Spatially explicit life cycle assessments reveal hotspots of environmental impacts from renewable electricity generation

Phases of the life cycle assessment

The LCA is carried out in four steps in accordance with DIN ISO EN 14040²². Phase 1 includes the determination of the goal and scope of the analyses. The functional unit provided by the four case studies is defined as 1 kWh of net electricity produced and results are related to this. The system boundary is defined to determine which processes along the process chain of the four case studies are considered in the LCA. Figure 1 shows that the electricity-producing processes are considered as well as the associated supply chains, respectively (only a selection shown in Fig. 1). If necessary, an allocation with respect to different output products is performed (bagasse). For the bagasse, two different system boundaries are considered in comparison which is described in detail below. The overall objective is to compare the different case studies in a meaningful way with regard to a broad range of environmental impacts and to identify hotspots of environmental burdens, which also defines the scope.

In the second phase, inventory analysis, the LCA models are designed in the software openLCA. They are based on existing datasets from the LCA database ecoinvent 3.5²³ which are extended or modified with the help of case study-specific data, as the focus of this study is not on the presentation of an extensive life cycle inventory. After a collection, description, and review of case study-specific data elementary flows from and to the environment are derived and added to the models as inputs and outputs. Taking into account the great differences between the case studies, an attempt was made to process the inventories with the same degree of accuracy and to consider the same elementary flows, where possible. Moreover, the inventory is adapted to include elementary flows relevant for the assessment phase. Product systems are created, linking the models of the case studies to the LCA database and providing the supply chain. In the case of eight mineral resources, the upstream chains in the database have been modified in advance.

In the third phase, the impact assessment, the product systems are evaluated using selected indicators. In the steps characterisation and (optionally) normalisation and weighting, the contribution of the product system to different impact categories is determined.

In the last phase, evaluation, the contributions are analysed spatially explicit and uncertainty analyses are carried out against the background of the framework conditions from phase 1. Mid-point approaches refer directly to the environmental impact categories, whereas end-point approaches describe impacts on the protected goods Human Health, Ecosystem Quality, and Resources. Both approaches are used in this study, whereas end-point impacts on Human Health and Ecosystem Quality are assessed. LCA is considered here from the use of raw materials to production (cradle to gate). The results for the construction phase of the facilities (buildings, infrastructure, machinery) and the operation phase (production) are presented separately, which is well suited for the comparison of energy systems²⁴. The construction phase is also related to the functional unit of 1 kWh.

Life cycle inventory analysis

In the following, the LCA models of the four case studies are described comprising general information, the used ecoinvent 3.5²³ dataset, modifications for the construction and operation phase, a definition of the functional unit as well as the allocation approach along with information about special data handling (access of life cycle inventory see “Data availability” statement). The four case studies have been selected in the course of a research project which was carried out in cooperation with local practice partners. The practice partner have provided internal knowledge and data at first hand which has opened up the possibility of comparing the four case studies in terms of a wide range of environmental impacts based on real data.

Case study (a), the ROR hydropower plant at the Danube, consists of six barrages between the Bavarian cities Oberelchingen and Faimingen. The plants have been built from 1960 to 1965 and are each equipped with two double-regulated Kaplan turbines with stationary shaft and one directly mounted synchronous generator each. Drop heights are 5–7 m, the capacities are 7–10 MW and an average of approximately 50 GWh is annually generated per barrage (information from the website of the operator Bayerische Elektrizitätswerke GmbH). An ecoinvent 3.5 dataset for an average European ROR hydropower plant is taken as basis to model the case study. For the construction phase, areas that were transformed during construction or are occupied with case study facilities are analysed by evaluating freely accessible satellite images. So called transformation from wetland, from water bodies, to industrial area, to traffic area, and to water bodies (visible in satellite images from bulges, Supplementary Notes 3 and Supplementary Fig. 8) as well as the occupation of river are considered as elementary flows. For the other case studies, a similar approach was taken according to the available information from the operators and satellite image evaluations. For the operation phase, a turbine water use of 111 m³ kWh⁻¹ is calculated using the equation P = Q × h × c₁ (P: capacity in W, Q: water flow rate in m³ s⁻¹, h: drop height in m, c₁ = 8.5 KN m⁻³, the latter including gravity, density of water and a plant efficiency of 85%). Evaporation losses from additional impoundment during the operation phase are calculated at 0.02 m³ kWh⁻¹ taking into account a total transformation to water bodies by the six barrages of approximately 1 million m² and an evaporation rate of 643 l m⁻² a⁻¹ as described in literature for similar latitudes²⁵. The difference to the input water, i. e. the water flowing through minus the evaporation losses, is modelled as emission to water. Due to their close spatial proximity, the six barrages were balanced together. The functional unit of the operation phase is 1 kWh for all case studies. The inventory of the construction phase is also related to 1 kWh using a factor which is derived from the total capacity of the six barrages of 52 MW, an annual production of 50 million kWh, and a life time of 80 years²⁶ for the cement in dams, tunnels and control units (the latter only considered in the conversion for the construction phase). A shorter lifetime for the steel for turbines and tubes of 40 years is already considered in the original ecoinvent 3.5 process.

