A systematic review of the impacts of climate variability and change on electricity systems in Europe

Understanding the impacts of CV&C2 (climate variability and change) on electricity systems is paramount for operators preparing for weather-related disruptions, policymakers deciding on future directions of energy policies and European decision makers shaping research programs. This study conducted a systematic literature review to collate consistent patterns of impacts of CV&C on electricity systems in Europe. We found that, in the absence of adaptation and for current capacity, thermal electricity generation will decrease for the near term to mid-21st century3 (NT-MC) and the end of the 21st century4 (EC). In contrast, renewable electricity generation will increase for hydroelectricity in Northern Europe (NT-MC and EC), for solar electricity in Germany (NT-MC) and the United Kingdom and Spain (NT-MC and EC) and for wind electricity in the Iberian Peninsula (NT-MC) and over the Baltic and Aegean Sea (NT-MC and EC). Although the knowledge frontier in this area has advanced, the evidence available remains patchy. Future assessments should not only address some of the gaps identified but also better contextualise their results against those of earlier assessments. This review could provide a starting point for


1-INTRODUCTION
Devastating consequences of extreme weather are repeatedly making the front pages of the media across Europe, as they challenge the provision and security of critical services (e.g. BBC (2015;; Gayle (2015)). Understanding the impacts of climate variability and change (CV&C) on electricity systems 4 is increasingly important not only for electricity companies providing such critical services, but also for policymakers in charge of ensuring the security of a country's electricity supply. As energy infrastructures form the central nervous system of all economies, interruption of electricity provision can have consequences reaching far beyond the electricity systems themselves.
Although the global impacts of CV&C on the energy sector have been explored in the literature (Ebinger andVergara 2011, Bruckner T., Bashmakov et al. 2014), the impacts of CV&C on the electricity systems have received less attention and regional, national and local assessments are still rare (Chandramowli and Felder 2014).
Existing studies of impacts of CV&C on electricity systems can be divided into three strands. First, some studies use the findings from empirical literature to assess the impacts of CV&C beyond electricity systems. For example, Mideksa and Kallbekken (2010) examine the impacts of CV&C on demand and supply in the electricity markets whilst Rübbelke and Vögele (2011; investigate the impacts of global warming on trade in electricity between European countries and on national electricity prices. Schaeffer, Szklo et al. (2012) explore the literature on the impacts of CV&C on resource endowments, energy supply, and energy use and infrastructure.
Second, some assessments, such as Klein, Olonscheck et al. (2013), construct indices to assess the susceptibility of the energy sector to the impacts of CV&C: they compare the impacts on energy systems in 21 European countries using an index based on variables such as summer temperature increases, discrepancies between production and consumption and the volume of imports and exports. Bardt, Biebeler et al. (2013) in turn compute risks and opportunities posed by changing climatic conditions for energy sectors in France, Germany, Norway and Poland on the basis of expert interviews.
Third, some assessments focus on the statistical relationships between climatic and energy variables. They use the outputs of climate modelling experiments as inputs in electricity generation and network impact models. Peer-reviewed articles using this approach were the objects of this systematic review. The systematic review approach was used in order to collate, evaluate and interpret all the results of such research.
This review aims to identify the impacts of CV&C on electricity systems in Europe to answer the questions: i) what patterns of impacts of CV&C on electricity systems can be identified by collating the results of peer-reviewed articles? ii) are any of these patterns robust?
The rest of the article is divided into four sections. Section two describes the method used in the systematic review and the data. Section three presents the results of the systematic review, including robust patterns of impacts of CV&C on electricity systems in Europe. The final two sections discuss the implications of the results for further studies and for decision-making and conclude.

