Article https://doi.org/10.1038/s41467-022-33976-5 Energy requirements and carbon emissions for a low-carbon energy transition 1 1,2 3 Received: 26 July 2021 Aljoša Slameršak , Giorgos Kallis & Daniel W. O’ Neill Accepted: 10 October 2022 Achieving the Paris Agreement will require massive deployment of low-carbon energy. However, constructing, operating, and maintaining a low-carbon Check for updates energy system will itself require energy, with much of it derived from fossil fuels. This raises the concern that the transition may consume much of the 1234567890():,; 1234567890():,; energy available to society, and be a source of considerable emissions. Here we calculate the energy requirements and emissions associated with the global energy system in fourteen mitigation pathways compatible with 1.5 °C of warming. We find that the initial push for a transition is likely to cause a 10–34% decline in net energy available to society. Moreover, we find that the carbon emissions associated with the transition to a low-carbon energy system are substantial, ranging from 70 to 395 GtCO2 (with a cross-scenario average of 195 GtCO2). The share of carbon emissions for the energy system will increase from 10% today to 27% in 2050, and in some cases may take up all remaining emissions available to society under 1.5 °C pathways. The IPCC’s Special Report on Global Warming of 1.5 °C concludes that climate mitigation could lead to a period of reduced energy we can still meet the 1.5 °C target and that by doing so, we would availability11,12, and at the same time, also consume a large share of the reduce climate impacts and limit the risk of exceeding the tipping remaining carbon budget13. Recent studies, however, find the points of the climate system1. The report provides a range of low- hypothesis of lower energy availability might be exaggerated due to carbon energy pathways compatible with limiting global warming to overestimating the EROI of fossil fuels14,15 and underestimating 1.5 °C. However, at present, there is no estimate of how much energy improvements in the EROI of renewable energy technologies16,17. would be needed to build and maintain a low-carbon energy system, or Alongside EROI, life-cycle assessment is another accounting what amount of greenhouse gas emissions would be associated with technique that has been used to quantify climate change impacts from such a transition2–4. This is an important gap in knowledge, as previous different energy generation technologies. However, life-cycle studies research suggests that rapid growth of low-carbon infrastructure typically only estimate the impacts of present-day energy technologies could use a substantial amount of the global energy supply5,6. More- applied to a particular case study18–20. Life-cycle assessment has rarely over, since the global energy supply is currently derived mostly from been used in a dynamic analysis where the impacts of technologies fossil fuels, the transition itself may become a source of significant change over time, or to assess the cumulative impacts of decarbonis- emissions7,8. ing the entire global energy system. Some studies suggest that renewables have a lower energy return A notable exception is a study by Pehl et al.21 who used a dynamic on energy invested (EROI) compared to the current energy system9,10. approach to estimate the energy requirements and emissions for the Lower EROI implies less energy delivered to society relative to the construction, operation, and maintenance of power plants. The energy required to supply the energy, leading these studies to con- authors combined a dynamic life-cycle assessment framework with an clude that a low-carbon energy transition may result in less energy Integrated Assessment Model (IAM), estimating that emissions asso- available to society. The energy required for the transition might push ciated with power plants would lead to 82 GtCO2eq of cumulative society into an “energy–emissions trap”, where achieving ambitious emissions from 2010 to 2050. In another study, Di Felice et al.8 1 The Institute of Environmental Science and Technology, ICTA-UAB, Autonomous University of Barcelona, Barcelona, Spain. 2Catalan Institution for Research and Advanced Studies, ICREA, Barcelona, Spain. 3Sustainability Research Institute, School of Earth and Environment, University of Leeds, Leeds, UK. e-mail: aljosa.slamersak@gmail.com Nature Communications | (2022)13:6932 1 Article https://doi.org/10.1038/s41467-022-33976-5 conducted a life-cycle assessment of the indirect emissions associated research objectives pursued26. As a result, EROI values are often not with the EU’s renewable energy strategy, calculating that 25 GtCO2eq directly comparable between different energy carriers and between would be emitted in the decarbonisation of the EU’s electricity gen- different studies27. To address this shortcoming, we apply a consistent eration from 2020 to 2050. These studies, however, only cover elec- energy system boundary to all energy technologies. This boundary tricity generation, which currently represents just ~20% of global final extends from the point of extraction (primary energy stage) to the energy use. Moreover, each study only analysed one specific low- point of use (final energy stage), as suggested by EROI analysts14,28. carbon pathway. To represent a range of plausible EROI transitions, we develop Here, we estimate how much energy would be required, and how three EROI scenarios. In the high-EROI scenario, we assume a fast much carbon would likely be emitted, to construct, operate, and increase in the EROI of renewables from the high end of present-day maintain the global energy system during a low-carbon energy tran- EROI values. In this scenario, we assume the EROI of bioenergy at the sition. Our study separates the energy and emissions associated with primary energy stage remains near the median of present-day values. the energy system from the energy and emissions remaining for other This scenario can be interpreted as a future of high innovation and societal uses. We thus provide complementary information to existing broad policy support for renewables, alongside efficient and sustain- mitigation pathways. Moreover, by modelling dynamic changes in the able harvest of biomass for energy. In the low-EROI scenario, we EROI of the energy system in fourteen different mitigation pathways assume a gradual increase in the EROI of renewables from the median produced by six IAMs, we provide a holistic picture for a range of of present-day EROI values, with the EROI of bioenergy remaining near distinct energy transitions, all in line with the ambitious goal of stabi- the lower end of present-day values. Such a scenario corresponds to a lising global warming below 1.5 °C. We also assess the future of moderate innovation and balanced policy support for energy–emissions trap hypothesis, considering the latest literature on renewables, and low efficiency in the management of bioenergy. In the the EROIs of different energy technologies. In doing so, we follow a median-EROI scenario, we assume a gradual increase in the EROI of consumption-based accounting approach using an EROI analysis to renewables from the median of present-day EROI values, with bioe- estimate both direct (on-site) and indirect (upstream) energy use and nergy remaining near the median of present-day values. In all three emissions associated with constructing, operating, and maintaining scenarios, we assume a gradual decline in the EROI of fossil fuels at the the energy system and the energy supply