Case study (b), the CSP Noor I, which started operation in 2016 is located in the Moroccan desert near the city Ouarzazate. The location has one of the highest solar radiation levels in the world with 2635 kWh m⁻² annually. The sun is shining almost 365 days. The 160 MW plant consists of a solar field, a power block, and a thermal energy storage. In the solar field, parabolic trough collectors use the solar radiation to heat up a heat transfer fluid. The power block, consisting of steam generation system, super-heater, turbine, re-heater, condenser, pre-heater, optional boiler, heat-exchangers, cooling tower, and pumps, receives this fluid to convert it into electricity. The thermal storage, to produce electricity in the absence of solar radiation, is based on molten salt, a mixture of 60% sodium nitrate and 40% potassium nitrate²⁷. In contrast to the follow-up Noor projects in the region, the cooling system of Noor I still relies on water, and water is also needed to remove sand from the solar panels. It is taken from the nearby reservoir El Mansour Eddahbi. An ecoinvent 3.5 dataset for a 50 MW CSP is taken as basis to model the case study. For the construction phase, occupation of approximately 40 million m² by industrial area is considered and a total water use of 0.3 million m³ reported by the operator is added to the inventory. For the operation phase, the elementary flow energy, solar, converted is added to account for the energy input from the sun. It is calculated by dividing the energy output of 1 kWh by 25%, which corresponds to the thermal-energy-to-electricity efficiency²⁸. For the LCA analysis, not the total energy input is taken into account, but the efficiency of the system after the conversion of the solar heat into thermal energy. If the efficiency of the conversion of solar energy to thermal energy of about 59%²⁸ is considered, the overall efficiency of the conversion of solar energy to electricity would be 15%, which is also used in other studies²⁹. The water demand was modelled in accordance with the information provided by the operator: A water use of approximately 0.005 m³ kWh⁻¹ is considered for cooling purposes and cleaning of the solar panels to remove sand. The water input is accounted for as evaporation loss as no water is recharged to the reservoir but collected on-site in evaporation ponds or reused if possible. The inventory of the construction phase is related to 1 kWh using a factor that is derived from the total capacity of 160 MW, a net annual production of 370 million kWh (information kindly provided by the operator), and a life time of 30 years³⁰ (the latter only considered in the conversion for the construction phase).