2.1-Method
The peer-reviewed articles included into this study were selected using a systematic literature review (SLR, see Berrang-Ford, Pearce et al. (2015)). A literature review is "systematic" when it is based on a clearly formulated question, identifies relevant studies, appraises their quality and summarises their evidence (Khan, Kunz et al. 2003). The SLR methodology is explicit and contains enough information to be reproducible. SLRs collate, evaluate and interpret all research available and relevant to a particular question, topic area, or phenomenon of interest. SLRs are widely used in medical research but they are still under-utilised in other disciplines including in climate science (Porter, Dessai et al. 2014).
The well-defined methodology makes SLRs less likely to be biased. SLRs can also provide information about the effects of a phenomenon across a wide range of settings and empirical methods; if the studies yield consistent results, the reported effects can be considered robust. If, on the other hand, the SLR yields inconsistent results, these dissimilarities can be analysed further (Biondi-Zoccai, Lotrionte et al. 2011).
SLRs have also their shortcomings. They are time-sensitive snapshots of the literature on their subject. Another drawback is closely linked to the type of evidence commonly used in SLRs: significant results published in peerreviewed articles, which leads to under-representation of non-significant results.
The results of the reviewed articles were collated to assess whether robust patterns of impacts of CV&C can be identified at regional, national or subnational scales on any parts of the electricity systems. The term "robust" does not refer here to "statistical robustness" as is sometimes done in climate science where future changes are considered robust "when i) present-future model ensemble mean difference is significant at the 95 % confidence level according to the Wilcox-Mann-Whitney test applied to the whole model ensemble (adapted from Jacob, Petersen et al. (2014)) and ii) at least 12 models out of 15 agree on the sign of change" (Tobin, Vautard et al. 2015). In this SLR we use Lloyd (2015) definition of robustness as "the standard convergence of predictions/retrodictions of multiple instantiations of variants of the model-type, as well as exploration and empirical confirmation of an array of empirical model assumptions, which can be seen as aspects of random, well-supported experiments when a variety of evidence inferences to support the core structure are used". This is a more qualitative take on robustness, in which the convergence of the results of independent empirical studies corroborates a given phenomenon.
The SLR was carried out in three successive steps: a) search for peerreviewed articles in Scopus using different keyword combinations; b) screening of the returned articles by applying inclusion and exclusion criteria and a star-rating scorecard, and; c) collation and analysis of the results from the subset of included articles.
First the accuracy of the search strategy was ensured by comparing the returned articles resulting from searches in Scopus to a benchmark collection of relevant studies collated from previous work (Bonjean Stanton, Dessai et al. 2016). Then, 734 searches were run in Scopus using the improved keyword combinations. The searches yielded a total of 24463 articles (including duplicates). Once imported into the EndNote software, the articles were screened using inclusion and exclusion criteria. The retained peerreviewed articles were in English, with European coverage (as defined by the United Nations Statistics Division), and focussing on the impacts of CV&C on electricity generation and networks in the near-, medium-and long-term. Following Porter, Dessai et al. (2014), the retained articles were screened using a scorecard to differentiate between rigorous and less rigorous publications. The scorecard's star-rating scheme ranges from zero to five stars. In a five star article the study design and methods are highly appropriate for the research question and they are clearly outlined and justified. Several climate models and scenarios are used for assessing impacts for several time-periods, annually and seasonally. The information on the calibration and validation of the climate and impact models used is explicit. The results are triangulated and set in the context of other studies (e.g. Finger, Heinrich et al. (2012); Majone, Villa et al. (2015); See Supplementary Material). In a four star article, the methods are clearly justified and several climate models and scenarios are used in the assessment but information on model calibrations, study limitations, or result triangulation is missing. In a three star article, the chosen method is appropriate for the assessment to be carried out. Information on the number and types of climate scenarios and climate and impact models used and their calibration is mentioned but not explained in detail. The results are clearly presented but their implications are not outlined explicitly nor triangulated against other studies. Articles using a single climate scenario, 1-2 climate model(s) and pre-compiled climate variable datasets were also classed as three star articles. Articles scoring less than 3 stars were excluded; such articles provided too little information on the method and the datasets used in the assessment and hence the results of such studies were not considered to be sufficiently rigorous to be included in this review.
Out of the 50 peer-reviewed articles retained for review, 9 were classed as five star, 29 as four star and 12 as a three star. Using the latest climate models or scenarios (e.g. the Representative Concentration Pathways, RCPs) did not automatically qualify the article as five star; all the scorecard attributes were considered conjointly to assign an article to a star category.

2.2-Data
There were 50 articles scoring three stars or more. They were retained for further analysis and labelled #1-50 (See Supplementary Material). Their publication dates range from 1997 to 2015: there are more publications for years 2012 and onwards compared to the earlier years (Figure 1). A third of the articles are on hydroelectricity generation, followed by articles on wind electricity (28%), thermal electricity (14%), solar electricity (13%), bioenergy (7%), and wave energy (3%). One article focused on the electricity networks (2%).