to society. Based on our primary energy stage from the present-day median value towards the results, we suggest that the energy requirements and emissions of the present-day low-EROI value, in line with historical trends and the energy system should be explicitly modelled in the next generation of existing literature5,14,29. For a detailed overview of the assumptions low-carbon mitigation pathways. across all of the EROI scenarios, see Supplementary Tables 2–7 and Supplementary Figs. 3 and 4. Results We calculate energy system emissions for different Estimating energy requirements and emissions 1.5 °C–compatible mitigation pathways as the product of energy for We refer to the energy that would be required during a low-carbon the energy system and the carbon intensity of the energy system. We energy transition as the “energy for the energy system” and the carbon divide the energy for the energy system into four energy carriers that that would be emitted as the “energy system emissions”. Energy for the each have different carbon intensities and distinguish between three energy system includes the energy required for the construction life-cycle stages: construction, operation and maintenance, and (including decommissioning), operation, and maintenance of energy decommissioning at the end of lifetime. The carbon intensity of energy facilities like power plants, mines, and refineries, as well as the energy carriers changes over time, primarily depending on the share of con- required to transport the energy carriers from the point of extraction ventional fossil fuels (i.e. fossil fuel technologies without carbon cap- to the end-user. ture and storage) in the energy mix of each carrier. By combining the To estimate the energy for the energy system, we apply the effects of technological improvements in the EROIs of energy tech- method of net energy analysis, calculating energy return on energy nologies with changes in carbon intensity due to the declining share of invested (EROI) at the final energy stage. EROI at the final energy stage fossil fuels, we capture the dynamic evolution of carbon emissions tells us how much of the total final energy is used by the energy system associated with the energy system over time. to extract, process, convert, and deliver a unit of energy to the point of We illustrate our findings using the four illustrative pathways from use for society22–24. Net energy at the final energy stage is defined as the the IPCC’s Special Report on Global Warming of 1.5 °C. Three of these difference between total final energy and the energy for the energy pathways were taken from the Shared Socioeconomic Pathways (SSP- system and represents the part of energy production that can be used 1.9) scenario study30, while one originates from the Low Energy for societal work6. We calculate EROIs and the energy for the energy Demand (LED) scenario31. These pathways represent the archetypes of system for twenty-seven energy conversion technologies, which cover different possible futures in terms of energy use, greenhouse gas the entire energy system, from 2020 to 2100 (see Methods). We dis- emissions, and preferences for energy conversion technologies, yet all tinguish between four different energy carriers (electricity, gases, manage to stabilise global warming below 1.5 °C (see Table 1). LED and liquid fuels, and solids), following the approach of Arvesen et al.25. To S1-A are pathways of rapid decarbonisation, achieved by phasing out obtain the share of energy for the energy system, we divide the energy more than 50% of fossil fuel energy by 2040, accelerating growth in requirements of the energy system by total final energy. renewable energy, and decreasing energy demand. S5-R is a pathway of In our calculations, we combine a range of EROI estimates of slower decarbonisation, long-term growth in final energy, and large- present-day energy technologies (see Supplementary Figs. 1 and 2 and scale carbon removal (which compensates for the higher emissions at Supplementary Table 1), with projections of future changes in EROI the beginning of the transition). S2-M is a “middle of the road” pathway due to technological improvements that we estimate using energetic that combines decarbonisation with slow growth in final energy and experience curves16. To account for the range of present-day esti- moderate carbon removal. For a complete representation of different mates, the uncertainty of technological change, and resource avail- 1.5 °C–compatible futures, we also analyse ten additional pathways ability, we report estimates of the energy requirements for each energy produced by Rogelj et al.30 (Supplementary Table 8), and present technology using low, median, and high-EROI values representing the average values for all fourteen pathways. first, second, and third quartiles of the inter-quartile range of our Each pathway has different “total cumulative emissions”, which estimates, respectively (see Methods). There is a divergence in EROI depend on the quantity of carbon sequestration it includes32. From the values in different studies, which can be traced to the distinct defini- perspective that interests us here, the total cumulative emissions that tions of energy system boundaries, which can vary depending on the are compatible with 1.5 °C of warming can be partitioned into energy Nature Communications | (2022)13:6932 2 Article https://doi.org/10.1038/s41467-022-33976-5 Table 1 | The four IPCC illustrative pathways Pathway Scenario assumptions Energy mix and emissions LED: Low Energy Demand1,31 Moderate population growth. Moderate Average annual emissions reduction rate (2020–2040): 6.5% decrease in energy and material use. High (rapid decarbonisation) innovation and fast adoption of sustainable Change in energy use (2020–2100): −44% technologies. Convergence to sustainable, Cumulative negative emissions from BECCS (2020–2100): low-carbon diets. 0 GtCO2 Share of cumulative final energy (2020–2100): —Renewables: 42.8% —Nuclear: 6.9% —Fossil fuels: 37.3% —Bioenergy: 12.9% S1-A: Sustainable Development30,58,76 Low population growth. Stable energy Average annual emissions reduction rate (2020–2040): 5.5% consumption and slow material growth. (rapid decarbonisation) High innovation and fast adoption of sus- Change in energy use (2020–2100): −7% tainable technologies that improve energy Cumulative negative emissions from BECCS (2020–2100): efficiency. Convergence to low-waste and 150 GtCO2 low animal share diets. Share of cumulative final energy (2020–2100): —Renewables: 44.1% —Nuclear: 5.6% —Fossil fuels: 39.9% —Bioenergy: 10.5% S2-M: Middle of the Road30,59,76 Moderate population growth. Moderate Average annual emissions reduction rate (2020–2040): 5.0% growth in energy and material use. Gradual (moderate decarbonisation) institutional and behavioural changes with Change in energy use (2020–2100): +40% slower technological innovation. Con- Cumulative negative emissions from BECCS (2020–2100): tinuation of historical dietary transition 415 GtCO2 trends. Share of cumulative final energy (2020–2100): —Renewables: 33.7% —Nuclear: 13.1% —Fossil fuels: 36.2% —Bioenergy: 16.9% S5-R: Fossil-fuelled Development30,60,76 Low population growth. High growth in Average annual emissions reduction rate (2020–2040): 3.8% energy and resource use. Delayed energy (slower decarbonisation) transition allowed by high innovation and Change in energy use (2020–2100): +76% large-scale adoption of negative emissions Cumulative negative emissions from BECCS (2020–2100): technologies. Diets with high animal