the span between the minimum and maximum estimates is substantial (Fig. 3;  insert).
In the case of EC (Fig. 4), select renewables such as geothermal and biomass are found to play a significant role here compared to RW/EW, and EC2015 is estimated at a higher level than EC1980-2015. As seen in the insert (Fig. 4), a large span is seen between the maximum and minimum values As compared to Fig. 3, in this case it is difficult to assert the validity of the median estimates since no figures are reported for validation in the literature. Fig. 5 summarizes analysis step 'd' above by providing an overview of the estimated divison of cooling technologies within each energy source (sources with negligible water use are omitted) and the corresponding water source, that is, freshwater or saline water, for EU28. Despite a higher total level, the generation of electricity from nuclear sources (830 TWh) shows lower water-consumption figures than electricity generation from coal sources (791 TWh). This is related to the division of energy sources in between countries and their access to cooling water. For example, Germany produces a large share of its electricity based on coal as an energy source in inland locations. From the figure it is also clear that renewable technologies employ negligable amounts of water, though this omits biofuel production or evapotranspiration from hydropower reservoirs. Fig. 6 compares country-level estimated withdrawals (EW2015, top right) and estimated consumption levels (EC2015, bottom right) respectively with the reported values (RW2015, top left) based on the more detailed country-level analysis for 2015. From RW2015 it is clear that energy systems in Central Europe, in particular in France and Germany, are particularly water-intensive with regard to total volumes of water compared to the rest of EU28. This pattern is less significant when correcting for the electricity produced (insert, top left), where instead Belgium, Bulgaria, Estonia, Finland, Hungary and the Netherlands stand out. Furthermore, it is seen that the estimated water withdrawals for Europe generate country-level 'patterns' that resemble those of RW (see also Fig. 7). In general, RW is found to exceed RW2015 for most countries (bottom left) with some notable exceptions like Germany, Lithuania and the United Kingdom, although the latter's deviation is likely to be related to low absolute withdrawal levels due a high degree of seawater cooling. Slovenia also stands out (approx 100% overestimate). This is likely to be related to the single nuclear plant in Slovenia, which uses approximately 74% of all national cooling water according to the calculations (EW). Therefore the RW/EW difference may be related to differences between the median values, given the conditions of this single plant. In Finland, most of the larger electricity plants are located along the coastline while still having a freshwater use (RW) of approximately 5000 M m 3 . While most biomass plants in Finland are Fig. 7. Estimated and reported freshwater withdrawals, as well as estimated freshwater consumption with ranges based on minimum and maximum water-usage levels for each energy source and cooling technology. Cyprus, Luxembourg, Malta and Slovakia have no reported water withdrawals (marked with asterisk). located inland, adding to the country's freshwater use, a high relative difference of approximately 50% is still seen. For Italy, little attention has been given to investigating the reasons for the large spread between RW and EW, since the latest RW reporting dates from 1980, rendering any attempts at validation futile (the 1980 level is used throughout the period). Fig. 7 shows the water usage results for EW2015 (top panel) and EC2015 (bottom panel) for each country using the median, minimum and maximum levels. For EW2015, the results are shown in conjunction with RW2015 to enable their validation by country. Assuming this to be a first-order estimate of the 'true' associated uncertainty, the estimates (EC2015) approximate well to the corresponding reported values (RW2015), and all estimates are within the illustrated range associated with the country-level estimates except in the case of Malta, where information on local energy sources is lacking.

Discussion
As shown above, the estimates of water withdrawal rates compare reasonably well with the independent validation data from Eurostat and with the spatiotemporal scales assessed here, both in reproducing key aspects of the time series data from 1980 to 2015 and in representing the variability in water use for energy by country. Below the implications of the analysis presented here are discussed alongside recommendations for the direction of future data provision within the water-energy nexus.