Figure 1:
Retained articles by publication year (1a) and by electricity system focus (1b) 1a) 1b) Information was collated on the authorship, assessment methods, results, limitations and research gaps of each retained article by using a qualitative record sheet template. In particular, it was discerned: i) what are the projected impacts of CV&C (positive, negative, no significant impact) on the electricity systems for the period of assessment in the articles? and ii) whether these A total of 43 articles on the impacts of CV&C on hydro-, wind, thermal and solar electricity generation were analysed and the results are reported in the next section. Results from the articles focusing on bioenergy, wave energy and electricity networks (n=7) were not included in the analysis because of the limited and conflicting evidence base they provided but are presented in the Supplementary Material. The remaining 43 articles had assessment periods chosen for reasons of their own (See Supplementary Material). In some articles, the choice was justified by invoking the electricity infrastructure lifespan, whereas others provided little or no justification for the chosen assessment period. The heterogeneity of used assessment periods made it difficult to gain an overall view of the results. To address this challenge, we re-mapped the articles and their results onto two time periods, near term to mid-21 st century and the end of the 21 st century. Near term to mid-21 st century (NT-MC) covers the period from the present until 2071, while the end of the 21 st century (EC) covers the period from 2061 until 2100. There were 22 articles covering near term to mid-21 st century and 11 articles covering the end of the 21 st century. Both periods were covered by 10 articles.
Each article was scrutinised for its results, and an individual result was chosen as the unit of analysis. A result is "individual" if the article outlines it explicitly and its interpretation is not left to the discretion of the reader. An individual result can be explicitly outlined in a table (e.g. Table 2 in Lehner, Czisch et al. (2005)), a figure (e.g. Figure 4 in Crook, Jones et al. (2011)) or in the text (e.g. Baltas and Karaliolidou (2010)). Some articles have several individual results (e.g. Van Vliet, Vögele et al. (2013)) whereas others only have a single one (e.g. Baltas and Karaliolidou (2010) Individual results from the 43 articles were organised by i) the type of electricity generation (hydro-, wind, thermal and solar electricity generation), ii) geographical coverage (regional, national and sub-national scale) and iii) assessment period (near term to mid-21 st century or the end of the 21 st century). Each combination could have more than one individual result, one individual result, or no result. A pattern of impacts of CV&C was identified when all relevant individual results were consistent, with the pattern direction of change (positive or negative) reflecting the envelope of individual results. When the individual results were inconsistent, no pattern was attributed. If a single individual result existed, a pattern was attributed only if several climate models or scenarios were used in the generation of the individual result. In total our sample contained 498 individual results.
Some limitations remain in the reported systematic review. We used the UN Statistics Division's clustering of countries to define European regions (Northern, Western, Eastern and Southern Europe). However, as some articles give limited information on their spatial coverage, the exact match of the results with the UN Statistics Division's clustering of countries cannot be fully guaranteed. Also, some articles cover a long time span including both near term to mid-21 st century and the end of the 21 st century: this makes it difficult to distinguish which impacts to allocate to which assessment period. Therefore, these individual results were allocated to both assessment periods (e.g. #11: 2010-2080; #29: 2020-2080; #30: 1990 -2080/2100). Articles on the same type of electricity generation were collated regardless of some differences in addressed generation technology and infrastructure. For example, articles on hydroelectricity generation included impact assessments for run-of-the-river and storage reservoir plants, and articles on thermal electricity generation examined generation from fossil fuels and nuclear fuels. The statistical significance of individual results was indicated in some articles but not in others; individual results with no mention of their statistical significance were still included, but non-significant results were not when explicitly characterised as such. Finally, all the reviewed articles are in English, disregarding results reported in other languages. Funding information, where available, revealed that the European Commission, national research councils and ministries, and academic institutions (e.g. university research departments) financed most of the studies, with the exception of one study (#29), commissioned directly by a national energy association.