shares 1190 GtCO2 and high waste. Share of cumulative final energy (2020–2100): —Renewables: 37.6% —Nuclear: 8.4% —Fossil fuels: 31.3% —Bioenergy: 22.8% The table summarises the fundamental assumptions and characteristics of the four alternative energy transitions that were selected as illustrative pathways in the IPCC’s Special Report on Global Warming of 1.5 °C. The aim of the illustrative pathways is to show different possible futures that lead to a stabilisation of global warming. The pathways differ with regards to socioeconomic, behavioural, and technological assumptions. For the “SN-X” pathways, the number N refers to the scenario narrative from the SSPs, while the letter X denotes the model that produced a particular mitigation pathway. BECCS bioenergy with carbon capture and storage. system emissions and emissions for other societal uses. To obtain the system during a low-carbon energy transition (see also Fig. 3 and share of energy system emissions in any given year, we divide energy Supplementary Fig. 5). system emissions by total emissions in that year (where the latter is Energy system emissions become more important over time, as obtained from the pathway data). they take up an increasing share of total emissions, leaving fewer emissions for other uses in society. We estimate that the share of Energy system emissions during the transition are substantial energy system emissions will increase to 2–5 times its current value by We find that the cumulative carbon emissions associated with the 2060, depending on different EROI assumptions (Fig. 2). After 2060, energy system during the transition are substantial, and represent a the share of emissions stabilises in most pathways, as the pathways considerable share of total cumulative emissions under different achieve a high degree of decarbonised energy. The share of emissions 1.5 °C–compatible scenarios (Fig. 1). The fourteen-pathway average is for the energy system in pathways S1-A, S2-M, and S5-R is much higher 195 GtCO2 for the median-EROI scenario and ranges from 185 GtCO2 than in the LED pathway, which completely decarbonises its energy for the high-EROI scenario to 290 GtCO2 for the low-EROI scenario. system. The fourteen-pathway average of energy system emissions These results correspond to an average of 21% of total emissions for increases from 10% of total emissions in 2006–2015 (for the median- the fourteen energy pathways under median-EROI assumptions, or EROI scenario) to 27% in 2050, and reaches 40% by the end of the 20% for high- and 31% for low-EROI assumptions. century. For the low-EROI scenario, the share increases from around Figure 1 shows the difference in cumulative energy system emis- 12% in 2006–2015, to 39% in 2050, and 59% by the end of the century. sions among the IPCC’s four illustrative pathways. Cumulative emis- In the high-EROI scenario, the change is from 9% in 2006–2015 to 26% sions for median-EROI values range from 70 GtCO2 (12% of total by 2050, and 31% by the end of the century. cumulative emissions) for LED, which is a low-energy-demand/no- The increase in the share of energy system emissions means that BECCS pathway, to 220 GtCO2 (20% of total cumulative emissions) for the decarbonisation of the energy system and energy supply is slower S5-R, which is a high-energy/high-BECCS pathway. Generally, in slower than the decarbonisation of the overall economy. A high share of decarbonisation pathways with higher energy use and higher deploy- emissions for the energy system may impose—particularly under the ment of BECCS, more carbon emissions are associated with the energy low-EROI scenario—a tight constraint on the “residual emissions” Nature Communications | (2022)13:6932 3 Article https://doi.org/10.1038/s41467-022-33976-5 Fig. 1 | Energy system emissions for each of the four 1.5 °C illustrative pathways. BECCS and the land-use sector (AFOLU). Each pathway allows for different total Energy system emissions (orange columns) are compared to total cumulative carbon emissions (and hence different total cumulative emissions) as each pathway emissions (blue columns). Orange error bars indicate the range of energy system assumes different amounts of carbon sequestration and non-CO2 greenhouse gas emissions calculations from high- to low-EROI model runs. Net emissions are equal emissions. See Supplementary Fig. 5 for all fourteen 1.5 °C pathways. to total emissions vented into the atmosphere minus carbon sequestration from remaining for activities such as aviation, steel and cement production, 3% of total carbon emissions in 2020, and does not undermine the and load-following electricity, which are difficult to decarbonise and target of keeping global warming below 1.5 °C. currently generate ~9 GtCO2 per year33,34. For some of the fourteen Overall, the benefits of rapid decarbonisation far outweigh the pathways, the energy system requires all of the residual emissions by extra emissions from the small jump. In pathways of rapid dec- 2080 under all three EROI scenarios—leaving no emissions for activ- arbonisation and lower energy use, the increase in emissions due to the ities such as air travel, or steel and cement production. upfront energy requirement of low-carbon infrastructure is small. In A high share of energy system emissions in some of the pathways S1-A, emissions increase by only 0.6 GtCO2 per year from 2020 to 2030 suggests the models may have been overly optimistic in their calcu- for the median-EROI scenario. Moreover, the phasing-out of fossil fuels lations of residual emissions. If this is the case, then the models need to leads to a rapid reduction in energy system emissions, starting as early either reduce the emissions allocated to other economic activities in as 2025. society or adjust their choice of energy technologies to reduce energy Over the long term, the quantity of emissions depends on the system emissions, as in the LED pathway. The pathways may be amount of fossil fuels remaining in the energy system and the choice of defended by assuming that technological innovation will make it low-carbon energy technologies in each of the pathways. Energy possible to cut emissions to zero in the sectors that are difficult to transitions that rapidly phase-out fossil fuels and prioritise renewables decarbonise today. However, this assumption is highly speculative, and nuclear energy over bioenergy technologies (BECCS in particular) given the essential role of fossil fuels in the production of steel and achieve lower cumulative energy system emissions. The reason is that cement, which are critical materials in the economy25,35. the emissions associated with renewables converge to zero (as in the LED pathway), while the emissions in pathways with BECCS only level Energy transition leads to a small jump in emissions off in the second half of the century (Fig. 3). Pathways that combine an Our results suggest that the upfront energy required to build a low- extended use of conventional fossil fuels with BECCS have higher carbon energy system would only lead to a small jump in annual energy energy system emissions, as they fail to completely decarbonise the system emissions, with the most notable increase taking place in the energy supply. Moreover, BECCS is a low-EROI technology (see also pathways of higher energy use and continued reliance on fossil fuels Supplementary Table 1 and Supplementary Figs. 3, 4). It has a low- beyond 2030 (e.g. S5-R; Fig. 3). Average energy system emissions in the energy conversion efficiency, and substantial energy requirements are S5-R pathway