In the Rio dos Patos basin, Brazil, sugarcane is cultivated on a total available area of 65,000 ha of formerly degraded pasture (case study (c)). The sugar cane is processed during nine months of the year by the sugar mill units Jalles Machado and Otávio Lage which are both located in Goianésia and started operating in 1980 and 2011. At first, the fresh plants are grinded to separate plant fibres from sugar cane water, which is further processed to produce primarily sugar and ethanol as well as yeast as a by-product. Distillery wastewater, so-called vinasse, is returned to the fields as irrigation and fertiliser to complete the cycle. The pressed plant fibres, so-called bagasse, are burned to produce electricity via a system of boilers, steam turbine, and generator. The electricity is partly used for self-supply and otherwise fed into the electricity grid. Additionally, generated heat is fed into the sugar fermentation process. As 54% of the yearly produced sugar cane is irrigated and the region may frequently be exposed to water scarcity during the dry season, the operator makes extensive efforts to steadily reduce water consumption in agriculture and industry: In addition to elaborated planting and harvesting strategies or the use of efficient plants, the focus is on strategic water management: While in 2018 45% of irrigation consisted of salvage irrigation (one single application of approximately 40 mm of surface water with a boom traveller during the growth period), 17% was deficit irrigation with between 25 and 50% of the plant water deficit supplied. The return of vinasse and residual process water from the mill to the fields is another key element of the irrigation strategy and accounted for 37% of total irrigation in 2018. Dryer years may demand more salvage irrigation which can results in a shift of shares. This information was kindly provided by the operator Jalles Machado S/A Açúcar e Álcool. The LCA model was applied with two different allocation approaches for the analysis of electricity production from bagasse (Fig. 1) which differ in that (1) bagasse is considered as a by-product of sugar and ethanol production and (2) as a waste product thereof. From an LCA perspective, this is a critical question: In the Brazilian case study, the bagasse is seen as pure waste product as sugar cane is grown exclusively for the production of sugar and ethanol. Furthermore, there are no disposal costs for the bagasse which would classify it perfectly as waste. But since the electricity is sold, bagasse could also be assigned an economic value as an energy source. Therefore, many authors are performing a by-product allocation from the outset (for example Botha and Blottnitz 2006³¹, Lopes Silva et al. 2014³², Mashoko et al. 2013³³ and Ramjeawon 2008³⁴). In order to be able to deal with different realities, especially in an international comparison, both approaches have been considered in this study comparatively. The LCA by-product approach is based on a dataset from ecoinvent 3.5 for electricity from sugarcane³⁵. For the construction phase, a 40 MW gas turbine and elementary flows for transformation from approximately 500.000 m² pasture to industrial area and occupation of industrial area are added. For the operation phase, the calorific value of sugarcane of 5 MJ kg⁻¹, transformation from pasture to irrigated and non-irrigated annual crop, occupation by irrigated and non-irrigated annual crop and an evapotranspiration of 0.38 m³ kg are considered in the process step sugar cane production (information kindly provided by operator). In the process step electricity generation, a net water demand of 0.008 m³ kWh⁻¹ due to evaporation losses from the boiler system is considered (information kindly provided by operator). For the by-product model, an economic allocation of sugar cane juice, which serves as starting material for the production of sugar and ethanol, and sugar cane fibres, i. e. bagasse, is performed in an upstream production step with respect to current market prices (Supplementary Data 13). The amount of bagasse used per kWh produced is calculated from its high heating value³⁶ of 16 MJ kg⁻¹ at 0.22 kg kWh⁻¹. The LCA waste model is based on the same ecoinvent 3.5 data. For the construction phase, only the 40 MW gas turbine is considered without modifications. For the operation phase, only the process step electricity generation is considered with a net water demand of 0.008 m³ kWh⁻¹ (see above). The inventory of the construction phase is related to 1 kWh using a factor that is derived from a net annual production of bagasse of 700 million kg, a turbine capacity of 40 MW, a lifetime of 50 years (information kindly provided by the operator) and an estimated share of bagasse processing infrastructure in the total sugar mill of 5%.

Case study (d), the CPP Heyden in Petershagen, at the Weser River serves as reference for conventional electricity generation within this study. Commissioned in 1987 it is still Germany’s most powerful power plant with a net capacity of 875 MW that will probably operate until the end of 2025. The fuel is hard coal which is delivered mainly from Russia. Waste gases from combustion are purified by passing them through denitrification, dust removal, and desulphurisation plants gradually. Waste water is also treated, including one of the world’s first ultrafiltration plants. During operation, the CPP produces by-products such as flue ash, gypsum, and slag which are reused for different purposes. This information was kindly provided by the operator Uniper Kraftwerke GmbH. An ecoinvent 3.5 dataset for an average European CPP is taken as basis to model the case study and no modifications are made for the construction phase. For the operation phase, the net water demand is put at 0.001 m³ kWh⁻¹ which represents the loss from cooling water input, and a multitude of material inputs and emissions to air and water, such as cadmium and mercury, are added (kindly provided by the operator). The economic allocation of electricity and the by-products is neglected, as their economic value is too low. The inventory of the construction phase is related to 1 kWh using a factor which is derived from a total electricity production of 10¹¹ kWh over a lifetime of 35 years with the German coal phase-out also considered and a capacity of 920 MW (information kindly provided by the operator).