3.1-Landscape of methods of analysis
The reviewed articles use quite different methods of analysis. The simplest ones take climate data as proxy for the impacts of CV&C (e.g. # 10), whereas more complex ones use outputs of climate model experiments as inputs to comprehensive impact models (e.g. #27).
The climate data used in the assessments can be taken directly from existing climate change projection datasets (e.g. UKCP09 in #6) or be simulated by a) combining emissions scenario(s) and climate model(s)/projection(s) (e.g. #2, #13, #27, #43) or b) by rearranging observed time series with respect to a given linear trend for a selected variable (e.g. STARS 5 in #24). The statistical measures of climate data (e.g. mean, median, distribution) used as inputs to the impact models, also vary.
The impact models used in the articles vary from validated and widely accepted models (e.g. IHACRES 6 ) to models specifically developed for the articles and conveyed by a single equation or more complex computations. Impact models also tend to reflect the dominant impact pathway.
Hydroelectricity generation depends directly on the hydrological cycle. CV&C affect hydroelectricity generation through the availability of excess water (precipitation minus evapotranspiration) and the seasonal pattern of the hydrological cycle in regions where snowmelt is a relevant factor for generation (Schaeffer, Szklo et al. 2012). The impacts of CV&C on hydroelectricity generation are assessed using hydrological models (e.g. rainfall-runoff models such as IHACRES, TOPKAPI 7 or HBV Model 8 , GEOTRANSF 9 ) or models simulating hydroelectric power plant operations.
Energy contained in wind is proportional to the cube of the wind speed (Pryor and Barthelmie 2010) and thus variations in wind speed can have significant effects on generation. Schaeffer, Szklo et al. (2012) indicate that wind speed varies significantly with height and that little is known about likely future wind speeds at the hub height of a wind turbine (above 50 m). In the reviewed articles, the impacts of CV&C on wind electricity generation is assessed either by taking future wind projections (e.g. GCM geostrophic wind) as proxy for wind power production, or by extrapolating wind speed for the specific height of the hub of the analysed wind turbine model. Thermal electricity generation using coal, natural gas, nuclear isotopes, geothermal energy and biomass depends on the availability and temperature of cooling water. Its efficiency depends on the heating and cooling needs of both Rankine and Brayton cycles, which in turn vary according to the average ambient conditions such as temperature, pressure, humidity and water availability (Schaeffer, Szklo et al. 2012). Reliability of supply can also be threatened by water abstraction and regulations on discharge water temperature (Naughton, Darton et al. 2012). Water use models (e.g. WaterGAP3 10 ), eco-hydrological models (e.g. SWIM 11 ), hydrological models and specific models of thermal electricity generation were all used. Solar electricity generation can be impacted by extreme weather events, changes in snow and cloud cover and air temperature increases. Changes in air temperature not only modify photovoltaic (PV) cell's efficiency and reduce generation (Pašičko, Branković et al. 2012), but also negatively affect temperature-sensitive Concentrated Solar Power (CSP) systems. The impacts of CV&C on solar electricity generation are assessed by using the delta change method, assessing the differences between simulated current and future climate conditions, by developing models of PV power generation, or by deriving the power output from irradiance and ambient temperature data.
Some of the reviewed articles explain the rationale for the choice of the assessment period(s) and used climate and impact models but most do not. Many articles develop their own methods of analysis, combining a unique set of climate data and impact models. Most articles (with the exception of e.g. Hoffmann, Häfele et al. (2013)) also assess the impacts of CV&C on the basis of climate signals only, and neglect to consider feasible adaptation measures or future change in policies and regulations. Impact models developed in some of the reviewed articles are based on the existing types of electricity infrastructure, designed on the basis of historical meteorological records and not future climate projections. The articles also assume that no new electricity infrastructure will be built and that generation capacity will remain constant. Moreover, all but a few articles consider only one technology for a given type of electricity generation. Lehner, Czisch et al. (2005) do consider both run-offthe-river and reservoir solutions for hydroelectricity generation, Crook, Jones et al. (2011) include in their analysis the two most widely installed solar technologies for large-scale electricity generation, namely photovoltaic (PV) and concentrated solar power (CSP)) and Van Vliet, Vögele et al. (2013) assess different types of thermal electricity generation plants. As a consequence, the methods of analysis were not examined further in the analysis.

3.2-Consistent patterns of impacts of CV&C
This section explains the consistent patterns of impacts of CV&C on hydro-, wind, thermal and solar electricity generation at the regional and national scales. The robustness of the patterns of impacts of CV&C is indicated for the regional and national scales, for which there were more often more than one individual result available (in bracket and in italic; NT-MC: near term to mid-21 st century and EC: end of the 21 st century). We use the number of available and consistent individual results as a proxy for robustness; a pattern of impacts of CV&C identified from four or more individual results is considered more robust that one derived from a single result. Robustness is not considered at the sub-national scale because only single individual results were available at this scale. At sub-national scale, impacts were mostly derived from one individual results per location, not allowing for any pattern to be extrapolated. As such, subnational scale impacts of CV&C are only discussed in the Supplementary Material. Figure 2 summarises the annual consistent patterns of CV&C on hydro-, wind, thermal and solar electricity generation at regional scales. Positive patterns can be observed for renewable electricity generation in Northern Europe and negative patterns for both renewables and traditional electricity generation for the Western, Eastern and Southern Europe.

Hydroelectricity generation
Hydroelectricity generation from the installed hydropower capacity is expected to drop from 10% of the EU27 electricity generation in 2013 to less than 6% by 2050 as the result of future changes in rainfall (#12).
Hydroelectricity generation will increase in Northern Europe (2 individual results available for NT-MC and 1 for EC) and decrease in Western (NT-MC: 1; EC: 1) and Southern Europe (NT-MC: 2; EC: 2) by near term to mid-21 st century and by the end of the 21 st century. In Eastern Europe, hydroelectricity generation will decrease in the near term to mid-21 st century (1).
Hydroelectricity generation is projected to increase in winter in Northern Europe (1) and decrease in summer for Southern Europe (1) for the end of the 21 st century.