from 2020 to 2030 are 4.0 GtCO2 per year for the associated with bioenergy supply36,37. As such, BECCS has higher median-EROI assumption, which is 1.0 GtCO2 more than during the energy system emissions per unit of energy generated when compared 2006–2015 period. Such an increase in emissions represents less than to renewables (see Supplementary Fig. 6). Nature Communications | (2022)13:6932 4 Article https://doi.org/10.1038/s41467-022-33976-5 Fig. 2 | Share of energy system emissions over time, as a percentage of total shaded envelopes show the full range of results for the fourteen pathways. If the emissions, for three different EROI scenarios. a Low, b median, c high. In each values reach 100%, the energy system emissions exceed the total emissions in the panel, the solid line shows the average result of all fourteen pathways, while the respective pathway. result for each of the four illustrative pathways is plotted as a dashed line. The Avoiding the energy trap the S5-R pathway, the decrease is only temporary. Net energy primarily Our findings suggest that a low-carbon energy transition would drive depends on the growth in final energy and less on changes in the up the share of total energy generation going towards the construction energy for the energy system. Therefore, net energy declines sub- and operation of the energy system, and maintenance of the energy stantially in pathways that increase both the energy for the energy supply, compared to the current energy system. A higher share of system and reduce final energy. Net energy per capita could drop by energy for the energy system would contribute to a decrease in net 28–34% by 2030, compared to 2015, for pathways of rapid dec- energy available to society. Depending on the mitigation pathway, the arbonisation such as S1-A and LED. decrease in per capita net energy could be as low as 10% or as Our results are similar to those of King and van den Bergh12, who high as 34%. estimated a 24–31% reduction in net energy per capita for the IEA low- In pathways of lower energy use and rapid decarbonisation, the carbon transition pathway. However, King and van den Bergh’s increase in the share of energy for the energy system would be largest reduction takes place over a longer period (from 2015 to 2050), during the initial push for the transition when the upfront energy whereas our results indicate that a low-energy transition could lead to requirement to construct low-carbon energy infrastructure would a major reduction in net energy per capita in a single decade (Fig. 4b). consume an increasing proportion of total final energy (Fig. 4a). In In the pathways of slower decarbonisation (S2-M and S5-R), net energy pathways of moderate and slower decarbonisation, the energy share per capita declines later, and by less, with a decrease of 10% by 2040 increases in the second half of the century. The average share of energy compared to 2030 in the median-EROI scenario. Net energy in these for the energy system in the fourteen pathways for the median-EROI pathways also only declines temporarily (until 2050), after which it scenario during the 2020–2030 period is ~14%, with the highest returns to growth. increase occurring in the S1-A pathway, and the lowest in the S2-M In contrast to what has been argued in previous studies9–12,38, we pathway (Fig. 4a). find that a low-carbon energy transition would not necessarily lead to a Net energy available to society declines in all of the pathways we decline in the EROI of the overall energy system in the long term analysed, albeit at different rates and over different periods (Fig. 4b). In (Fig. 4c). The EROI of the overall energy system depends on the choice Nature Communications | (2022)13:6932 5 Article https://doi.org/10.1038/s41467-022-33976-5 Fig. 3 | Energy system emissions and energy use in the four illustrative path- bioenergy, and low-carbon technologies that include renewables, hydrogen, and ways. The left panels a, c, e, and g show annual carbon emissions associated with nuclear energy. The right panels b, d, f, and h show the final energy consumption in the energy system under the median-EROI scenario. Energy system emissions are the pathways. divided between three types of energy conversion technologies: fossil fuels, Nature Communications | (2022)13:6932 6 Article https://doi.org/10.1038/s41467-022-33976-5 Fig. 4 | Share of energy for the energy system, net energy per capita, and EROI will be left for society as net energy per capita. Panel c shows the evolution of the of the energy system for the four illustrative pathways. Panel a shows the share EROI of the overall energy system. All panels show a range of three estimates: high- of final energy that is required for the construction and operation of the energy EROI (dashed line), median-EROI (solid line), and low-EROI (dotted line). system. Panel b shows final energy per capita (solid black line), and how much of it of energy conversion technologies. EROI declines in pathways that constrained possibilities of energy growth in the models, but because prioritise bioenergy and fossil fuels with carbon capture and storage the models assume more efficient energy use, which makes such (e.g. S2-M and S5-R), and increases in pathways that focus on deploying pathways cost-effective31. renewable energy technologies (e.g. S1-A). Our results are consistent The prospect of more efficient energy use in society means that with the latest findings in the literature, which suggest that the EROI of fundamental energy services such as heating, lighting, and trans- renewable energy is comparable to (or higher than) the EROI of fossil portation could still be provided even if less net energy were avail- fuels at present, and likely to increase16, while the EROI of fossil fuels is able. Access to fundamental energy services could be maintained in likely to decrease14. However, pathways, where EROI is likely to decline, high-income countries, and increased in lower-income countries, at can still provide more net energy to society by increasing energy much lower net energy levels39–41. A good life could be achieved at production, even though they require more energy to support the lower per capita energy use by improving the efficiency of energy energy system (e.g. S5-R). using technologies (e.g. by replacing gasoline-powered cars with All pathways suggest an inevitable decline in per capita net energy electric cars), by shifting from consumption choices with higher at some point during the transition. However, this finding does not energy intensities to choices with lower energy intensities (e.g. from mean that energy scarcity is an unavoidable feature of any low-carbon cars to bicycles), and by avoiding the most inefficient alternatives energy transition. The projected net energy decline is not due to altogether (e.g. flying)42. Nature Communications | (2022)13:6932 7 Article https://doi.org/10.1038/s41467-022-33976-5 Fig. 5 | Analysis of factors affecting energy system emissions. The figure shows period following this push (2041–2100). Panel b shows average energy system the relationship between average energy system emissions and different factors for emissions in relation to the average share of energy generated from conventional each of the fourteen pathways compatible with 1.5 °C, for the median-EROI fossil fuels, over the two periods. Panel c shows average energy system emissions as assumption. Panel a shows average energy system emissions in relation to average a function of the average share of energy for the energy system for the two