Regionalisation of supply chains of selected mineral commodities

The supply chains of the mineral commodities aluminium, copper, coal, cement, iron and steel, lithium, and phosphorus are regionalised at mine site level and inserted into the LCA database ecoinvent 3.5. A detailed description of the procedure and the associated data is presented in Supplementary Notes 7. Essentially, global production data for the selected mineral commodities are taken (see “Data availability” statement) to select those countries which account for the largest share of world production, so that overall 80% of world production was covered. Mine sites in the selected countries (see data availability statement) are clustered using a hotspot analysis, taking into account distance and regional water stress, resulting in a maximum of five mine sites per country, each representing an entire mining region. These sites are added to the LCA database as single processes and linked to the existing supply chains in the database according to their share of world production. For example, for the CPP, it is known that the hard coal is sourced exclusively from Russia, but where no case study-specific data are available, the upstream chains regionalised in the described way are linked to the case studies. These supply chains represent the most likely origin of a particular mineral commodity, based on global production volumes.

Life cycle impact assessment

In order to cover a wide range of environmental impacts, a number of LCIA indicators are considered. Primarily the resource footprints, which already cover more than 80% of all environmental impacts¹², are to be taken into account. For the climate³⁷, energy, land, material, and water footprint methods are selected, that quantify and assess elementary flows, as footprint is understood here as a value weighted according to certain criteria. In addition, the indicator set should meet the requirements of the German environmental impact assessment, which is an environmental policy instrument in Germany to evaluate environmentally relevant projects for possible environmental impacts before approval, is also included. In contrast to it, the LCA assessment is not restricted to the planning phase. Moreover, LCA analyses also evaluate remote environmental impacts associated with the upstream supply chain which is missing in the German environmental impact assessment. The approach presented here is seen as a possible interface between science-based indicators and practical applications.

Based on the Cumulative Energy Requirements Analysis^38,39 the sub-indicator Fossil Cumulative Energy Demand (CED_fo) from the LCA implementation of Hischier et al. 2010⁴⁰ is used within this study as energy footprint: It summarises the energy provided by the fossil energy carriers hard coal, lignite, crude oil, natural gas, coal mining off-gas as well as peat, uranium and wood and biomass from primary forests along the supply chain⁴⁰ and assesses them according to the energy content in MJ equivalents m⁻³, kg⁻¹ or MJ⁻¹. This is implemented by multiplication with corresponding characterisation factors.

For the climate footprint³⁷, the global warming impact is calculated according to the LCA implementation IPCC 2013 using the impact category climate change GWP100a (GWP₁₀₀)⁴⁰. The elementary flows of carbon dioxide, carbon monoxide, chloroform, dinitrogen monoxide, different ethane and methane compounds, nitric oxide, nitrogen fluoride, sulphur hexafluoride as well as volatile organic compounds are summarised along the supply chain and assessed with respect to their global warming potential in kg CO₂-equivalents kg⁻¹.

Two indicators represent the product material footprint⁴¹: For the Raw Material Input (RMI) the input of a multitude of abiotic materials from aluminium to zirconium is summarised along the supply chain and assessed with respect to “the ratio of the mass of the extracted raw material (used extraction) to the mass of the respective abiotic material in the extracted raw material in kg kg⁻¹”⁴². The Total Material Requirement (TMR) comprises the input of similar abiotic materials, which are assessed with respect to “the ratio of the mass of unused extraction and the mass of the extracted primary material for the production of the material measured in kg kg⁻¹”⁴².