Wind electricity generation
No consistent patterns of impacts of CV&C on wind electricity generation are projected for Northern Europe for the near term to mid-21 st century (3). For Northern Europe, an annual increase (3) and an increase for the winter months (1), and a decrease for the summer months (1), are predicted for the end of the 21 st century. For Southern Europe, wind electricity generation is predicted to decrease in the near term to mid-21 st century and for the end of the 21 st century (NT-MC: 1; EC: 2). A decrease in generation is also predicted for summers in Western Europe (1) and summers (1) and winters (1) in Southern Europe for the end of the 21 st century. The decrease for Southern Europe is consistent with a decrease in annual wind electricity generation in the Mediterranean Sea for the near term to mid-21 st century and the end of the 21 st century (NT-MC: 2; EC: 2).

Thermal electricity
Annual thermal electricity generation is projected to decrease in Western Europe (1) and Southern Europe for the near term to mid-21 st century (2). This projection resonates with the projections for decreasing precipitation for Southern Europe , reducing the volume of runoff available for use as cooling water.

Solar electricity generation
Annual solar electricity generation is projected to increase in Western Europe (1) and to decrease in Eastern Europe for the near term to mid-21 st century (1).

End of the 21st century
Icons made by Freepik from www.flaticon.com

End of the 21st century
Icons made by Freepik from www.flaticon.com

End of the 21st century
Icons made by Freepik from www.flaticon.com

3.2.2-Patterns of impacts of CV&C on hydro-, wind, thermal and solar electricity generation at national scale
Figures 3 and 4 present the annual patterns of impacts of CV&C on hydro-, wind, thermal and solar electricity generation at the national scale and in the Baltic and Mediterranean seas and Iberian Peninsula for the near term to mid-21 st century and the end of the 21 st century, respectively. The figures also indicate where no pattern could be identified.

Figures 3
and 4 indicate that national scale assessments of impacts of CV&C are still largely missing for wind, thermal and solar energy generation for the near term to mid-21 st century and the end of the 21 st century. More individual results are available for the near term to mid-21 st century than for the end of the 21 st century. There is more agreement between individual results for the end of the 21 st century than for the near term to mid-21 st century, resulting in more consistent patterns of impacts of CV&C for the later period. This is consistent with stronger climate signals towards the end of the century.

Hydroelectricity generation
Finland is the only country with a confirmed positive pattern of increased hydroelectricity generation for the near term to mid-21 st century (4). Northern European countries of Estonia (2), Finland (3), Iceland (2), Latvia (2), Norway (3) and Sweden (3) and Belarus (2), and the European part of the Russian Federation (2) in Eastern Europe, are also projected to experience an increase in hydroelectricity generation in the end of the 21 st century.

Wind electricity generation
There is substantial uncertainty associated with assessing projected changes in wind (Pryor, Barthelmie et al. 2005). Despite this, reviewed articles indicate some patterns. An increase in annual wind electricity generation is projected for the Baltic and the Aegean Seas for the near term to mid-21 st century and the end of the 21 st century (respectively for the NT-MC: 2, 1; EC: 2, 3) and for the Iberian Peninsula (1) for the near term to mid-21 st century. An annual decrease is projected for the Mediterranean Sea for the near term to mid-21 st century and the end of the 21 st century (NT-MC: 2; EC: 2).
Wind electricity generation is projected to increase in summers for the Baltic and Aegean Seas (respectively: 1 and 1) and in winters (November to February) for Germany (1) and Ireland (2) in the near term to mid-21 st century, and for the United Kingdom (1) for the end of the 21 st century.
A decrease in wind electricity generation is projected for summers for Ireland (2) and Germany (1) in the near term to mid-21 st century, and for France (1), the United Kingdom (2), Germany (2) and Poland (1) for the end of the 21 st century. A decrease is projected for springs and autumns for the Iberian Peninsula for the end of the 21 st century (2).

Solar electricity generation
Annual solar electricity generation is projected to increase for the United Kingdom, Germany and Spain for the near term to mid-21 st century ( (3), (4), (4)), and for the end of the 21 st century ( (3), (4), (4)).