To achieve the above-mentioned results an 'ensemble' of water-usage factors from a range of studies (cf. Table 1) was initially considered. While the authors by no means consider this factor set to be exhaustive, in each of the studies reviewed the estimated water uses (withdrawal and/or consumption) by different energy (cooling) technologies were inferred from local investigations. There was a tendency for them to originate in the USA in particular, possibly due to commercial interests resulting in data not being directly obtainable (Rothausen and Conway, 2011). One example of openly available data originating from the US is (Maupin et al., 2014), which contains very detailed information over time on, for example, water withdrawals per source (surface/groundwater) and per sector (industry, agriculture, housing), with numerous subcategories. As a result of the excessive data from the US, it might be expected that slight differences in local management practices and technology implementation between the US and Europe (and the rest of the world) might introduce a bias. A study by (Macknick et al., 2012), which also examines the water-usage data sources used in the present study, states that some median values are created based on the midpoint between the range endpoints, which may introduce an underestimation bias. Moreover, for some energy production and/or cooling technologies only a very few studies exist (cf. Table 1). In summary, these potential sources of uncertainty and biases imply a lack of complete statistical robustness with regard to the term 'median'. However, the median estimates in the studies considered here inherently agree well (results not shown), suggesting that the median factor seems to represent a fairly robust metric with which to represent yearly water usage on the larger European scale. Conversely, and not surprisingly, there is greater variability in the case of the minimum and maximum factor values (e.g. as illustrated by the inserts in Figs. 3 and 4), which are used here to represent the uncertainty associated with the factor estimates.
Another source of uncertainty is the assumption that the lowgeneration plants have the same technology/water source characteristics as the top 75% of electricity plants (calculated on the basis of electricity generation; see above). Smaller plants tend to have the oldest technologies (i.e. once-through cooling) using more water per unit of electricity generated (as seen in the databases listed above and in (U.S. Dept. of Energy, 2006)), which would imply a negative bias in respect of the estimates of water withdrawals. Here this bias is regarded as largely negligible for EU28 based on the low extent of countries and energy sources where not all plants were assessed.
Mapping European energy production convincingly represents another considerable source of uncertainty. In this study information from several sources is combined to represent energy production and dominant technologies across EU28, as well as to detect and account for obvious gaps or flaws in the underlying data sets. The availability of accurate information in this regard is clearly critical for the results, as this might otherwise lead to erroneous results and/or introduce significant biases in estimated water usages. That said, even for the relatively data-sparse case of modelling water use by energy sources for EU28 between 1980 and 2015, a decent level of agreement with the validation data was found (e.g. Figs. 3 and 7).
A key motivation for conducting the present study estimating water usage in electricity production is the lack of data available to analyse the water-energy nexus properly in quantitative terms, as also highlighted by (Chini and Stillwell, 2017;Larsen et al., 2016). Optimally, such data should be freely available, have wide (global) coverage, have a high temporal resolution, have detailed information on a single plant level, as well as with regard to water (source/sink), have shared and userfriendly formats, be collected using the same conventions, be forced by the same conditions (e.g. for future scenarios) and be congregated in a single database for easy acquisition. Taken together these suggestions are fundamentally out of reach for the near future for reasons of policy (Scott et al., 2011), coordination, financing and commercial interests (Goldstein et al., 2008). However, every step in this direction would improve the possibilities and range of the analytical steps and assumptions needed for most nexus analyses and, in essence, facilitate the emergence of firmer conditions for studying the implications of anthropogenic energy-related activities and water-management perspectives, as (Hussey and Pittock, 2012) also conclude.
By embedding the assembled median water-usage estimates within energy-system models such as TIMES (Simoes et al., 2013) or OSeMOSYS (Howells et al., 2011), the water-requirement impacts could be dynamically simulated and used to investigate a range of water management-and policy-related issues more reliably. In the context of the water-energy nexus, another potential issue to investigate is the impact of introducing more renewable energy technologies into the energy grid and/or add water shortages and related market demands.

Conclusions
This paper has investigated the extent to which estimates of an ensemble of factor sets representing volumes of water use per unit of electricity produced by different energy sources, in conjunction with a comprehensive review of the individual cooling technologies and water sources for individual electricity plants, provide an adequate predictor of water withdrawal and consumption levels generation by energy production. Based on validation data, the factor set estimates are found to generate a skilful reproduction of reported withdrawals on the coarse scales assessed here, which, on a yearly basis, include historical levels for EU28  and individual country levels (2015). In the present demonstration, country-level information was extracted from a number of databases, some of which were found to be ambiguous, inadequate and/or contradictory. To support quantitative studies of the water-energy nexus at different levels and thus facilitate the improved management of water resources, the authors therefore recommend sustained and coordinated efforts towards improving the availability of data linking observations and projections of, for example, energy and water systems at relevant spatio-temporal levels using common scenarios (including climate change) and assumptions. Appendix A Median water withdrawals. (continued on next page)