periods. annual energy use during the initial push for transition (2020–2040), and the Factors driving energy system emissions emissions (as shown by the wide scatter in the blue markers in Fig. 5c). Our results suggest that energy system emissions are substantial, and This is not to say that EROI is not a relevant factor, as a lower EROI increase as a share of total emissions over time. We use a panel data means that the energy system requires a larger share of total energy. analysis (Supplementary Table 9) to analyse the underlying factors However, a lower EROI may be counterbalanced by a lower share of behind energy system emissions during the transition while control- fossil fuels in the energy system (e.g. due to faster decarbonisation), or ling for heterogeneity across the fourteen analysed pathways over by lower energy use. time. Our analysis shows that energy system emissions depend on the growth in final energy use and the choice of energy technologies Discussion during the transition. A decrease in the overall EROI of the energy In this article, we have calculated the energy for the energy system, system contributes to an increase in energy system emissions during and the corresponding energy system emissions, for fourteen the initial push for the transition (from 2020 to 2040) but does not 1.5 °C–compatible mitigation pathways used extensively in the IPCC have a clear effect on emissions thereafter. The pathways that provide literature. Although energy for the energy system and energy system more energy to society have higher energy system emissions. emissions are implicitly accounted for in these pathways, they are not The different relationships in the two periods can be seen in Fig. 5. quantified and reported as separate quantities. By providing a separate The relationship between energy use and energy system emissions is picture of energy system emissions, we complement existing IAMs, particularly strong during the initial push for the transition (as shown yielding three core insights. by the orange markers in Fig. 5a). From 2020 to 2040, a 100 EJ increase First, we find that energy system emissions are substantial. On in annual energy use is associated with a 0.8 GtCO2 increase in annual average, we estimate that energy system emissions for a low-carbon energy system emissions. The relationship weakens after 2040 as the transition would amount to 195 GtCO2, which corresponds to ~5 years energy system is gradually decarbonised (see blue markers in Fig. 5a). of global CO2 emissions at their 2021 level. Based on the modelled From 2041 to 2100, the share of energy from fossil fuels is the most linear relationship between total cumulative emissions and global important factor contributing to energy system emissions (Fig. 5b and warming46, this figure implies that a low-carbon energy transition Supplementary Table 9). would lead to approximately 0.1 °C of additional global warming. In theory, energy system emissions could be decoupled from the Therefore, although the cumulative energy system emissions are scale of the energy system completely by fully substituting fossil fuels substantial, their overall climate impact is small compared to the with energy from renewables and nuclear energy5. However, such a amount of carbon saved over the long term by rapid decarbonisation. transition would require even more dramatic upscaling of these Second, we do not find a large jump in energy system emissions in technologies than currently assumed in the IPCC literature43. More- the short run from intensifying efforts to decarbonise the energy over, this upscaling could be constrained by other factors, such as the system. On the contrary, we find energy system emissions to be higher supply of materials required for energy infrastructure44,45. Such issues in pathways that decarbonise slowly, that use more fossil fuels to may be best addressed by improved models that explicitly calculate produce energy in the short term, and that rely on negative emission the energy and material requirements of the transition, beyond what is technologies to compensate for higher cumulative emissions (Fig. 3 covered by existing IAMs. and Supplementary Fig. 6). Contrary to previous concerns about Finally, we find a weak relationship between the share of energy emissions associated with a transition to renewable energy increasing for the energy system (defined as 1/EROISYS) and energy system emissions in the short run, we identify a longer-term problem in emissions from 2041 to 2100. During the latter years of the transition, pathways of slower decarbonisation and large-scale carbon removal, as the overall EROI of the energy system becomes a secondary factor for energy system emissions in these pathways continue well into the Nature Communications | (2022)13:6932 8 Article https://doi.org/10.1038/s41467-022-33976-5 future (Fig. 3). Although modest on their own, these emissions are system, and what remains for other socioeconomic activities outside comparable in magnitude to the residual emissions from aviation, steel of the transition. and cement production, and load-following electricity. Our results complement studies that find that carbon removal by BECCS is a much Methods less efficient mitigation approach than assumed in existing pathways, Energy transition pathways due to upstream emissions from biomass supply chains36,47 and land- Fourteen 1.5 °C–consistent (RCP1.9) mitigation pathways were selec- use change48,49. ted for this study. For illustrative purposes, we focus on four pathways Third, we find a comparable reduction in net energy, and in the (LED, S1-A, S2-M, and S5-R) from the IPCC’s Special Report on Global share of energy available to society, during the low-carbon transition Warming of 1.5 °C, which the IPCC selected as illustrative archetypes of to that found in previous studies12. However, reductions in our study alternative low-carbon transitions1. These transitions are model inter- tend to come earlier (within the first decade of efforts), especially for pretations of four distinct narratives that describe possible socio- mitigation pathways of fast decarbonisation. Pathways with faster economic and technological developments in a world that limits decarbonisation and lower energy demand have lower energy system climate change to 1.5 °C. The narratives are: Low Energy Demand emissions, but this comes at the cost of lower net energy for society. (LED)31, Sustainability (S1)58, Middle of the Road (S2)59, and Fossil- Lower net energy does not need to lead to energy scarcity. A consensus fuelled Development (S5)60. is emerging regarding the enormous potential to use energy more To capture a wider range of assumptions and modelling frame- efficiently50,51, and the possibilities of providing a decent life with much works beyond the four illustrative pathways, we complement the less energy than is currently consumed in wealthy nations39,41,52. analysis with ten additional pathways, which were produced in the In general, our study demonstrates the importance of calculating same study as S1-A, S2-M, and S5-R by Rogelj et al.30. Five of these are the energy requirements and emissions associated with the transition, modelling representations of S1, three of S2, and one each of S4 and S5. to get a more complete picture of energy system dynamics and to S4 is also known as the “world of deepening inequality” narrative61. quantify the remaining emissions available to society. Further research could explore the energy required and emissions associated with the Energy requirements and EROI