4-DISCUSSION
Robust negative patterns of impacts of CV&C were identified for thermal electricity generation for the near term to mid-21 st century and the end of the 21 st century. In contrast, positive patterns were identified for renewable electricity generation; robust positive patterns of impacts of CV&C can be found from the projections for increased generation of hydroelectricity in most of Northern Europe in the near term to mid-21 st century and end of the 21 st century, for solar electricity in Germany in the near term to mid-21 st century and in the United Kingdom and Spain in the near term to mid-21 st century and end of the 21 st century, and for wind electricity in the Iberian Peninsula in the near term to mid-21 st century and over the Baltic and Aegean Sea in the near term to mid-21 st century and end of the 21 st century.
Future climate projections are in agreement about an increase in temperature throughout Europe, and about increasing precipitation in Northern Europe and decreasing precipitation in Southern Europe (Jacob, Petersen et al. 2014). Episodes of high temperature extremes are also expected to become more frequent (high confidence) and so are meteorological droughts (medium confidence) and heavy precipitation events (high confidence) ). These climatic projections resonate with the patterns of impacts of CV&C on electricity systems identified in this systematic review. Increased ambient air temperatures will decrease the efficiency of thermal generating plants and reduce thermal electricity generation across Europe. Higher precipitation will be favourable to hydroelectricity generation in Northern Europe, but decreasing precipitation will reduce hydroelectricity generation in Southern Europe (Figures 3 and 4).
The results of this review also highlight further the vulnerability to CV&C of more traditional electricity generation technologies such as thermal power plants.
The key issue in managing such assets in the face of future changes is that the past can no longer be assumed to be the best guide for the future. As such infrastructure managers should not rely only on past conditions but also consider a range of future scenarios. They should also envisage potential adaptation options for not only climate-proofing traditional technologies but also diversify their electricity generation asset portfolio and encourage the penetration in the energy mix of less climate vulnerable electricity generation technologies such as renewables. Transitioning towards more renewable sources of electricity could also simultaneously support the achievement of the European Union's commitment to reduce GHG emissions from 1990 levels by 40% by 2030 and by 80-95% by 2050, to retain global warming below 2ºC (European Commission 2011). It would also help achieving the binding EU target of covering at least 27% of the European energy consumption from renewable sources by 2030 (European Commission 2014).
A systematic review of the assessments of impacts of CV&C on electricity systems makes several contributions. First, validation and invalidation of specific results can lower uncertainty and remove barriers from decisionmaking. Second, as most individual results are not directly transferable to other locations (e.g. Gaudard, Romerio et al. (2014)) or attributable to other electricity infrastructure assets, a systematic review can help to assemble the puzzle of the future impacts of CV&C on electricity systems. Finally, the envelopes of results represent versions of possible futures that policymakers and electricity operators will have to prepare for. They can inform policymakers' plans for a future energy mix capable of withstanding the impacts of CV&C, and interruptions related to them, to ensure the reliability and security of electricity provision. Electricity operators can use such evidence to re-think future investments in electricity generation infrastructure, especially those with long-term lifespan such as hydroelectric dams, and thus limiting the risks of stranded assets. Electricity companies, carrying out their own CV&C risk assessments can also use such evidence to triangulate and reinforce their own findings.
This systematic review identified robust patterns of impacts of CV&C from peer-reviewed articles published in English. Although the knowledge frontier in this area has advanced, the evidence available is still sparse. Little robust assessments still exist on thermal generation (combustible fuel and nuclear power plants) for the near term to mid-21st century and the end of the 21st century. As thermal electricity is the main source of electricity in Europe at present 12 and is likely to remain very prominent in the future electricity mix, understanding more consistently the impacts of CV&C on thermal power plants is paramount to better plan for energy security in the future. Some articles also explored the impacts of CV&C on renewable electricity but to the authors' knowledge no study exists looking more holistically at the potential for future renewable installation capacity at European or national levels and at the effects of renewable penetration on future electricity systems. Additionally, most existing articles assess near term to mid-21 st century impacts and fewer articles cover end of the 21 st century impacts (Figures 3 and 4). Even fewer articles consider intra-annual or seasonal variations. The spatial coverage of assessments is also uneven. Few assessments focus on the impacts of CV&C at national scale on thermal, wind electricity and solar electricity generation. Sub-national and infrastructure scale assessments are also largely missing, yet they would be key in supporting decision-making. Furthermore, many articles have quite static approach; climate parameters are often the only variables and the energy mix, the commissioning and decommissioning of assets, and the technical parameters for electricity generation are considered constant. Technology innovation is not taken into consideration and nor are future technologies with increased energy efficiencies.
There are inherent cascading uncertainties associated with the climate and impact models used in the assessments, and yet these uncertainties are rarely discussed explicitly in the reviewed articles. There is also little reflection on what the implications of these uncertainties are in practice and how confident the readers and users can be in the results. Future assessments of impacts of CV&C on electricity systems should tailor the communication of results and uncertainties associated with them to specific audiences. Latest literature on communicating climate science would help to better understand the target audiences' needs and preferences, and to tailor the communication of results accordingly (e.g. EU FP7 Euporias 13 ). Furthermore, future assessments should communicate uncertainties and confidence in the results more explicitly (Lorenz, Dessai et al. 2013). For example, the latest IPCC AR5 report uses two metrics for communicating the degree of certainty in key findings: confidence in the validity of a finding, based on the type, amount, quality and consistency of evidence and a quantified measure of uncertainty in a finding expressed probabilistically (Intergovernmental Panel on Climate Change (IPCC) 2014).
The articles should also be more explicit about their limitations and outline if possible what the implications of their results are for the stakeholders. For example, few of the reviewed assessments reflect on how to adapt the electricity systems to the impacts of CV&C found in their results.
Assessments of impacts of CV&C on electricity systems often assess the impacts of a single climate variable (a proxy for climate change) on one type of electricity generation or infrastructure asset. To the authors' knowledge, no article has yet looked at the impacts of a climate variable along the whole chain of electricity provision (e.g. the impact of decreasing rainfall on electricity generation and network infrastructure) or investigated the impacts of concomitant weather events on one type of electricity generating technology (e.g. the simultaneous impact of a massive earthquake and a tsunami like in Fukushima in Japan in 2011). Little is also still known about the impacts of CV&C on sector interdependencies. For example, reduced rainfall could lead to droughts, which in turn could translate into not only decreased thermal electricity and hydroelectricity but also into bans and levies on water extraction for irrigation or human consumption. Finally, another area of importance for future modelling is adaptation. Adaptation options should be included in future assessments of impacts of CV&C on electricity infrastructure and the technological and economical efficacy of such option evaluated for different climate scenarios. Such studies could be invaluable to help infrastructure managers to climate-proof their assets, to ensure national electricity security and to avoid potential maladaptation.

