Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA

STOTEN-21577; No of Pages 13 Science of the Total Environment xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA Catherine E. Raptis ⁎, Justin M. Boucher, Stephan Pfister Ecological Systems Design Group, Institute of Environmental Engineering, ETH Zurich, Zurich 8093, Switzerland

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

• The impact of heat emissions from power plants were studied within the LCA framework. • The highest polluters and thermally polluted watersheds were identified. • The Mississippi, Danube and Yangtze basins are among the most impacted. • Characterization factors for use in LCA & other high resolution data are provided. • These data can be used in further analyses or environmental mitigation strategies.

a r t i c l e

i n f o

Article history: Received 19 October 2016 Received in revised form 7 December 2016 Accepted 8 December 2016 Available online xxxx Editor: D. Barcelo Keywords: Electricity generation Heat emissions Thermal pollution Impact assessment LCA Characterization factors

a b s t r a c t Freshwater heat emissions from power plants with once-through cooling systems constitute one of many environmental pressures related to the thermoelectric power industry. The objective of this work was to obtain high resolution, operational characterization factors (CF) for the impact of heat emissions on ecosystem quality, and carry out a comprehensive, spatially, temporally and technologically differentiated damage-based environmental assessment of global freshwater thermal pollution. The aggregation of CFs on a watershed level results in 12.5% lower annual impacts globally and even smaller differences for the most crucial watersheds and months, so watershed level CFs are recommended when the exact emission site within the basin is unknown. Long-range impacts account for almost 90% of the total global impacts. The Great Lakes, several Mississippi subbasins, the Danube, and the Yangtze are among the most thermally impacted watersheds globally, receiving heat emissions from predominantly coal-fuelled and nuclear power plants. Globally, over 80% of the global annual impacts come from power plants constructed during or before the 1980s. While the impact-weighted mean age of the power plants in the Mississippi ranges from 38 to 51 years, in Chinese watersheds including the Yangtze, the equivalent range is only 15 to 22 years, reflecting a stark contrast in thermal pollution mitigation approaches. With relatively high shares of total capacity from power plants with once-through freshwater cooling, and tracing a large part of the Danube, 1 kWh of net electricity mix is the most impactful in Hungary, Bulgaria and Serbia. Monthly CFs are provided on a grid cell level and on a watershed level for use in Life Cycle Assessment. The impacts per generating unit are also provided, as part of our effort to make available a global dataset of thermoelectric power plant emissions and impacts. © 2016 Elsevier B.V. All rights reserved.

⁎ Corresponding author. E-mail address: [email protected] (C.E. Raptis).

http://dx.doi.org/10.1016/j.scitotenv.2016.12.056 0048-9697/© 2016 Elsevier B.V. All rights reserved.

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

2

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

1. Introduction The pressures on the environment caused by electricity generation are manifold, with local or global outreach, and quantifying the associated emissions (to air/water) and water consumption, along with their ensuing impacts, has been the focus of numerous studies (for example (Feeley et al., 2008; Laurent and Espinosa, 2015; Macknick et al., 2012; Madden et al., 2013; Masanet et al., 2013; Meldrum et al., 2013; Miara and Vörösmarty, 2013; Miara et al., 2013; Peredo-Alvarez et al., 2016; Spang et al., 2014; Stewart et al., 2013; Vassolo and Döll, 2005; Zhao et al., 2008)). In view of climate change, which is generally expected to exacerbate the impacts of power production (Förster and Lilliestam, 2010; Fricko et al., 2016), the need to achieve global coverage on related analyses becomes even more pressing. At the same time, the ability to quantify the power production-related impacts on a high resolution remains pertinent, given the often highly regionalized nature of these impacts (Mutel et al., 2012). Efforts towards global coverage have increased in recent years, although, with the exception of few studies with high resolution such as (Vassolo and Döll, 2005), most investigations on the impacts of global electricity production present a coarse geographical differentiation, involving either countries or greater, multi-country aggregated regions (Fricko et al., 2016; Mekonnen et al., 2015; Pfister et al., 2011). Recently, we have begun a concerted effort to create a global inventory of power plants emissions for use in all kinds of environmental impact assessment studies (Raptis and Pfister, 2016). Our aim is to build a comprehensive, georeferenced dataset by modelling emissions to air and water, as well as water consumption on a generating unit level, which, among other things, will enable a near complete, bottom-up regionalized assessment of the impacts of global power production. We have commenced this effort by focusing on the thermoelectric sector, and specifically, on modelling the heat emissions from power plants that lead to the thermal pollution of freshwater bodies (Raptis and Pfister, 2016). The thermal pollution of freshwater bodies from power generation (one of the principal causes for this type of pollution (Hester and Doyle, 2011)) is brought about by steam-electric power plants employing once-through cooling systems. In contrast with wet tower cooling, where almost all the heat absorbed during the steam cycle is removed via evaporation and dissipated into the atmosphere, oncethrough cooling involves the direct rejection of the heat back into the water body. This thermal effluent can lead to significant water temperature increases, the adverse effects of which have been reported in multiple publications (Bush et al., 1974; Caissie, 2006; Coutant and Brook, 1973; Davidson and Bradshaw, 1967; de Vries et al., 2008; Hester and Doyle, 2011; Langford, 1990; Souchon and Tissot, 2012). Along rivers, the situation is compounded by the presence of many power plants within a certain reach (Madden et al., 2013). In order to protect aquatic ecosystems, there are regulations in place in both the United States and Europe, which impose thresholds on surface water temperatures (European Parliament and Council of the European Union, 2006; U.S. EPA, 2014). Several U.S. states impose an upper temperature threshold of 32 °C, which is often exceeded (Madden et al., 2013). In countries affected by the European Freshwater Fish Directive, downstream temperatures from the point of heat discharge should not exceed 21.5 °C and 28 °C (or 1.5 °C and 3 °C above the natural water temperature) in salmonid and cyprinid waters, respectively (European Parliament and Council of the European Union, 2006). According to a recent study, however, as a result of thermoelectric power cooling water emissions, over one third of the total flow of the Rhine experiences an average river temperature increase ≥ 5 °C above the natural temperature throughout the entire year, and the water temperature increase equals or exceeds 3 °C over significant stretches of the Danube and the Mississippi rivers (Raptis et al., 2016). Steps have been taken to evaluate the environmental impacts of freshwater thermal emissions also within the framework of Life Cycle

Assessment (LCA) (Pfister and Suh, 2015; Verones et al., 2011, 2010). LCA is a powerful decision support tool that quantifies the environmental impacts of products and processes throughout their entire life cycle (Finnveden et al., 2009; Hellweg and Mila i Canals, 2014; International Organization for Standardization, 2006, 2005). This is achieved by relating the emissions and resource inputs of each process in each life stage to their impacts via characterization factors (CF). CFs are calculated by modelling the relevant emission-to-impact pathway and express via appropriate indicators the potential impact on ecosystem quality or human health that one unit emission results in. LCA permits the identification of improvement strategies and its strength lies in its holistic approach, not only with regard to the entire value chain, but also with regard to the different incurred impacts considered: by quantifying a multitude of environmental impacts, decision makers can avoid burden shifting from one impact to another. In recent years, advances in two fronts have greatly improved the potential of LCA, namely the inclusion of impact categories for a wider range of emissions or resource uses and their assessment in a regionalized manner e.g. (Azevedo et al., 2013; Chaudhary et al., 2015; Cucurachi and Heijungs, 2014; de Baan et al., 2013; Helmes et al., 2012; Núñez et al., 2013; Pfister et al., 2009; Verones et al., 2013). Regionalized assessments provide a more accurate view of the environmental impacts and their spatial distribution, and, given the international nature of value chains, most regionalized impact assessment methodologies in LCA strive to achieve global coverage. Despite the advances in regionalized impact assessment methodologies, these have not always been matched by advances in regionalized emission inventories. The global regionalized heat emission data modelled by (Raptis and Pfister, 2016) have the potential to be combined with a global regionalized impact assessment method for freshwater thermal pollution. In the first of the two aforementioned studies that have brought thermal pollution into the LCA spotlight, CFs were calculated for thermal emissions along the Rhine only, and were tested in a case study with thermal emissions from a single nuclear power plant in Switzerland (Verones et al., 2011, 2010). In the second study, a more broadly applicable impact assessment model was developed, capable of producing spatially explicit CFs for river systems in the United States, and was applied in a case study for generic state-of-the-art gas and coal power plants (Pfister and Suh, 2015). In view of the importance of identifying global hotspots of freshwater thermal pollution and the value for environmental decision making of doing so within the framework of LCA, the goal of this study is after a) refining the temporal resolution of the mean heat rejection rates of Raptis and Pfister (2016) from yearly to monthly, b) improving the impact assessment model of Pfister and Suh (2015) and extending its coverage to global scale, to c) combine the updated inventory of heat emissions with the updated impact assessment model to obtain operational characterization factors and a comprehensive global environmental impact assessment of thermoelectric power cooling water emissions, which is spatially, temporally, and technologically differentiated. 2. Methods 2.1. Temporal update of global thermal emission data Our ongoing power plant emission inventory work expands on the Platts UDI WEPP (World Electric Power Plants) dataset, version March 2012 (Platts, 2012). Raptis and Pfister (2016) georeferenced and populated the relevant data gaps of the global thermoelectric sector, including cooling system information and thermodynamic steam conditions, achieving a respective coverage of ~92% and 100% in terms of the total installed capacity of all operational thermoelectric facilities as reported in the WEPP dataset. Based on these and localized environmental data, the authors systematically solved the thermodynamic cycle for every generating unit worldwide, differentiating between simple, reheat and cogenerative Rankine cycles. As an outcome of this analysis, the cycle