replacement of machines and infrastructure at the consumption end We estimate the energy for the energy system during transition by of the energy system (e.g. electric vehicles, their charging stations, and applying the analytical framework of Energy Return on Investment energy storage solutions). Calculating such emissions would be (EROI) at the final energy stage. EROI describes a ratio between the worthwhile, as the transformation of the consumption end of the amount of net energy delivered to society (E NET ) and the total amount energy system could potentially take up a large part of the remaining of energy that is required to extract, convert, and deliver this energy carbon budget for 1.5 °C. (E REQ ), which we also refer to as the energy for the energy system23,62. The research on energy transitions should go beyond the sce- EROI is a measure of energy system efficiency, as it compares the narios produced by IAMs, and also include scenarios from alternative amount of energy that enters the productive economy with the energy “normative” energy modelling approaches5,43,53. IAMs focus on optimal- that is associated with total (gross) energy production14,63. The lower cost pathways of decarbonisation, and therefore do not cover the the EROI, the greater the energy requirements, and the lower the net whole range of possible energy transitions54. IAMs have been found to energy that is available for productive socioeconomic activities (see be biased towards technologies that are direct substitutes for con- Eq. 1). ventional fossil fuels, such as BECCS and fossil fuels with carbon cap- ture and storage55, which is why they tend to underestimate the E NET E GROSS  E REQ E GROSS EROI = = = 1 ð1Þ realistic deployment potential of intermittent renewables56. E REQ E REQ E REQ In our analysis, we find that a preference for direct substitutes for conventional technologies leads to higher energy system emis- We define the system boundaries of our EROI analysis at the final sions and lower net energy. A discussion of whether a low-energy energy boundary, also known as the point-of-use boundary, which transition based on renewables would be a preferable mitigation describes the point where energy carriers enter the productive strategy is beyond the scope of this article. However, our analysis economy64. EROIFIN includes all of the direct inputs along the energy suggests that explicit modelling of energy system emissions and supply chain required to extract (E EXT ) and refine energy resources dynamic EROIs from different energy technologies could add sup- (E REF ), the energy used to transport the energy from the primary port to the case for renewables over technologies relying on carbon energy stage to the point of use for society (E TRA ), as well as the energy capture and storage. requirements associated with construction (E CON ), decommissioning Questions remain regarding the extent to which the production- (E DEC ), and operation and maintenance of energy infrastructure based approach of IAMs accounts for the upstream emissions asso- (E O&M ), such as power plants and refineries, as shown in Eq. 224: ciated with different energy generation technologies21, and also the extent to which IAMs capture effects from changes in the EROI of the E REQ = E EXT + E REF + E TRA + E CON + E O&M + E DEC ð2Þ energy mix57. Further research should explore the possibility of inte- grating EROI analysis into IAMs to produce internally more consistent Energy for construction refers to the energy that is used to man- energy and emissions pathways, which would likely change the mod- ufacture and build energy infrastructure like power plants and refi- els’ choice of energy generation technologies. Such integration could neries. Energy for decommissioning accounts for the energy required involve EROI scenarios tailored to the narratives of the mitigation to dismantle, remove, and dispose of obsolete energy infrastructure. pathways, and link EROI calculations to specific narrative assumptions Energy for operation and maintenance includes energy used to extract (e.g. about technological change, international cooperation, land-use, primary energy resources, and the energy required to convert primary and innovation). energy into useful energy carriers and deliver them to the end-user. Overall, our study demonstrates the importance of accounting for Energy for operation and maintenance also includes all of the energy net energy and energy system emissions. Future mitigation pathways inputs for the energy industry’s own use, from the primary to the final would be improved by explicitly modelling the energy requirements energy stage. and emissions associated with a low-carbon energy transition. Doing By convention, energy conversion losses from primary to final so would allow us to better understand the trade-off between the energy and energy losses in distribution, transmission, and storage energy and carbon required to transition to a low-carbon energy (E LOSS ) are not counted among the energy requirements of energy Nature Communications | (2022)13:6932 9 Article https://doi.org/10.1038/s41467-022-33976-5 conversion technologies10. These losses are already accounted for in We calculate the energy for the operation and maintenance of the energy balances of the original data from the pathways and result energy infrastructure as a product of the energy intensity of capital in lower final energy relative to total energy generation. Moreover, the and the operation and maintenance costs per unit of generated energy energy requirements do not include the raw energy embodied in (C O&M ) multiplied by the total energy generated per year (E GEN ), as energy resources (e.g. the heating value of gas) that are to be con- shown in Eq. 5: verted into useful carriers. The energy requirements only account for energy inputs that are needed to procure and process the resources E O&M ðt Þ = ε × C O&M × E GEN ðtÞ ð5Þ into useful energy carriers, and to deliver these carriers to the end- user. See Supplementary Fig. 7 for a complete illustration of our energy In estimating the energy required for the decommissioning of system boundaries and the representation of energy flows from pri- energy infrastructure at the end of its life we apply the assumption of mary energy sources to net energy. Hertwich et al.35, who estimates that decommissioning represents We estimate the energy requirements for twenty-seven energy roughly 10% of the energy required for construction (see Eq. 6). conversion technologies that are represented in the mitigation path- ways (see Supplementary Figs. 1 and 2 for a detailed overview of all E DEC ðtÞ = 0:1 × E CON ðt  τÞ ð6Þ energy conversion technologies in our model). These technologies describe different pathways of energy conversion from fossil fuels and In the following steps, we describe the calculation of the energy biomass alongside energy generated from non-biomass renewables, requirements of processes for obtaining raw fuels before they are nuclear energy, and hydrogen. In our calculations of energy for the refined into useful energy carriers that can be delivered to end-users. energy system, we distinguish between four types of energy carriers To estimate the energy used in the extraction, mining, or har- that are represented in the mitigation pathways: electricity, refined vesting of raw fuels, we collect a series of present-day EROI estimates at liquid fuels, gases, and solids (coal and combustible biomass). the standard energy system boundary28 (e.g. farm-gate or mine-mouth; denoted