5-CONCLUSION
This systematic review is the first attempt at collating the impacts of CV&C on electricity systems in Europe from peer-reviewed literature published in English. The review indicates that although the evidence base is improving and yields some robust patterns, there is still a need for additional empirical research.
In future assessments there is a need to better contextualise the results against those of earlier assessments. This review can provide a starting point for doing so. Future assessments should also link their results and their implications to user needs and consider how the results are best communicated. Few attempts have been made to date to integrate the assessments of impacts of CV&C on supply and demand of electricity (e.g. Chandramowli and Felder (2014); Ciscar and Dowling (2014)). Such could be the next step in assessment of risks CV&C pose for electricity systems.
This review identified some consistent patterns of CV&C impacts on electricity systems in Europe. As the climate is changing so should energy infrastructure management, policies and the future directions of research. This work could inform not only infrastructure managers trying to climate-proof their assets and avoid resource misallocation but also policymakers shaping future European Energy policies and the European Commission when shaping the future research and funding programs.

List of Appendices:
Appendix A-Detailed method followed in the systematic review Appendix B-Data: Peer-reviewed articles included in the systematic review and their characteristics Appendix C-Peer-reviewed articles included in the systematic review but excluded from the analysis Appendix D-Impacts of Climate Variability and Change (CV&C) on hydro-, wind, thermal and solar electricity generation at sub-national scale

Appendix A-Detailed method followed in the systematic review
The systematic review was carried out in three steps as illustrated in Figure  5. Figure 5: The four-step process followed to carry out the systematic literature review Step 1: Scopus keyword combination searches The keywords used Table 1 presents the keywords that were combined for the searches.

The search process
Each search was carried out using the following combination of keywords: "One keyword word from Level 1 AND One keyword from Level 2 AND One keyword from Level 3".
Several combinations of keywords were tested. Results with search terms x and y returned few relevant articles. The relevant articles returned were already covered by other search terms combination This led to 734 search combinations returning 24463 resources (including duplicates).
Step 2: High level screening of the articles returned for each of the keyword combination search The articles returned for each keyword combination search were screened and only retained if they met all of the following inclusion criteria: -Content relevant for Europe / Assessment made for a European country or region (as defined by the United Nations Statistics Division 14 ) -In peer-reviewed journals -In English (both Abstract ad Full Text) -Articles focusing on impacts of climate variability and change (CV&C) on electricity generation and transmission in the xxx Note: studies on energy resource endorsement were excluded (e.g. impacts of CV&C on coal mining when coal is used as a fuel for thermal electricity generation) Step 3: Screening using a star-rating scorecard The remaining articles were then further assessed using the star-rating scorecard outlined in Table 2. A 5* paper is a paper that includes all the individual attributes outlined in the scorecard.
Only fifty articles in total were retained in this study as a result of the systematic review. Their full references can be found in Table 3 in Appendix B. There is a good balance in the paper between the methods and the results section (some paper have a lot of info on assessment method but the result section is rather underdeveloped even if the key messages are there OR the paper described the model used in details in another paper and concentrates on the results) Methods