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

efficiency was obtained for every unit worldwide, and, for those units with once-through freshwater cooling systems (~2400), the maximum heat rejection rate was also obtained as a direct output. Marine or estuarine thermal emissions were not considered, since the focus was on freshwater thermal pollution. Moreover, heat emissions into artificially constructed cooling ponds were also excluded, in that these are generally tightly controlled purpose-built arrangements (Raptis and Pfister, 2016). Raptis and Pfister (2016) also provide mean annual thermal emission rates for all relevant generating units (Fig. S1 in the Supporting information (SI)), calculated by multiplying the maxima by mean annual capacity factors (ranging from 0 to 1) per fuel group for 2011 based on Energy Information Authority (EIA) data (US Energy Information Administration, 2011). The authors used data for 2011 instead of 2012 because the total installed electrical capacity calculated from the EIA 2011 dataset was closer in value to that of the WEPP March 2012 version. To better represent the temporal variation of electricity demand, and hence of the freshwater thermal emissions, due to climatic conditions and societal needs, where possible, we calculated mean monthly capacity factors per fuel group in this work. We retrieved monthly net electricity supply and net electrical capacity data from the IEA (International Energy Agency, 2011), and calculated monthly capacity factors by taking the ratio of net monthly electricity supply over the theoretical maximum net generation in 2011. The supply and capacity data provided by the IEA have a coarse fuel category differentiation, only two of which were applicable in this study: nuclear and combustible fuels. Moreover, monthly resolution data are only provided for OECD countries. In view of these limitations, we calculated mean monthly capacity factors for all countries and fuel categories available (~50% of the total units), and for the rest, we took the mean annual capacity factors reported in Raptis and Pfister (2016) and assumed them to be constant over the year. We then scaled the maximum thermal emission rates reported in Raptis and Pfister (2016) by multiplying with the updated capacity factors to produce mean monthly heat rejection rates for each generating unit with once-through freshwater cooling. Subsequently, we calculated the total monthly heat emissions and aggregated them on a 0.5° × 0.5° grid cell resolution. 2.2. Update of impact assessment model Pfister and Suh (2015) developed a regionalized (0.5° × 0.5°) fate model for heat emissions and combined it with an effect model based on Verones et al. (2010, 2011) to calculate freshwater thermal pollution CFs for the United States. The two-step model estimates the fate and effect of thermal emissions on a short- (grid cell of release) and longrange scale. The fate factor describes the temperature (heat concentration) increase in the system (a volume of water) during the residence time of a heat emission. The short-range fate factor FFshort range accounts for fate of the heat emissions within the grid cell of release, whereby nonlinear effects are considered. Its calculation depends on the change in temperature caused by the heat emission, the local residence time, the flow rate, and the inverse heat emission rate. The residence time in the grid cell of release is calculated by the average flow distance within the grid cell of release (~50 km) divided by the flow velocity. The longrange fate factor FFlong range accounts for the fate of the heat emissions in the downstream section of the watershed, where the temperature change is considerably lower than in the grid cell of release. Its calculation depends on the remaining residence time in the watershed and the average share of energy in the water when it reaches the sea with respect to the point of release. The units of the fate factor are °C m3 yr MJ−1. The fate model by Pfister and Suh (2015) was only applicable to river systems. In this work we extended the fate model to include the larger modelled residence time of emissions in lakes, based on flow velocity studies of lake-river systems in Switzerland. The reported flow velocity in Lake Constance was compared with the flow

3

velocity of the inflowing river Rhine, and the same was done with Lake Geneva and the river Rhone. The average retardation of velocity in lakes between these two lake-river systems was a factor of 12, and this correction factor was used to calculate the residence-time equivalent distance to sea across lake-river systems globally (see Section S2.2 of the SI for further details). The effect model estimates the potentially disappeared fraction (PDF) of species as a function of water temperature increase, based on species sensitivity distributions (SSD) for temperature induced mortality of aquatic species. PDF is the terminology used in LCA equivalent to PAF (potentially affected fraction) from ecotoxicology for the effect of chemicals on species. Previously available SSDs were applicable for species in temperate and subtropical climate zones (de Vries et al., 2008). Here, we also estimated SSDs for the tropical and boreal climate zones, allowing for a better geographical representation of the effects of increased temperatures on freshwater species. A linear regression of the temperature at which 50% of the species could potentially disappear (T50%) [taken from the SSD curves developed by (de Vries et al., 2008)] versus the average summer river water temperatures in the temperate and subtropical zones (Fig. S5), enabled the estimation of T50% in the tropical and boreal climate zones, using the average summer river water temperatures in these two extra zones as predictors. Under the assumption that the new SSDs follow the already existing SSDs for temperature induced mortality, these were then shifted until the estimated T50% values for the tropical and boreal climate zones were reached. Section S2.2 of the SI provides more detail. The short-range effect factor EFshort range is found by the slope of the line formed on the appropriate SSD curve by connecting the background, naturalized water temperature and the temperature to which the water is heated upon receiving the heat emissions (Fig. S3). The long-range effect factor EFlong range is also calculated by the slope between the background and the heat temperatures on the appropriate SSD curve, but the heated temperature is assumed to be only marginally higher than the naturalized water temperature (dT of 0.5 °C). See the SI for further details. The units of the effect factor are PDF °C−1. For every 0.5° × 0.5° grid cell where heat emissions were present, inputs to the final model included the mean monthly thermal emission rates, mean monthly discharge values as well as mean monthly naturalized water temperatures estimates (averaged over the period 1971– 2000) (van Vliet et al., 2012a). Monthly global gridded freshwater thermal pollution CFs (PDF m3 yr MJ−1) were obtained according to Eq. (1): CF i; j ¼ FF short range;i; j  EF short range;i; j þ FF long range;i; j  EF long range;i; j

ð1Þ

Freshwater thermal pollution impacts (in units of PDF yr m3) were then obtained by combining the heat emissions with the corresponding CFs according to Eq. (2): Impacti; j ¼ Emissionsi; j  CF i; j

ð2Þ

for every grid cell i (where thermal emissions were present) and for every month j. 2.3. Reducing the error due to the model resolution. Global gridded data have inherent uncertainty depending on their resolution, and the combination of different sets of spatial data in a model has the potential to amplify this uncertainty. This is exemplified by Fig. 1a.i: Raptis and Pfister (2016) provided exact coordinates for the power plants with once-through cooling, and, evidently, the gas power plant pointed at by the red arrow lies along the Mississippi river. However, the underlying grid-based discharge data (example for July) do not follow the exact path of the river at this point. Consequently, when the thermal emissions from this plant are also aggregated, they are allocated to a grid cell that appears to be ‘outside’ the main Mississippi flow, with a discharge that is over 300 times lower. The expected effect of

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

4

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

Fig. 1. a.i) The emissions from a single power plant along the Mississippi (pointed at by red arrow) are misallocated upon aggregation to a grid cell ‘outside’ the main river flow with over 300 times lower discharge (pointed at by black circle and arrow); the resulting original CF is much higher than the CF in all grid cells surrounding the original one (b.i); taking the geometric mean of the CF in all 9 indicated grid cells attenuates the effect of overly high or overly low CFs resulting from such misallocation (b.i). The dashed arrows (a.i, b.i) point to the grid cell that the Mississippi power plant emissions should have been allocated to (at least in terms of background river flow). a.ii) The emissions from two power plants along the Rhine (pointed at by red arrow) are correctly allocated to the grid cell with the underlying main river flow (pointed at by black circle and arrow); the resulting original CF and geometric mean CF are pretty close (b.ii). b) The monthly variation of thermal emissions, CFs, and impacts for the two examples of misallocated (i) and correctly allocated (ii) power plants upon aggregation to 0.5° × 0.5° grid cell resolution. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

such a misallocation would be an overly high CF. Analogous instances of emissions being allocated to higher discharge streams, leading to lower CFs than expected, can also be envisaged with the given 0.5° × 0.5° aggregation level. In order to attenuate the effect of extremes in either direction, for every grid cell where the thermal emissions were originally allocated (black points in the centre of cells, Fig. 1a.i–ii), we also calculated the CF for the 8 surrounding cells. We then calculated the geometric mean CF and geometric standard deviation from the set of 9, comprising the original and surrounding cells. The geometric mean and geometric standard deviation characterize the log-normal distribution, which is the most appropriate model for environmental data, where the dominant interactions are multiplicative (Limpert et al., 2001). It was also found to fit the data much better than the normal distribution in each CF set.