EROIST) from the peer-reviewed literature, as listed in Sup- Energy requirements of fossil fuel and biomass technologies plementary Table 11. From the EROIST values of these selected studies, For energy conversion from fossil fuels and biomass, the main energy we calculate the lower, median, and upper inter-quartile range of the requirements are associated with the extraction, processing, and EROIST for each energy resource and use these values to determine a delivery of energy resources, whereas the construction, decom- range of estimated energy requirements associated with the appro- missioning, and operation and maintenance of the energy infra- priation of raw energy fuels. We assume the EROIST of fossil fuels will structure represent only a small share of total energy requirements25. continue to decline over time. We model the decline by following the By contrast, for non-biomass renewables, almost all energy require- approach of Dale et al.29 and Sgouridis et al.5, who use the equation of ments are from upfront energy demand for the construction of energy exponential decline from present-day values EROIST(0) shown in Eq. 7. infrastructure. This approach models the convergence of EROIST towards the mini- We estimate the energy requirements associated with construc- mum EROI (EROIST,low), which we assume corresponds to the lower tion as upfront energy invested during the first year of the energy inter-quartile range of present-day EROI estimates. The rate of decline facility’s lifetime. Similarly, the energy required for decommissioning (βC ) for each respective resource is calibrated from the historical trend is accounted for at the end of the energy infrastructure lifetime. The for the EROIST of fossil fuels, as published by Brockway et al.14. remaining energy inputs that are associated with energy system operations are counted every year during the lifetime of the energy EROIST ðt Þ = EROIST,low + ðEROIST ð0Þ  EROIST,low Þ × expβC × t ð7Þ infrastructure. To calculate the energy requirements to build, decommission, EROIST compares the raw energy content of energy resources and operate and maintain the energy infrastructure, we follow the such as wood, coal, gas, and crude oil (E RAW ) with the energy required previous work of Sgouridis et al.65. We calculate the energy required for to obtain these fuels (E EXT ; see Eq. 8), before they are converted into the construction and the energy embodied in the energy generation useful energy carriers. The efficiency of energy conversion (ηC ) machinery by estimating the energy intensity of capital (ε) and multi- depends on the respective energy conversion technology and may plying it first by the capital costs of infrastructure per unit of installed change over time. In this study, we apply the energy conversion power (C p ), and second by the newly installed power capacity in the coefficients from the representation of energy technologies in the respective year (P NEW ), as shown in Eq. 3. For infrastructure capital REMIND model66,68. The model assumes energy conversion efficiency costs, we use values from the REMIND IAM documentation66, which are in new energy infrastructure improves over time. We combine Eq. 8 provided in $US2015. We estimate the energy intensity of capital at and 9 to obtain an expression that links the energy requirements of 4.52 TJ/million $US2015, after adjusting for inflation the estimate of extraction (or harvest or mining) to the efficiency of energy conversion 5.49 TJ per million $US2007 from the abovementioned study, using and EROIST (Eq. 10). the producer price index from the PCU3336 industry group data67. Values of the parameters for different energy conversion technologies E RAW ðt Þ EROIST ðt Þ = ð8Þ are listed in Supplementary Table 10. E EXT ðt Þ E CON ðtÞ = ε × C p × P NEW ðtÞ ð3Þ E GEN ðtÞ = ηC ðtÞ × E RAW ðtÞ ð9Þ In our calculations of energy requirements associated with new E GEN ðt Þ energy infrastructure, we include the power capacity built to increase E EXT ðtÞ = ð10Þ ηC ðtÞ × EROIST energy production as well as the capacity that replaces the infra- structure that is decommissioned at the end of its lifetime (τ), as shown In estimating the energy required for the refining or processing of in Eq. 4: fuels (E REF ) we refer to the calculations from previous studies. For the  refining of crude oil, we use the estimates of energy intensity of maxð0, P ðt Þ  P ðt  1ÞÞ; t < τ P NEW =   ð4Þ refining in MJ per kg (μREF ) from Raugei and Leccisi63 and the “Ecoin- max 0, P ðt Þ  P ðt  1Þ + P NEW ðt  τÞ ; t ≥ τ vent Life-cycle Inventories of Oil Refinery Processing”69. For the Nature Communications | (2022)13:6932 10 Article https://doi.org/10.1038/s41467-022-33976-5 processing of raw fuels from biomass, we use estimates of energy Energy requirements of non-biomass renewables and nuclear intensity from an extensive literature review by Fajardy et al.36,37. We energy define the energy used in refining as a product of the mass of the The largest energy requirements of non-biomass renewables (i.e. solar respective fuel and the energy intensity of refining, as shown in Eq. 11. photovoltaics, wind, geothermal, and hydropower) are related to the We calculate the mass of the fuel by dividing the raw energy content of manufacturing and construction of energy infrastructure25,35. For energy resources (E RAW ) by the higher heating value (HHV), described renewables, the energy required for operation is much lower than by Eq. 12. We do not assume specific energy requirements for the technologies that produce energy from raw fossil fuels, as renewable processing of natural gas and coal, consistent with previous EROI and sources do not require energy to be extracted, transported, and pro- life-cycle studies63,70. cessed. For nuclear energy, the energy to maintain the energy supply chain also includes energy requirements for the extraction, enrich- ment, and transportation of uranium. Here, the energy requirements E REF ðtÞ = M FUEL ðtÞ × μREF ð11Þ for operating energy infrastructure and maintaining the energy supply are substantially higher compared to the construction of energy infrastructure. E RAW ðt Þ M FUEL ðt Þ = ð12Þ To obtain estimates of the energy requirements of renewables HHV and nuclear energy over the lifetime of each technology, we collected a series of present-day EROI estimates for each technology at the final To calculate the energy requirements for transportation, we energy boundary, from a number of peer-reviewed studies (see the assess global trade routes of coal, gas, and crude oil in the year 2019, by studies listed in Supplementary Table 1). From these studies, we cal- using the flows of these fuels from the international trade balance culated the lower, median, and upper quartiles of the range of EROI sheets of the BP Statistical Review of World Energy 2020. For biomass, values for each energy source. These quartiles are classified as low, we use data on the global flows of wood pellets from Junginger et al.71. median, and high-EROI estimates. We partition the trade routes (indexed with l) into different stages by We divided the energy requirements between the energy required transportation type, estimating the average trade distance in each for construction (E CON ) and decommissioning (E DEC ) of the energy route. For example, the oil route from Baghdad (Iraq) to Houston infrastructure, and the annual energy requirements to operate the (USA) consists of an onshore pipeline of 970 km from Baghdad to energy infrastructure and maintain the energy supply (E O&M ), follow- Ceyhan (Turkey), a