The water consumption is determined and evaluated as midpoint LCA water scarcity footprint¹⁴. It “assesses the on-site and remote probability of natural freshwater scarcity for humans and nature caused by water use along human supply chains in a spatially explicit way”¹⁴. The Quantitative Water Scarcity Footprint (WSF_quan) represents the quantitative water consumption through evapotranspiration, product-incorporated water and water transfer across basin boundaries in regionally weighted m³ of water. The weighting, which relates to the water stress level in a country, is carried out with the LCIA method AWARE¹⁵, respectively. The Qualitative Water Scarcity Footprint (WSF_qual), which is the regionally weighted virtual volume of water in m³ required to dilute process-related aluminium emissions into water bodies to safe concentrations, is not included in this study. Schomberg et al. 2021¹⁴ have already pointed out that the WSF_qual is mainly due to waste treatment in global supply chains, because of high aluminium emissions. Since we cannot provide any new information on these upstream chains, especially not spatially explicit, the WSF_qual would not provide any new insights.

To assess land use impacts on biodiversity there is currently no midpoint LCIA available that is suitable for the examined case studies. The Life Cycle Initiative made an interim recommendation for the indicator potential species loss from land use⁴³ in 2016, but also stated, that it is not suitable for comparative assertions. In a newer report from 2019, the LANCA^®44 approach is recommended to assess land use impacts on soil quality. However, for the purpose of this study, it is rather the total area occupied in the context of the case studies and the encroachment on the natural ecosystem by land use that are of interest. Hence, the pure land occupation is summed up along the supply chains of the case studies in m² × a without any weighting in order to provide first information on possible teleconnections as already Kaiser et al. 2021⁴⁵. Land use changes are not evaluated in the absence of a suitable method so far.

The endpoint LCA method ReCiPe Endpoint (H,A)⁴⁶ is used to reveal damages on the LCA protected goods human health and ecosystem quality. The sub-indicator Human Health (HuHe) summarises impacts in the categories climate change, human toxicity, ionising radiation, ozone depletion, particulate matter formation, and photochemical oxidant formation and assesses them according to modelled and harmonised impact pathways in points kg⁻¹ or m⁻² or m⁻² a⁻¹, respectively. The sub-indicator Ecosystem Quality (ECO) summarises impacts in the categories agricultural land occupation, climate change, freshwater ecotoxicity, freshwater eutrophication, marine ecotoxicity, natural land transformation, terrestrial acidification, and terrestrial ecotoxicity and assesses them analogously.

For LCIA indicators with sub-categories, the results of the single sub-categories are added up to receive the total indicator results: The indicator CED_fo for example consists of the sub-categories fossil, nuclear and primary forest, whose individual results were added together. Occasionally, especially for the WSF_quan, this approach removed negative values that can for example occur when datasets in the LCA database contain rounding errors.

Hotspot analysis of LCIA results

An LCIA not only provides an overall result for an environmental pressure or impact, but also shows the contributions of individual activities. The term activity refers to the processes that make up the upstream chain of a case study. A systematic analysis of the extensive information provided by the LCIA is the main focus of this study and several methodological steps are carried out to present relevant information in a comparative manner:

(1) As an average supply chain can be made of ~100,000 single activities, only processes that contribute more than 1% to the total result of an environmental pressure or impact per case study are selected. This represents at least 48%, but on average 79%, of a total indicator result (Supplementary Data 14). The difference to 100% is mostly due to many thousands of processes that contribute very little, respectively, the analysis of which would go beyond the scope and distract from focal points.

(2) To make activities of different environmental pressures and impacts, which have different units, comparable, a normalisation step is carried out. For each environmental pressure or impact p, single activity results of a case study c, x_p,c,i, are normalised by the median med_p of all activities of all case studies⁴⁷. The sample for determining this median includes all activity results of all case studies of an environmental pressure or impact (Eq. 1, ROR: run-of-river hydropower, CSP: concentrated solar power, BBY: bagasse as by-product, BWA: bagasse as waste, CPP: coal-fired power plant, operation, and construction phase not listed separately). The normalised values are calculated per environmental pressure or impact p by dividing single activity results by the median (Eq. 2).