The log-normal distribution is a positive, left-skewed distribution, where the corresponding interval of μ ± σ (containing ~ 68% of the data) and μ ±2σ (containing ~95% of the data) for the normal distribution is found by multiplication and division: μ⁎ × ÷σ⁎ and μ⁎ × ÷ (σ⁎)2, respectively, where μ⁎ is the geometric mean (scale parameter) and σ⁎ is geometric standard deviation (shape parameter). The log-normal distribution is self-replicating under multiplication by a constant, whereby μ⁎ changes accordingly, but σ⁎ remains the same. σ⁎is a dimensionless measure of dispersion and, analogously to the coefficient of variation of the normal distribution, can be used to compare the spread between different distributions(Limpert et al., 2001). Referred to from now on are the estimators for μ⁎ and σ⁎ that we calculated, namely x and s⁎. For every grid cell with thermal emissions then, we calculated two CFs: one according to the original grid cell values, and a second by taking

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

the geometric mean of that and of the 8 surrounding grid cells. Section S3.0 of the SI elaborates on why the latter is the most suitable. In every following reference, therefore, the grid cell level CFs adopted and assumed are those calculated via the geometric mean of the original and surrounding cells. We subsequently aggregated the CFs on a watershed level, where both the original and geometric CFs were again found to be lognormally distributed, so x and s⁎ were appropriate measures for this level of aggregation. Watershed boundaries were taken according to (Alcamo et al., 2003). The relevant 165 watersheds (with the exception of those in Australia and New Zealand) are shown in Figs. S6 and S7. Watershed names (where available) reflect either the actual basin name or the name of the largest river within it. 3. Results

5

residence time of the emitted heat in the freshwater body, has a major influence on the CF. In that respect, the long-range component of the CF tends dominate the overall CF when the grid cell of emission is at a large distance from the sea (Fig. S14). The influence of the fate factor on the CF can be observed in the coastal area of the United States as well as along the Yangtze river (Figs. S11b.ii and S13b.ii), where coastal emission sites end up with a clearly lower CF compared to sites more inland. Evidently, the distance to the sea is not the major contributing factor in all cases, as can be seen in the plots of CF as a function of distance to the sea for the Mississippi and the Yangtze (Fig. S15). The effect factor comes into play and has a greater influence on the CF in locations where the water temperature is cold, e.g. in mountainous areas: any extra heat added to the system in these cases will be at the bottom end of the SSD curve, where the gradient is low, resulting in a smaller effect factor and bringing down the CF as a result (Fig. S3).

3.1. From grid cell to watershed level: characterization factors and impacts 3.1.1. Grid cell level CFs and impacts Over 40 times more aggregated emission sites (0.5° × 0.5° grid cells) were present in the northern hemisphere, hence these give a more complete picture of the results. In the northern hemisphere the lowest and highest freshwater thermal pollution CFs on a grid cell level are observed in January and July, respectively (Fig. 2b), whereas in the southern hemisphere the extremes occur in the same months, but reversed (Fig. S9a). The northern hemisphere median CF in July is almost 3 orders of magnitude larger than the equivalent January CF, a stark difference which is also observed on a grid-by-grid comparison, whereby over 50% of the grid cells globally have N3 order of magnitude difference between the extreme warm and cold month CFs, in 1/5th of which this difference goes over 4 orders of magnitude. This monthly variation in CFs is attributed to the effect factor, and the sigmoidal shape of the SSD curves used to calculate it (Fig. S3): as the background water temperature increases, so does the slope of the SSD curve of the particular climate zone, and any additional temperature increase due to power plant thermal emissions will result in an increased temperature derivative, so long as the background water temperature is not overly high, at which point the SSD curve plateaus and any extra heat would not result in a high effect (Fig. S3). Fig. 2c presents the variation of the freshwater thermal pollution impacts over the months. The median difference between July and January impacts on a grid-by-grid basis is also almost 3 orders of magnitude in the northern hemisphere. Given that the variation in thermal emissions is small by comparison (the extremes are b1.5 times different, globally, though larger in some of the OECD countries where the emission variation was modelled, see Fig. S10), it is clear that the impact variation is driven mainly by the CFs. Figs. S11–S13 showcase the entire analysis sequence for January and July, with maps of the major affected regions depicting on a grid cell level the thermal emissions, the CFs, and the resulting impacts. Pfister and Suh (2015) reported that the fate factor, in other words, the

3.1.2. Watershed level CFs, impacts and uncertainties Fig. 3a presents the freshwater thermal pollution CFs for January and July aggregated on a watershed level, calculated by taking the geometric mean of all grid cell level CFs in each basin. Fig. 3b presents the thermal pollution impacts per watershed, calculated by multiplying the watershed level CFs (Fig. 3a) by the sum of the heat emissions per watershed (Fig. S16). The watersheds of central-eastern United States and Europe together with the Yangtze stand out with their high summer CFs (Fig. 3a.ii). In over 60% of the watersheds worldwide, the CF during the warm months is over 2 orders of magnitude higher than that during the cold months, and in half of those this difference is over 3 orders of magnitude. The stark difference in CFs between warm and cold months is preserved when aggregating from a grid cell (Fig. S9) to a watershed level, though it is slightly reduced. When it comes to the impacts (Fig. 3b), the watersheds standing out include those with high CFs (Fig. 3a), but also ones where the heat emissions play a larger role in the final impact. Fig. 4 allows us to get a better idea of the relative influence of emissions inventory and CF on the final impact. In this scatterplot of impacts versus CFs, each point represents a watershed and is scaled by the total emissions for the given month. As an example, the Danube and the Yuan watersheds have more or less the same CF in July, but the Danube receives over 3.5 times more thermal emissions, resulting in a considerably higher impact. On the other hand nearly the same amount of heat enters both the Rhone and the Danube, but this heat remains in the shorter Rhone for much less than they do in the Danube, resulting in a CF and impact that are over 5 times smaller. For further such comparisons refer to Figs. S18, where the watersheds are labelled, and S17, where the emissions, CFs, and impacts are explicitly represented in bar charts for the most impacted watersheds worldwide. The points in Fig. 4 are colour-coded according to the geometric standard deviation s⁎ of the CF and of the impacts (these are equal because upon multiplication of the lognormal distribution by a constant (here, the emissions) s⁎ remains the same). The error bars extend from x  s to x  s and contain ~68% of the data. Watersheds with

Fig. 2. Tukey boxplots for northern hemisphere monthly grid cell level freshwater a) heat emissions (only for OECD countries here; for a complete country breakdown see Fig. S10), b) thermal pollution CFs, and c) thermal pollution impacts.

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

6

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

Fig. 3. Freshwater thermal pollution CFs (a) and impacts (b) on a watershed level for January (i) and July (ii). The impacts shown in (b) were calculated by multiplying the sum of the thermal emissions per watershed (Fig. S16) with the watershed level CFs (a).

no colour and error bars are ones with only one emission site, for which the variability in CF and impacts is currently unknown. Being dimensionless and multiplicative, s⁎ provides a means of comparing the variability between distributions, and in this case allows a comparison not only between watersheds, but also between months: in July, for CFs and impacts alike, ~70% of the watersheds with N1 emission site have a geometric standard deviation between 1 and 2 (overall median: 1.6), whereas in January the respective share of watersheds is 45% (overall median 2.2). The dispersion factor k, quoted as an alternative variability measure (Slob, 1994), indicates the extent that a variable X  deviates from x : Prob {xk bXbkx } ~ 0.95, i.e. k = (s⁎)2. Fig. S20 presents the same plot as Fig. 4 for January and July but with watersheds colourcoded according to the k dispersion factor and with error bars covering ~95% of the data. The median dispersion factor for the CF and the impacts is 5.0 in January and 2.6 in July. The above analysis and Figs. 4, S17 and S19 indicate that the relative variability of the CFs within a given watershed is lowest during the warm months when their magnitude is highest (Figs. S9b and 3a). Fig. S21 presents these differences on global maps, and also reveals the influence on s⁎ of the number of emissions sites per watershed: basins with 5 or less emission sites tend to have a high CF variability, whether this is in January or July. In absolute terms, however, due to the multiplicative nature of s⁎, the variability of the watershed CFs in the warm months is greater (Fig. S17) and dominates the annual variability.