M1
The method used for the assessment, etc is outlined M2 The method used for the assessment, etc is clearly outlined. The information given about the assessment method are enough to allow the study to be reproduced for a different location M3 The method clearly explains why one climate model, impact model, region of assessment was chosen over another) M4 The method uses several climate models to create an envelope of climate data / uses ensembles of climate data References: -"Ensemble means have proven to be more accurate than individual models in reproducing the instrumental observational period" (From: Gleckler, P.J., Taylor The method uses several climate scenarios to forecast different future conditions M6 The method assesses the impact in the near term to mid-21st century and the end of the 21st century M7 "The information on the calibration and validation of the climate and impact model used is explicit The climate models were rigorously tested before they are applied Reference: Refsgaard, J.C., Madsen, H., Andréassian, V., Arnbjerg-Nielsen, K., Davidson, T.A., Drews, M., Hamilton, D.P., Jeppesen, E., Kjellström, E., Olesen, J.E., Sonnenborg, T.O., Trolle, D., Willems, P., Christensen, J.H., 2014. A framework for testing the ability of models to project climate change and its impacts. Climatic Change 122, 271-282 M8 The method assesses annual changes as well as seasonality (intra seasonal variations) M9 The impact model used has been widely applied and tested in various contexts Results

R1
The results are explicit R2 The results are consistent and answer the question raised R3 The paper mentioned further information about the results. This can be for example limitations associated with the method that influence the results, uncertainties associated with the results, confidence intervals of the results, taking the results with caution etc.

R4
The paper mentions what the results could be used for and by whom and / or some adaptation to palliate to the impacts identified by the results of the study R5 "The results are triangulated with one or several studies. None of the author from the assessment study is an author or co-author of a study used for triangulation of the results" # Article full reference 27 Majone, B., F. Villa, R. Deidda and A. Bellin (2015). "Impact of climate change and water use policies on hydropower potential in the south-eastern Alpine region." Science of the Total Environment. 28 Maran, S., M. Volonterio and L. Gaudard (2014 Mimikou, M. A. and E. A. Baltas (1997). "Climate change impacts on the reliability of hydroelectric energy production." Hydrological Sciences Journal 42 (5): 661-678. 31 Naughton, M., R. C. Darton and F. Fung (2012). "Could climate change limit water availability for coal-fired electricity generation with carbon capture and storage? A UK case study." Energy and Environment 23 (2)  Results of the global mean warming -regional climatescaling scaling methodology The future local scale meteorological time seriesnamely daily mean precipitation and temperatureare generated by perturbing the observed series for a control period according to the method of Shabalova et al (2003). In this method, the perturbation of local scale precipitation and temperature is based on the corresponding regional scale outputs of a Regional Climate Model (RCM) for the same control and future period. 2070-2099 1961-1990 Hydro The authors develop a model of PV power generation based on a) the change in global radiation and b) the averaging due to the distribution of orientations and the tilt angles of PV modules within a region.
Wind: Use of an Enercon E40 wind turbine with a rated power of 500 kW, a cut-in wind speed of 2,5 m/s and a rated wind speed of 13,0 m/s.

Appendix C-Peer-reviewed articles included in the systematic review but excluded from the analysis
Results from the articles focusing on bioenergy, wave energy and electricity networks were not included in the analysis because of the limited and conflicting evidence base they provided. Only four articles examine the impacts of CV&C on electricity generation from bioenergy (# 4, 9, 45, 46). They model the yields of different bioenergy crops in future climate conditions. No consistent patterns of impacts of CV&C could be extrapolated from the results of these four articles.
Two articles focus on electricity generation from wave energy. The first article (#21) quantifies how changes in the mean wind speed (a proxy for climate change) influence electricity generation by a Wave Energy Converter (WEC) in Western Scotland (UK). Harrison and Wallace (2005) demonstrate that under fixed conditions, WEC generation changes by up to 800 MWh/year (42%) for a 20% wind change. The second article (Reeve, Chen et al. (2011); #38) assesses the impacts of CV&C on generation by the Wave Hub WEC in Cornwall (UK). Although generation is projected to decrease by 2-3% under the A1B and B1 emissions scenarios for 2061-2100, this could be mainly due to the low efficiency of generation from steeper waves by the examined WEC (Reeve, Chen et al. 2011).
A single article examines the impacts of CV&C on electricity networks (#29). McColl, Palin et al. (2012) first formalise the current relationships between five types of weather-related faults and weather, and then use climate projections from a Regional Climate Model (RCM) to quantitatively assess how fault frequency could change in the 2020s-2080s. Their results suggest that lightning and solar heat faults are likely to increase but snow, sleet and blizzard (SSB) faults are likely to decrease (McColl, Palin et al. 2012). There are uncertainties regarding future wind, gale and flooding related faults.
The two articles on wave energy and the one on energy networks do not provide sufficient evidence to enable the identification of consistent patterns of impacts of CV&C. They also have limited spatial foci and thus limited value from a European perspective. For these reasons they were excluded from further analysis.