(3) A normalised value represents the ratio of the activity result to the median, meaning that the activity result is norm_p,c,i times as large as the median. For the spatial hotspot analysis, normalised values are presented on a scale from 1 to 100. Values below 1 represent activity results smaller than the median which are hence no hotspots, values above 100 are set to a maximum of 100 (dark orange hotspots in Fig. 5) to keep the scale manageable. It can be assumed that a result that is more than one hundred times the median is a hotspot in any case. This approach is inspired by the calculation approach of the water stress indicator AWARE¹⁵, which is already widely accepted in the LCA community. A colour scale is used to provide orientation and to distinguish hotspots of different severity according to their normalised value: 1–5 light blue, 5–10 dark blue, 10–30 pink, 30–50 yellow, 50–100 orange, >100 dark orange. Land occupation and environmental impacts, which represent LCIA endpoint approaches are not included in the spatial hotspot analysis. As regards land occupation, spatial information is lacking for a correct assessment; in case of the environmental impacts, it must be assumed that, as a result of the endpoint perspective, impacts do not also occur at the location of the associated activities.

(4) Locations can occur more than once because different activities take place in the same place or an activity causes more than one pressure. Hence, for each location, all single normalised activity results are summed up per case study.

(5) The level of regionalisation strongly depends on the input data and can reach from point coordinates to global, which is equivalent to unknown. Due to the large volume of data involved in the analysis of international supply chains, the data may contain inaccuracies in the regionalisation. To address this issue and to be able to see the level of regionalisation at a glance, a quality index is provided for locations: Locations of quality 1 are point coordinates. Locations of quality 2 represent country or sub-country level, while locations of quality 3 are regions of two or more countries. Spatial allocations that are at least questionable, e. g. treatment of waste in Switzerland which is part of the supply chain of the sugar mill in Brazil, are assigned quality 4, while unknown locations, referred to as global or rest-of-world, are assigned quality 5. Locations of quality 1 are added in the course of this study, locations of quality 2–5 are taken from the ecoinvent 3.5 database. Locations of quality 4 and 5 are excluded from the spatial hotspot analysis. Quality indices are provided for all analysed processes in the Supplementary Information and are used to identify activity groups with poor regionalisation.

(6) To assess the relevance of single activities for the total environmental burden of the case studies independent of the presence of spatial information, they are grouped in the categories mining, hard coal mining, forestry, refining, burning of diesel, production of petroleum and (natural) gas, production of electricity and heat, production of polyethylene, treatment of waste, on-site activities (that take place at the location of the case studies) and other activities. A representation as a matrix allows a quick evaluation of the relevance of a category for the respective case study (Fig. 6) to identify particularly important environmental burdens. In addition, for each activity the number of LCIA indicators the respective activity influences, i.e., is responsible for impacts in the categories of the indicator, is given by numbers from 1 to 8. A comparison of the matrix to activities with poor regionalisation, identified in step 5 of the hotspot analysis, allows to estimate the relevance of activities with a poor regionalisation.

Data quality evaluation, assumptions, and limitations

As close cooperation is established with the operators of the case studies, many data are directly provided by the operators (see data availability statement). In these cases, no evaluation of the data sources is carried out. This may result in discrepancies with other studies: For example, Aqachmar et al. 2019²⁷ reported approximately 0.006 m³ kWh⁻¹ water use for the Moroccan CSP, in contrast to 0.005 m³ kWh⁻¹ reported by the operator within this study. Moreover, Verán-Leigh & Vázquez-Rowe 2019⁴⁸ considered a lifetime of 50 years for the permanent structural items and a reservoir water evaporation of 0.003 m³ kWh⁻¹, however, the examined case studies differ from this study in location and technical equipment so that the values are not transferable. As the reservoir water evaporation of 0.03 m³ kWh⁻¹ used in ecoinvent 3.5 for German non-alpine reservoirs, could not be verified, a value of 0.02 m³ kWh⁻¹ was calculated for this study. The measurement of additional impoundment areas from satellite images is subject to a great deal of uncertainty, as the images only represent a snapshot and it is unclear whether the images were taken at a time of high or low water levels (Supplementary Fig. 1). In the absence of more precise data, they are intended to give an idea of the possible magnitude of water consumption through evaporation. Biogenic greenhouse gas emissions from biogenic decay in reservoirs⁴⁹, which has been calculated for ROR hydropower plants with reservoirs⁴⁸, is not considered within this study, as the methodology is explicitly reported for dams and not additional impoundment areas. All other data used are taken from scientific publications. For all values that are added to the LCA models in the course of this study, a logarithmic normal distribution is assumed for the error distribution. These are included in the Monte Carlo simulations of the uncertainty analysis.