Fig. 4. For July, the scatterplot of watershed level impacts versus watershed level CFs, scaled by the thermal emissions per basin and colour-coded according to the geometric standard deviation. The error bars extending from the points cover the interval [x  s ; x s ] containing ~68% of the potential values along both axes. Note the logarithmic scale on both axes. Watersheds with no colour have only one emission site, so no tolerance interval could be calculated for these. See Fig. S18 for the same scatterplot with watershed labels, Fig. S19 for a comparison of the same plot in January, and S20 for the same plot with 95% tolerance intervals.

3.1.3. Grid cell vs. watershed level impacts In order to establish a degree of confidence in the global watershed level CFs, we compared the impacts from the watershed level approach with impacts calculated from a grid cell level approach. Fig. 5a reveals maps for January and July of the percentage relative difference in the final basin impacts between using watershed level emissions inventory and CFs, and grid cell level inventory and CFs. The highest relative differences globally are observed in the colder months (October–March; Fig. 5b), which are also the months during which the lowest CFs occur (Fig. S9), and vice-verse for the warmer months April–September (northern hemisphere watersheds dominate the results). Specifically, the median relative difference between the watershed and grid cell level

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

7

Fig. 5. a) For i) January and ii) July the percentage relative difference between freshwater thermal pollution impacts calculated by a combination of watershed level emissions and CFs & level impact−Grid cell level impacts summed over watershedÞ . b) Tukey boxplots of the aforementioned grid cell level emissions and CFs (impacts subsequently summed per watershed), namely 100ðWatershedGrid cell level impacts summed over watershed

percentage relative difference per month over all watersheds with N1 emission sites. c) Scatter plot of aforementioned percentage relative difference per watershed as a function of the variability in the estimated watershed level CF (s⁎) for all basins with N1 emission sites and all months (January and July highlighted). d) The monthly variation of global thermal pollution impacts as calculated via a grid cell and watershed level approach.

approaches varies over all watersheds (with N1 emission site) from −30% to −3% from January to July, with the interquartile range in January being over twice as large as that in July (Fig. 5b). It is unsurprising that the relative difference in impacts is overwhelmingly negative (though this is not always the case, see Section S3.1.3 of the SI). That is, the watershed level approach results in lower impacts than the grid cell level assessment (skewed boxplots in Fig. 5b): assuming a more or less uniform spread of thermal emissions among the watershed, as previously mentioned, the geometric mean used to aggregate all grid cell level CFs in each watershed, ‘favours’ smaller values by not being swayed by higher ones. A high variability in thermal emissions and/or CFs within a given watershed leads to greater relative differences between the two methods of calculating the impact. The left-pointing arrowhead-like shape of the plot in Fig. 5c demonstrates this effect: the higher the CF variability, the greater the deviations in the impact calculations in either way, but more pronounced in the negative direction. While the relative differences between the two approaches are smaller during the months when the impacts are highest (compare Fig. 5b and d), in absolute terms the difference during these months is greatly responsible for the divergence of the impacts in an annual assessment. Fig. 5d shows that the monthly variation in impacts is similar according to both approaches, with the warm months of June–August being responsible for ~ 70% of the annual impact in both cases (May– September: N 90% in both cases). Nevertheless, the watershed level approach results in a total of 12.5% lower annual impacts (Fig. S22) compared to the grid cell approach, with half of this difference coming from June–August alone. Overall, it is evident from the analysis in this and the previous section that when it comes to aggregating on a watershed level: a) the relative variability of the CFs within a watershed is lowest during the warm months when their magnitude is highest and for the most crucial basins, namely those with the highest CFs and emissions (Figs. 4 and S19); b) the relative divergence in watershed level impacts from the respective summed grid cell impacts is also lowest during the warm months,

when their magnitude is highest (Fig 5); but c) in absolute terms the CF variability and the difference in impacts are much more important during the warm months and dominate the annual results (Figs. 5, S17, S22). Based on these findings, for a global assessment of the impacts of thermal emissions we recommend in the following order the use of a) grid cell level CFs, when the thermal emission sites are known in detail, b) watershed level CFs, when no exact location is known within the watershed, and in either case to ensure the emissions for the warm months (May–September in the northern hemisphere) are included in yearly assessments, since the CFs during these months are orders of magnitude higher than those during the rest of the year (Fig. 2b, S9), a variation which is much more influential than that of the thermal emissions (Fig. 2a). 3.2. Global and regionalized allocation of impacts to power plant age and fuel Tracing the freshwater thermal pollution impacts back to the responsible generating units shows that they are dominated by coal and nuclear power plants globally (Fig. 6a, c) and over the entire year (Fig. S23). Raptis and Pfister (2016) found that the shares of the mean global annual thermal emission rates from coal and nuclear power plants were at ~40% each, yet the shares of the global annual impacts they are associated with amount to 52% and 37%, respectively (Fig. S24). On a monthly basis, the share of the impacts varies significantly. Capacity factors for nuclear power plants are typically high and constant. On the other hand, the operational flexibility of fossil fuel-driven power plants is exploited during periods of higher demand. As a result of these and other factors, more than half the total share of impacts is allocated to coal-fuelled power plants in July, but only a quarter in January (pie charts of Fig. 6a), during which month, however, the total impacts are significantly lower in absolute terms. Of the 10 most thermally polluted watersheds annually, 4 are subbasins of the Mississippi, 6 are located in the United States, 2 in Eastern Europe and 2 in Asia (Fig. 6a). The lead of the most thermally polluted

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

8

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

Fig. 6. a) Pie charts depicting the fuel-allocated shares of the global impacts in January and July. The bar chart disaggregates the annual impacts per watershed (for the 50 most thermally polluted ones globally) in terms of the fuel used (for fuel group breakdown refer to Raptis and Pfister (2016); hatched areas represent the short-range impacts. The numbers underneath the 10 most thermally polluted watersheds correspond to the % share of the short-range impacts. b) Tukey boxplots of the first year of commercial operation of the generating units with once-through cooling systems within each of the top 50 most thermally polluted watersheds; the boxplots are colour-coded according to the number of generating units contained and the arithmetic and impact-weighted means are overlaid for each watershed. c) The share of annual freshwater thermal pollution impacts (occurring in 2012) as a function of the decade during which the relevant power plants entered commercial operation. d) The variation over the months of the median freshwater thermal pollution impact per MJ electricity produced (by the relevant generating units with once-through cooling only), grouped according to the fuel group they pertain to. Fig. S27 presents all boxplots of the impacts per MJ electricity produced per fuel group and month.

basin can be explained by the fact that heat from power plants in the St. Lawrence watershed is released almost entirely into the Great Lakes. With the highest total heat emissions (Fig. S17a) and their high residence times in the lakes, it is no surprise that the impacts in the St. Lawrence watershed are higher than in all other watersheds. The hatched areas in Fig. 6a indicate the best estimates of the short-range impacts (these are only approximate because of the inability to linearly decouple the nonlinear CF, calculated as the geometric mean of the CF in the original grid cell of emission and in the 8 surrounding cells). For the top 10 most thermally polluted watersheds, the short-range impacts are responsible for 8%–17% of the total impacts, while globally this share is approximately 12%. These values demonstrate, again, the much larger influence of the long-range component of the CF, however it is important to note that there is a linear relationship between the short-range and total impact (Fig. S25), meaning that the conclusions drawn

regarding the ranking of thermally polluted watersheds would still generally be valid if only the short-range components were considered. Emissions from nuclear power plants are responsible for over half the total impacts in 11 of the top 20 most thermally polluted watersheds, whereas impacts due to gas-powered plants feature prominently in 10 out of the top 50 most thermally polluted watersheds (Figs. 6a and S26). In watersheds in China the impacts are caused almost entirely by coal-fuelled power stations, and, though much smaller in absolute terms, oil features greatly in two middle-eastern watersheds as the major source of freshwater thermal pollution impacts Figs. 6 and S26). Fig. 6b gives insight into the age of the generating units in the most affected watersheds. The impact-weighted mean first year of commercial operation for 30 out of the 50 most thermally polluted watersheds is between 1960 and the 1980. In the Mississippi subbasins, the impact-weighted mean first year of operation ranges from 1965 (Ohio –

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

Tennessee) to 1977 (Platte). Globally, over 80% of the annual impacts originate from power plants entering commercial operation during the 1980s or before (Fig. 6c). The decrease in total impacts in the following decades reflects a phasing out of once-through cooling systems in the United States and Europe (Raptis and Pfister, 2016). In some fast growing economies such as China, the construction of power plants with once-through cooling systems does not appear to have faced a similar phasing out (see cumulative capacity growth curves in (Raptis and Pfister, 2016)), and the impact-weighted first year of operation of the generating units within Chinese watersheds is clearly higher than that in all the rest (Fig. 6b). Notably, the impact-weighted mean first year of operation of generating units with once-through cooling in the Yangtze is the year 2000. Interestingly, while Raptis and Pfister (2016) reported the 1980s as the worst decade in terms of emissions, it is the 1970s that dominate the impacts (Fig. 6c), alone claiming 1/3 of the global impacts, a result which indirectly demonstrates regionalized impact assessment in action, since spatial patterns of emissions and CFs vary partially independently. The spread of values and the presence of many outliers per fuel group in the monthly boxplots of Fig. S27 show that the impact per MJ of electricity produced on a generating unit level is considerably variable, since it depends on many other aspects other than the fuel used at the particular station. Nevertheless, as a general trend, Fig. 6d reveals that on a generating unit level the median impact per unit of electrical energy produced for nuclear-fuelled units is larger than that of coal-, gas-, and oil-fuelled generating units, respectively, a trend which is maintained over all months. This reflects the fact that nuclear power plants have lower efficiencies than those fuelled by the other fuel groups, irrespective of the type of Rankine cycle employed (which also has a large influence on the efficiencies and is not reflected in this analysis) (Raptis and Pfister, 2016). 3.3. The varying impact of 1 kWh electricity production between countries The heat incorporated in the cooling water effluent is rarely reported for power plants with once-through cooling. At the moment, therefore,

9

applying the freshwater thermal pollution CFs calculated in this work at the impact assessment stage of LCA is somewhat problematic. In order to make freshwater thermal pollution impacts directly relatable to life cycle inventories, but also to gauge the relative thermal pollution impact of the production of 1 unit of electricity between different countries we carried out the following calculation: the grid cell impacts allocated to the generating units were summed per country and over the year; the annual country impacts were then divided by the total net electricity produced in that country over the entire year (data obtained from the EIA (US Energy Information Administration, 2011)). Both the bar chart and the map of Fig. 7 show the variation of the resulting freshwater thermal pollution impact per kWh electricity produced in countries where once-through freshwater cooling is employed (the bar chart includes 40/56 countries). The colour-coding of the bar chart reveals the share that thermoelectric power plants with once-through freshwater cooling occupy with respect to the entire country electricity production mix (in terms of installed capacity). In addition, shaded areas indicate the approximate contribution of the short-range component of the impact modelling, with a median share of 15% over all countries depicted. The scatterplot of Fig. S28 enables an even closer examination of the relative contribution of the short-range component with respect to the total impact per kWh electricity produced. 1 unit of electricity is most impactful in Hungary, followed by Bulgaria and Serbia, a fact that is unsurprising given that these countries trace a large part of the Danube, which is the 6th most thermally polluted watershed worldwide, and that the share of thermally polluting power plants in the three respective countries amounts to over 40%, 25% and 50% (Fig. 7). 6 of the top 10 countries with the most impactful unit of electricity are actually in Eastern Europe. There is a particularly high density of power plants with oncethrough cooling in the eastern United States and Canada, as well as in central Europe, and it is interesting to highlight some differences in the relative impact of 1 unit of electricity between neighbouring countries, in light of the interconnectivity of grids. In Canada and the United states the environmental damage of 1 kWh is almost the same. In Switzerland, on the other hand, the impact from the production of 1 kWh

Fig. 7. Bar chart and map of annual freshwater thermal pollution impacts per unit of net electricity produced per country – the entire electricity production mix of each country was considered. For clarity, only countries with impacts per kWh N 1.5 × 10−6 are shown in the bar chart. The bars are colour-coded according to the share that power plants with oncethrough freshwater cooling occupy with respect to the entire country electricity production mix; shaded areas correspond to the short-range component of the impact.

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

10

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

electricity is greater than in all its neighbouring countries (Figs. 7 and S28), and, specifically, 3 and 8 times as high compared to that in Germany and Italy, which can partly be explained by the fact that some of the largest rivers in these countries originate in the Swiss mountains and therefore the emissions from Swiss power plants have higher residence times (the short-range component of the impact is also higher, 2 and 5 times larger in Switzerland compared to Germany and Italy, respectively). Based on the allocation of the impacts to the different fuel categories, the damage caused by the production of 1 kWh electricity from nuclear, fossil fuels, biomass-waste and geothermal energy can also be assessed using further EIA data (which also dictates the fuel group aggregation). In this case, the annual impacts per fuel and country were divided by the annual total net electricity produced per fuel in the given country. The results for fossil fuels and nuclear power are shown in Fig. S29. Hungary and Bulgaria again occupy the top spots when it comes to the impact of 1 kWh of electricity by nuclear power, whereas Serbia has the highest impact per kWh of electricity by fossil, all of which provide some insight into the lead these countries hold in terms of impacts per 1 kWh of total electricity (c.f. Figs. 7 and S29). The freshwater thermal pollution damage from 1 kWh nuclear energy is over 3.5 times higher when produced in Canada compared to the United States. This difference can be explained by the fact that 94% of the thermal emissions from nuclear power plants in Canada are released into the Great Lakes (lake Ontario and lake Huron), compared to only 22% of heat from nuclear power plants in the United States entering lake Michigan and lake Ontario: the high residence time of the heat in the lakes compared to rivers closer to the coast leads to higher impacts. In fact the thermal emissions into the Great Lakes in Canada make up 80% of the total country freshwater thermal pollution impact. On the other hand, the damage from 1 kWh electricity from fossil fuelled power production is almost the same in the two neighbouring countries (Fig. S29). 4. Discussion 4.1. Appraisal of methodology: the LCA approach to thermal pollution This work constitutes the first damage-based global impact assessment of freshwater pollution from the thermoelectric power sector. Its great advantage is that is uses the most complete to-date global heat emission inventory calculated on a generating unit level. One of the limitations of this work is that it does not sufficiently reflect power plant adaptations, such as output reductions to prevent the overheating of freshwater bodies and comply with relevant legislation. Operational adjustments such as these can only be captured on a plant-by-plant basis, and were not accounted for in our work, at least not in their entirety since the mean monthly capacity factors used to calculate the heat emissions were, only where available, on a coarse country and fuel group level. Reports of decreased power plant operability during past heat waves are plentiful (see Cook et al., 2015; Förster and Lilliestam, 2010 and references therein), and the effect of elevated water temperature in combination with decreased river flow on thermoelectric power supply has been investigated by a multitude of researchers, including in the light of future increased energy demand and climate change estimations (e.g. Cook et al., 2015; Förster and Lilliestam, 2010; Koch and Vögele, 2009; van Vliet et al., 2012b). Our research does not cover this aspect of the water-energy nexus, however, its purpose is also a different one: by providing insight into potential impacts on freshwater aquatic biodiversity, our research highlights hotspots of thermal pollution globally and offers a suitable means of including these impacts within the framework of LCA. In terms of relevant LCA research, Pfister and Suh (2015) estimate that 95% of the CFs over all months and grid cells in the contiguous United States had an impact between 2.5 × 10−6 and 2.5 × 10−4 PDF m3yr MJ− 1. The impact assessment methodology was improved in our work resulting in the equivalent 2.5%–97.5% percentile range for the

contiguous United States over all months and grid cells becoming broader, extending more towards smaller values: (7.5 × 10−9 to 1.9 × 10−4 PDF m3 yr MJ−1). The upper level is very close, which is explained by the similarity of the locations and capacities of actual power plants at the higher end of the capacity spectrum, compared to fictional units all over the United States in Pfister and Suh (2015). Verones et al. report an average yearly CF of 7.2 × 10−6 PDF m3 yr MJ−1 for heat emissions from a single power plant in Switzerland. In our work, the annual average CF for the exact same grid cell is 11.6 × 10−6 PDF m3 yr MJ−1, which is a close enough estimate, considering the simplifications in the global model. It is evident from the analysis that the long-range component of the thermal pollution impacts dominates the results. At first glance this might appear counter-intuitive, with common notions of heat dilution dictating that any impacts other than the short-range one ought to be negligible. Nevertheless, the goal of LCA is to account for all impacts over space and time, and, in that respect, the impacts of diluted heat over large volumes of water are significant, especially when considering that there can be multiple power plants and other sources of heat emissions over the length of a river, resulting in the accumulation of downstream diluted thermal emissions. There is, however, considerable uncertainty in the long-range impacts due to the limitations of the applied equilibrium temperature modelling, especially concerning fate and effect in lakes and dams. For this reason we also provide the short-range CFs and impacts. 4.2. Uncertainties The model we used for the global scale assessment of the residence time of the heat in freshwater bodies is one-dimensional and thus neglects important mixing features occurring in reality, which can be better reflected in detailed models (Verones et al., 2011, Verones et al., 2010). Effects in lakes or dams, especially, are only roughly addressed. Our adjustment of residence time based on reduced velocity is uncertain due to the limited amount of data available and the highly specific behaviour of individual lakes. Stratification processes are particularly important, and these vary among lakes and over seasons. At this point, these specific aspects cannot be accounted for on a global level and our estimates of flow retardation and therefore increased residence time are a first approach to account for the influence on the fate factor. Additional heat exchange processes that can significantly affect residence time in specific cases (e.g. exchange of surface and groundwater) are neglected in our analysis too, adding to the uncertainty of the longrange and hence total CFs. Another major source of uncertainty can be found in the estimation of the SSD curves for the tropical and boreal climate zones, which were used in the effect factor. The size of the data sample used in the regression models for this purpose was particularly small. However, this was the best that could be done, in view of the limited data. Moreover, the new SSDs were constrained to follow the shape of the empirically developed subtropical and temperate zones, respectively, which is a simplification; the boreal zone, for instance, may actually be represented by a curve allowing for higher tolerance of aquatic species at low temperatures near the freezing point. Experimental thermal tolerance studies of species living within these zones could be conducted to improve the accuracy of the SSD curves. On a global scale, the influence of this simplified approach on the effect factor is not considered to be overly high, since, the limited number of thermally polluting power plants located within the tropical and boreal climates zones (41 and 18 out of 754, respectively), contribute only 3% and 0.3% of the global annual pollution impacts, respectively (for the temperate and subtropical climate zones, the respective shares are 86% and 10.7%). An additional point relating to the SSDs has to do with the fact that they were based on temperature tolerance limits; these, however, do not fully capture the effect of temperature change on the biological processes of aquatic species. Biological processes such as growth or reproduction are better

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

characterized by bell-shaped thermal performance curves, with minimum and maximum temperatures beyond which the process cannot proceed, and an optimum temperature related to a maximum process rate (Hester and Doyle, 2011). In addition, certain species exhibit a varying sensitivity to the thermal regime depending on their life stage, and this temporal species sensitivity is not captured in the effect factor (Souchon and Tissot, 2012). While the heat emissions were available on a generating unit level by Raptis and Pfister (2016), the minimum aggregation level on the inventory side had to be a 0.5° × 0.5° grid cell resolution, in order to match the impact assessment method. Uncertainties due to the spatial mismatch with background hydrological data were confronted by taking the geometric mean CF of the original and all surrounding grid cells, with the geometric standard deviation giving insight into the variability of each calculated value. While preserving the temporal dimension, for which the variability was shown to be considerably higher, monthly CFs were subsequently also aggregated on a watershed level. The rationale for the latter was that the impacts involve freshwater ecosystems, for which such an aggregation level, apart from making sense hydrologically, has been shown to be the most appropriate (Mutel et al., 2012). Using the watershed level CFs result in 12.5% lower global annual impacts compared to the grid cell level CFs, however, on a basin by basin level, deviations from the grid cell level approach in the final impacts were smallest for the most impactful months and watersheds. Therefore, in cases where the exact location of the heat emissions is unknown within the basin, these CFs provide a satisfactory alternative to estimating the thermal pollution impact, not least because this aggregation level coincides with that used for CFs of other impacts included in LCA, such as freshwater consumption, allowing geographically consistent assessment to be carried out in comprehensive future studies of power generation impacts. Further uncertainties are linked to the CFs and the impacts which have not been accounted for in this work, with a notable case being uncertainty in the heat emissions and their temporal variation. Approximately half of the global generating units (64% of gridded emission sites, 57% of the global heat emissions) were in OECD countries, permitting an estimation of the monthly variation of heat emissions, with mean annual heat emissions being applied for the rest. This limitation is likely to have manifested itself in the impact results, though the extent of this depends heavily on the local CF and environmental conditions. As an indication, however, 32% of the global annual freshwater thermal pollution impacts are calculated without accounting for monthly heat emission variations, 30% of which are located in China. A sensitivity analysis of the impact model inputs on the calculated characterization factor showed that the most sensitive input variable is the naturalized water temperature followed by the heat emissions. The sensitivity of the other inputs varies in the case of extreme combinations. For example, sites with a very high distance to sea have a higher sensitivity to the set average depth of the river, which helps to define heat release from the river water, since heat dissipation plays a larger role over longer distances. River depth begins to show a larger impact on the resulting CF when the distance to sea is increased. With a low distance to sea of 100 km and river depth of 5 m, an 80% decrease in the river depth decreases the CF by 4.7%. However, when the distance to sea is greatly increased to 10,000 km, an 80% decrease in river depth decreases the CF by as much as 78%. The same relationship can be seen for the sensitivity of the river heat exchange rate. When the heat exchange rate is decreased by 40% at 100 km to the sea, the CF increases by as much as 0.5%. When the distance to sea is increased to 1000 km, the same 40% decrease in heat exchange can increase the CF by as much as 58%. Further details are available in the Section S4.2 of the SI. While the sensitivites of these variables are relatively small in comparison to the magnitude of impact an uncertainty in the naturalized water temperature can have on the calculated CF, they are still worthy of consideration when reviewing modelled results.

11

4.3. Top polluters and polluted watersheds All results indicate that the majority of global freshwater thermal pollution impacts occurring today are caused by nuclear and coalfuelled power plants that have been in operation for over 30 years. Compared to nuclear power plants, coal-fuelled ones have lower impacts per generating unit, but they make up for this with their ubiquity worldwide, with new coal-fuelled units with once-through freshwater cooling entering commercial operation even in recent decades especially in countries with fast-growing economies. While the freshwater thermal pollution impacts per unit of electricity produced might point to nuclear as the worst polluter, it is important to keep in mind that a proper, comprehensive assessment of the impacts of power production on ecosystem quality should include a series of other impact categories. According to both Verones et al. (2010, 2011) and Pfister and Suh (2015) the impact of freshwater thermal pollution is small compared to other power production-related impacts on the aquatic environment, such as freshwater ecotoxicity, eutrophication and water consumption. As Pfister and Suh (2015) point out, however, in order to make a comparison between the different impact categories completely viable, these should be quantified in a consistent way. Recent approaches quantifying the impacts in terms of global species-equivalent losses rather than the approach employed in this work (potentially disappeared fraction of species in a given volume of freshwater) promise to make the assessment of impacts on ecosystem quality more systematic and comparable (Tendall et al., 2014; Verones et al., 2013). However, a full understanding of effects on ecosystem quality as a whole is still largely missing, and the focus in recent methodologies is on an extremely small set of species, which do not include the species used for developing the SSDs in this work. That being said, there are impacts from power-production that have not been quantified at all yet, with the most obvious example being nuclear power production, that often stands out as a climate friendly option in terms of its carbon footprint, but as yet features no major methodology accounting for the impact of accidents and the ensuing radioactive emissions, or for the long-term effects of nuclear waste storage. The regionalized assessment carried out in this work permits a detailed evaluation of the most affected regions, highlighting the Mississippi subbasins, central and eastern European watersheds, as well as ones in eastern China as especially thermally polluted due to cooling water emissions. The location of the emissions and the length of each affected river system are non-negligible, since the CF model is particularly sensitive to the distance to sea (Pfister and Suh, 2015). However, there is a limitation in the effect modelling in that it does not account for cumulative effect of thermal emissions along any river length. That is, when modelling the effect in a given grid cell, any upstream temperature increase due to emissions from preceding power plants is not accounted for. A proper, connected water temperature-hydrological model would be required to achieve this, such as the one applied in Raptis et al. (2016). 5. Conclusions Freshwater systems face a plethora of physical and chemical stressors that put pressure on their ecosystems, such as irrigation, dams, channelization and deforestation, as well as industrial and agricultural waterborne emissions (Brooker, 1985; Bunn and Arthington, 2002; Meybeck, 1989; Sweeney et al., 2004). Thermal pollution is but one possible physical stressor on freshwater ecosystems, yet the identification of areas of high thermal pollution is of high importance since when combined with chemical emissions it can potentially exacerbate any ecotoxicological effects (Heugens et al., 2002; Holmstrup et al., 2010). As such, this work provides valuable guidance to the watersheds that are already more vulnerable to the effect of chemical emissions. At the same time, there is still plenty of scope for extending this study to cover the impacts of thermal emissions in coastal waters, especially

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

12

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx

when our estimate of the total gross installed capacity of power plants with once-through brackish or sea water amounts to almost 30% of the total installed thermoelectric capacity as reported in the WEPP dataset. While the fate in the near future of nuclear and coal-fuelled power plants with once-through cooling systems that have been operating for N45 years already might be clear in countries where legislation has restricted the use of once-through cooling, such phasing out is not guaranteed to happen in all countries. The high impacts in Chinese watersheds from power plants built predominantly after the year 2000 points to a worsening situation in basins that are already highly stressed. Moreover, the authors of a recent study estimate the installed capacity of all thermoelectric power plants with once-through freshwater cooling in China as being 40% higher than the respective installed capacity identified from WEPP by Raptis and Pfister (2016) (116.1 GW vs. 68.5 GW, Zhang et al., 2016), meaning that the thermal pollution impacts could already be significantly higher than what we estimated in this work. Should the power plants with once-through cooling be phased out at some point in the future, the accompanying detailed data on a generating unit level provide a unique opportunity for future studies to quantify the impacts of replacement and avoid burden-shifting. As we continue our efforts to characterize all other environmental impacts associated with thermoelectric power production, the scope for such studies will become more complete. Data Monthly freshwater thermal pollution CFs and associated uncertainties are provided on multiple levels in the accompanying files: a) grid cell level (raster files and spreadsheet), b) watershed level (shapefile), c) country level (spreadsheet). Country level impacts per kWh net electricity mix, and monthly impacts per generating unit are also provided (spreadsheets). Acknowledgements We would like to thank the anonymous reviewers for their insightful comments and suggestions, which improved the quality of the article. We would also like to thank Stefanie Hellweg for her useful feedback and Engie for partially funding this work. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.scitotenv.2016.12.056. References Alcamo, J., Döll, P., Henrichs, T., Kaspar, F., Lehner, B., Rösch, T., Siebert, S., 2003. Development and testing of the WaterGAP 2 global model of water use and availability. Hydrol. Sci. J. 48:317–337. http://dx.doi.org/10.1623/hysj.48.3.317.45290. Azevedo, L.B., Henderson, A.D., van Zelm, R., Jolliet, O., Huijbregts, M.A.J., 2013. Assessing the importance of spatial variability versus model choices in life cycle impact assessment: the case of freshwater eutrophication in Europe. Environ. Sci. Technol. 47: 13565–13570. http://dx.doi.org/10.1021/es403422a. Brooker, M.P., 1985. The ecological effects of channelization. Geogr. J. 151, 63–69. Bunn, S.E., Arthington, A.H., 2002. Basic principles and ecological consequences of altered flow regimes for aquatic biodiversity. Environ. Manag. 30:492–507. http://dx.doi.org/ 10.1007/s00267-002-2737-0. Bush, R.M., Welch, E.B., Mar, B.W., 1974. Potential effects of thermal discharges on aquatic systems. Environ. Sci. Technol. 8:561–568. http://dx.doi.org/10.1021/es60091a009. Caissie, D., 2006. The thermal regime of rivers: a review. Freshw. Biol. 51:1389–1406. http://dx.doi.org/10.1111/j.1365-2427.2006.01597.x. Chaudhary, A., Verones, F., de Baan, L., Hellweg, S., 2015. Quantifying Land Use Impacts on Biodiversity: Combining Species–Area Models and Vulnerability Indicators. Cook, M.A., King, C.W., Davidson, F.T., Webber, M.E., 2015. Assessing the impacts of droughts and heat waves at thermoelectric power plants in the United States using integrated regression, thermodynamic, and climate models. Energy Rep. 1:193–203. http://dx.doi.org/10.1016/j.egyr.2015.10.002.

Coutant, C.C., Brook, A., 1973. Biological aspects of themal pollution II. Scientific basis for water temperature standards at power plants. CRC Crit. Rev. Environ. Control. 3, 1–24. Cucurachi, S., Heijungs, R., 2014. Characterisation factors for life cycle impact assessment of sound emissions. Sci. Total Environ. 468–469:280–291. http://dx.doi.org/10.1016/j. scitotenv.2013.07.080. Davidson, B., Bradshaw, R.W., 1967. Thermal pollution of water systems. Environ. Sci. Technol. 1:618–630. http://dx.doi.org/10.1021/es60008a606. de Baan, L., Mutel, C.L., Curran, M., Hellweg, S., Koellner, T., 2013. Land use in life cycle assessment: global characterization factors based on regional and global potential species extinction. Environ. Sci. Technol. 47:9281–9290. http://dx.doi.org/10.1021/ es400592q. de Vries, P., Tamis, J.E., Murk, A.J., Smit, M.G.D., 2008. Development and application of a species sensitivity distribution for temperature-induced mortality in the aquatic environment. Environ. Toxicol. Chem. 27:2591–2598. http://dx.doi.org/10.1897/08056.1. European Parliament and Council of the European Union, 2006. Directive 2006/44/EC of the European Parliament. Off. J. Eur. Union. Feeley, T.J., Skone, T.J., Stiegel, G.J., McNemar, A., Nemeth, M., Schimmoller, B., Murphy, J.T., Manfredo, L., 2008. Water: a critical resource in the thermoelectric power industry. Energy 33:1–11. http://dx.doi.org/10.1016/j.energy.2007.08.007. Finnveden, G., Hauschild, M.Z., Ekvall, T., Guinée, J., Heijungs, R., Hellweg, S., Koehler, A., Pennington, D., Suh, S., 2009. Recent developments in life cycle assessment. J. Environ. Manag. 91:1–21. http://dx.doi.org/10.1016/j.jenvman.2009.06.018. Förster, H., Lilliestam, J., 2010. Modeling thermoelectric power generation in view of climate change. Reg. Environ. Chang. 10:327–338. http://dx.doi.org/10.1007/s10113009-0104-x. Fricko, O., Parkinson, S.C., Johnson, N., Strubegger, M., van Vliet, M.T.H., Riahi, K., 2016. Energy sector water use implications of a 2 °C climate policy. Environ. Res. Lett. 11: 034011. http://dx.doi.org/10.1088/1748-9326/11/3/034011. Hellweg, S., Mila i Canals, L., 2014. Emerging approaches, challenges and opportunities in life cycle assessment. Science 344 (80):1109–1113. http://dx.doi.org/10.1126/ science.1248361. Helmes, R.J.K., Huijbregts, M.A.J., Henderson, A.D., Jolliet, O., 2012. Spatially explicit fate factors of phosphorous emissions to freshwater at the global scale. Int. J. Life Cycle Assess. 17:646–654. http://dx.doi.org/10.1007/s11367-012-0382-2. Hester, E.T., Doyle, M.W., 2011. Human impacts to river temperature and their effects on biological processes: a quantitative synthesis. J. Am. Water Resour. Assoc. 47: 571–587. http://dx.doi.org/10.1111/j.1752-1688.2011.00525.x. Heugens, E.H.W., Hendriks, A.J., Dekker, T., van Straalen, N.M., Admiraal, W., 2002. A review of the effects of multiple stressors on aquatic organisms and analysis of uncertainty factors for use in risk assessment. Crit. Rev. Toxicol. 31:247–284. http://dx.doi. org/10.1080/20014091111695. Holmstrup, M., Bindesbøl, A.-M., Oostingh, G.J., Duschl, A., Scheil, V., Köhler, H.-R., Loureiro, S., Soares, A.M.V.M., Ferreira, A.L.G., Kienle, C., Gerhardt, A., Laskowski, R., Kramarz, P.E., Bayley, M., Svendsen, C., Spurgeon, D.J., 2010. Interactions between effects of environmental chemicals and natural stressors: a review. Sci. Total Environ. 408:3746–3762. http://dx.doi.org/10.1016/j.scitotenv.2009.10.067. International Energy Agency, 2011. Electricity Information Statistics [WWW Document]. (URL http://www.oecd-ilibrary.org/energy/data/iea-electricity-information-statistics_ elect-data-en (accessed 3.1.16)). International Organization for Standardization, 2005. ISO 14044 Draft International Standard. Environmental Management – Life Cycle Assessment – Requirements and Guidelines (Geneva, Switzerland). International Organization for Standardization, 2006. ISO 14040 International Standard. Environmental Management – Life Cycle Assessment – Principles and Framework. second ed. (Geneva, Switzerland). Koch, H., Vögele, S., 2009. Dynamic modelling of water demand, water availability and adaptation strategies for power plants to global change. Ecol. Econ. 68:2031–2039. http://dx.doi.org/10.1016/j.ecolecon.2009.02.015. Langford, T.E.L., 1990. Ecological Effects of Thermal Discharges. Elsevier Applied Science, New York. Laurent, A., Espinosa, N., 2015. Environmental impacts of electricity generation at global, regional and national scales in 1980–2011: what can we learn for future energy planning? Energy Environ. Sci. 8:689–701. http://dx.doi.org/10.1039/C4EE03832K. Limpert, E., Stahel, W.a., Abbt, M., 2001. Log-normal distributions across the sciences: keys and clues. Bioscience 51:341–352. http://dx.doi.org/10.1641/00063568(2001)051[0341:LNDATS]2.0.CO;2. Macknick, J., Newmark, R., Heath, G., Hallett, K.C., 2012. Operational water consumption and withdrawal factors for electricity generating technologies: a review of existing literature. Environ. Res. Lett. 7:045802. http://dx.doi.org/10.1088/1748-9326/7/4/ 045802. Madden, N., Lewis, A., Davis, M., 2013. Thermal effluent from the power sector: an analysis of once-through cooling system impacts on surface water temperature. Environ. Res. Lett. 8:035006. http://dx.doi.org/10.1088/1748-9326/8/3/035006. Masanet, E., Chang, Y., Gopal, A.R., Larsen, P., Morrow III, W.R., Sathre, R., Shehabi, A., Zhai, P., 2013. Life-cycle assessment of electric power systems. Annu. Rev. Environ. Resour. 38 (38):107–136. http://dx.doi.org/10.1146/annurev-environ-010710-100408. Mekonnen, M.M., Gerbens-Leenes, P.W., Hoekstra, A.Y., 2015. The consumptive water footprint of electricity and heat: a global assessment. Environ. Sci. Water Res. Technol. 1:285–297. http://dx.doi.org/10.1039/C5EW00026B. Meldrum, J., Nettles-Anderson, S., Heath, G., Macknick, J., 2013. Life cycle water use for electricity generation: a review and harmonization of literature estimates. Environ. Res. Lett. 8:015031. http://dx.doi.org/10.1088/1748-9326/8/1/015031. Meybeck, M., 1989. The quality of rivers: from pristine stage to global pollution. Glob. Planet. Chang. 1:283–309. http://dx.doi.org/10.1016/0921-8181(89)90007-6.

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056

C.E. Raptis et al. / Science of the Total Environment xxx (2016) xxx–xxx Miara, A., Vörösmarty, C.J., 2013. A dynamic model to assess tradeoffs in power production and riverine ecosystem protection. Environ. Sci. Process. Impacts 15:1113–1126. http://dx.doi.org/10.1039/c3em00196b. Miara, A., Vörösmarty, C.J., Stewart, R.J., Wollheim, W.M., Rosenzweig, B., 2013. Riverine ecosystem services and the thermoelectric sector: strategic issues facing the Northeastern United States. Environ. Res. Lett. 8:025017. http://dx.doi.org/10.1088/17489326/8/2/025017. Mutel, C., Pfister, S., Hellweg, S., 2012. GIS-based regionalized life cycle assessment: how big is small enough? Methodology and case study of electricity generation. Environ. Sci. Technol. 46:1096–1103. http://dx.doi.org/10.3929/ethz-a-007586392. Núñez, M., Antón, A., Muñoz, P., Rieradevall, J., 2013. Inclusion of soil erosion impacts in life cycle assessment on a global scale: application to energy crops in Spain. Int. J. Life Cycle Assess. 18:755–767. http://dx.doi.org/10.1007/s11367-012-0525-5. Peredo-Alvarez, V.M., Bellas, A.S., Trainor-Guitton, W.J., Lange, I., 2016. Mandate a man to fish?: technological advance in cooling systems at U.S. thermal electric plants. Water Resour. Res. 52:1418–1426. http://dx.doi.org/10.1002/2014WR015716. Pfister, S., Koehler, A., Hellweg, S., 2009. Assessing the environmental impacts of freshwater consumption in LCA. Environ. Sci. Technol. 43, 4098–4104. Pfister, S., Saner, D., Koehler, A., 2011. The environmental relevance of freshwater consumption in global power production. Int. J. Life Cycle Assess. 16:580–591. http:// dx.doi.org/10.1007/s11367-011-0284-8. Pfister, S., Suh, S., 2015. Environmental impacts of thermal emissions to freshwater: spatially explicit fate and effect modeling for life cycle assessment and water footprinting. Int. J. Life Cycle Assess. 20:927–936. http://dx.doi.org/10.1007/s11367015-0893-8. Platts, U.P.G., 2012. UDI World Electric Power Plants Data Base, March 2012. Raptis, C.E., Pfister, S., 2016. Global freshwater thermal emissions from steam-electric power plants with once-through cooling systems. Energy 97:46–57. http://dx.doi. org/10.1016/j.energy.2015.12.107. Raptis, C.E., van Vliet, M.T.H., Pfister, S., 2016. Global thermal pollution of rivers from thermoelectric power plants. Environ. Res. Lett. 11:104011. http://dx.doi.org/10.1088/ 1748-9326/11/10/104011. Slob, W., 1994. Uncertainty analysis in multiplicative models. Risk Anal. 14:571–576. http://dx.doi.org/10.1111/j.1539-6924.1994.tb00271.x. Souchon, Y., Tissot, L., 2012. Synthesis of thermal tolerances of the common freshwater fish species in large Western Europe rivers. Knowl. Manag. Aquat. Ecosyst.:1–48 http://dx.doi.org/10.1051/kmae/2012008. Spang, E.S., Moomaw, W.R., Gallagher, K.S., Kirshen, P.H., Marks, D.H., 2014. The water consumption of energy production: an international comparison. Environ. Res. Lett. 9:105002. http://dx.doi.org/10.1088/1748-9326/9/10/105002. Stewart, R.J., Wollheim, W.M., Miara, A., Vörösmarty, C.J., Fekete, B., Lammers, R.B., Rosenzweig, B., 2013. Horizontal cooling towers: riverine ecosystem services and

13

the fate of thermoelectric heat in the contemporary Northeast US. Environ. Res. Lett. 8:025010. http://dx.doi.org/10.1088/1748-9326/8/2/025010. Sweeney, B.W., Bott, T.L., Jackson, J.K., Kaplan, L.A., Newbold, J.D., Standley, L.J., Hession, W.C., Horwitz, R.J., 2004. Riparian deforestation, stream narrowing, and loss of stream ecosystem services. Proc. Natl. Acad. Sci. U. S. A. 101:14132–14137. http://dx.doi.org/ 10.1073/pnas.0405895101. Tendall, D.M., Hellweg, S., Pfister, S., Huijbregts, M.A.J., Gaillard, G., 2014. Impacts of river water consumption on aquatic biodiversity in life cycle assessment—a proposed method, and a case study for Europe. Environ. Sci. Technol. 48, 3236–3244. U.S. EPA, 2014. Technical Development Document for the Final Section 316(b) Existing Facilities Rule (Washington). US Energy Information Administration, 2011. International Energy Statistics [WWW Document]. (URL http://www.eia.gov/cfapps/ipdbproject/IEDIndex3.cfm (accessed 1.22.15)). van Vliet, M.T.H., Yearsley, J.R., Franssen, W.H.P., Ludwig, F., Haddeland, I., Lettenmaier, D.P., Kabat, P., 2012a. Coupled daily streamflow and water temperature modelling in large river basins. Hydrol. Earth Syst. Sci. 16:4303–4321. http://dx.doi.org/10. 5194/hess-16-4303-2012. van Vliet, M.T.H., Yearsley, J.R., Ludwig, F., Vögele, S., Lettenmaier, D.P., Kabat, P., 2012b. Vulnerability of US and European electricity supply to climate change. Nat. Clim. Chang. 2:676–681. http://dx.doi.org/10.1038/nclimate1546. Vassolo, S., Döll, P., 2005. Global-scale gridded estimates of thermoelectric power and manufacturing water use. Water Resour. Res. 41, W04010. http://dx.doi.org/10. 1029/2004WR003360. Verones, F., Hanafiah, M.M., Pfister, S., Huijbregts, M.A.J., Pelletier, G.J., Koehler, A., 2010. Characterization factors for thermal pollution in freshwater aquatic environments. Environ. Sci. Technol. 44:9364–9369. http://dx.doi.org/10.1021/es102260c. Verones, F., Hanafiah, M.M., Pfister, S., Huijbregts, M.A.J., Pelletier, G.J., Koehler, A., 2011. Correction to characterization factors for thermal pollution in freshwater aquatic environments. Environ. Sci. Technol. 45:7608. http://dx.doi.org/10.1021/es102260c. Verones, F., Saner, D., Pfister, S., Baisero, D., Rondinini, C., Hellweg, S., 2013. Effects of consumptive water use on biodiversity in wetlands of international importance. Environ. Sci. Technol. 47:12248–12257. http://dx.doi.org/10.1021/es403635j. Zhang, C., Zhong, L., Fu, X., Wang, J., Wu, Z., 2016. Revealing water stress by the thermal power industry in China based on a high spatial resolution water withdrawal and consumption inventory. Environ. Sci. Technol. 50:1642–1652. http://dx.doi.org/10. 1021/acs.est.5b05374. Zhao, Y., Wang, S., Duan, L., Lei, Y., Cao, P., Hao, J., 2008. Primary air pollutant emissions of coal-fired power plants in China: current status and future prediction. Atmos. Environ. 42:8442–8452. http://dx.doi.org/10.1016/j.atmosenv.2008.08.021.

Please cite this article as: Raptis, C.E., et al., Assessing the environmental impacts of freshwater thermal pollution from global power generation in LCA, Sci Total Environ (2016), http://dx.doi.org/10.1016/j.scitotenv.2016.12.056