sea freight route of 12,500 km from Ceyhan to ing the approach of King and van den Bergh12. Energy requirements for Houston, and an onshore pipeline of 100 km on the US mainland. We operation are proportional to the total installed power (P) times the assume that the energy intensities of fuel transportation types remain capacity factor (CF) divided by the EROI of the technology12, as shown constant over time. in Eq. 16. CF is a dimensionless ratio that compares the actual annual We calculate the energy used in each stage of transportation generation of energy to the maximum potential energy output. The (indexed with j) by multiplying the amount of fuel transported by the parameter αtech describes the ratio between the energy requirements energy intensity of the transportation type and the distance over which of operation and the energy invested in construction over the lifetime the fuel is transported, as shown in Eq. 13. The parameters for the of the technology. transportation types are obtained from the life-cycle inventory data- base EcoInvent v3.272, and can be found among the parameters listed in αtech × PðtÞ × CFðtÞ Supplementary Table 10. E O&M ðtÞ = ð16Þ EROIðtÞ M FUEL,l × γTRA,j × distancel,j ð13Þ As described in Eq. 17, the energy requirements of construction are proportional to newly installed power (P NEW ), times the capacity factor of the respective energy conversion technology (CF), multiplied To estimate the average global energy intensity (ϵTRA ) associated by the lifetime of the technology (τ), and divided by the technology’s with the transportation of each fuel (in MJ/kg), we sum the energy use EROI. across the global trade routes and divide the sum by the global volume of trade flows (in tonne kilometres), defined as the global sum of ð1  α tech Þ × P NEW × CF × τ transported fuel multiplied by the distance, as described in Eq. 14: E CON ðtÞ = ð17Þ EROIðtÞ ! P P Energy associated with decommissioning is assumed to represent M FUEL,l × γTRA,j × distancel,j l j 10% of the energy used for construction, following Hertwich et al.35 ϵTRA = ! ð14Þ Energy associated with decommissioning is accounted for in the last P P M FUEL,l × distancel,j year of the energy infrastructure’s lifetime, as shown in Eq. 18: l j E DEC ðtÞ = 0:1 × E CON ðt  τÞ ð18Þ Finally, we obtain the energy required for the transportation of raw fuel by multiplying the mass of the fuel transported by the average We assume the historical trend of increasing EROIs of photo- global energy intensity of fuel transportation for each respective fuel, voltaic and wind power technologies will continue in the future. We as shown in Eq. 15: model the EROI dynamics of these technologies by applying “energetic experience curves” (see Steffen et al.16 and Loueven et al.17), thus esti- mating the reduction in the energy requirements for construction, and E TRA ðt Þ = M FUEL ðtÞ × ϵTRA ð15Þ operation and maintenance due to technological innovation. For a detailed explanation of how we calculated the future dynamics of EROI For a complete overview of our assumptions regarding the trade for photovoltaics, wind power, and hydrogen from electrolysis, see the routes of coal, natural gas, crude oil, and biomass, and our calculations “Note on EROI dynamics of wind and solar power” and the “Note on of the energy intensities of fuel transport, see Supplementary energy requirements for hydrogen from electrolysis” in the Supple- Tables 12–15. mentary Information. In estimating the energy requirements of Nature Communications | (2022)13:6932 11 Article https://doi.org/10.1038/s41467-022-33976-5 hydropower, geothermal, and nuclear energy, we refer to the present- requirements of the energy system and the carbon intensity of the day range of EROI estimates, due to a lack of studies on EROI dynamics energy for the energy system. Energy system emissions decrease with for these technologies. the decarbonisation of the energy supply and a reduction in energy requirements. EROI of the overall energy system To calculate the emissions associated with the construction of We calculate the EROI of the overall energy system at the final energy energy technologies, and the operation and maintenance of the energy stage, by applying Eq. 1, wherein we compare the total amount of gross supply over time, we first separate the energy requirements associated final energy production to the sum of the energy requirements for all with the construction, decommissioning, and operation and main- energy conversion technologies (here represented by the index i), as tenance of the energy supply for each energy technology into the four shown in Eq. 19. We use the same approach to calculate the EROI of energy carriers: electricity, gases, liquid fuels, and solids. This step is individual energy conversion technologies and the EROI of different crucial to adequately quantify energy system emissions, as the carbon carriers, such as the EROI of electricity from renewables. intensities of different energy carriers can differ substantially21, espe- cially given that electricity can be decarbonised much faster than other E GROSS carriers. We count hydrogen among the liquid fuels, assuming that EROISYS = P 1 ð19Þ E EXT,i + E REF,i + E TRA,i + E CON,i + E O&M,i + E DEC,i most hydrogen will be destined to replace liquid fossil fuels. In i decomposing the energy requirements into different energy carriers, We test our model by comparing our estimates of the EROI of the we follow the approach of Arvesen et al.25, who distinguish between the overall energy system with the results from the EROI literature. The four abovementioned energy carrier types, for each of the three life- note on “EROI estimates of different energy carriers” in the Supple- cycle assessment phases of construction, decommissioning, and mentary Information demonstrates that our calculations of the EROI at operation (see Eq. 20). The life-cycle phase of operation and main- the final energy stage are consistent with estimates from previous tenance includes both the energy requirements to operate the energy studies. infrastructure as well as the energy required to maintain the energy EROI values differ greatly depending on the energy system supply. We use the life-cycle assessment database from Arvesen et al. boundaries that the analyst uses22,64. For example, some studies mea- to decompose the energy requirements into fourenergy carriers by sure energy delivered at the point of energy extraction, while others multiplying the total energy requirements by the vector of the calculate energy delivered to the end-user, which is an expanded respective energy carriers shares, composed of electricity (e), gases analytical boundary of the system. Expanding the boundary results in (g), liquid fuels (l), and solids (s): lower EROI values, as it includes the additional energy required to convert the raw resource into useful energy and move or store it. We EREQ,i ðtÞ = E REQ,i ðt Þ × he,g,l,sii ð20Þ selected studies to match a consistent system boundary, which includes the energy investments for energy resource extraction, Energy requirements, decomposed into four energy carriers, are resource transportation, resource processing, the construction of multiplied by the carbon intensity vector containing the carbon energy conversion facilities, and the energy required for the operation intensities of energy carriers (CI), to obtain the energy system emis- of the facilities. sions from each respective energy generation technology, and the life- We assume a global average EROI for each energy conversion cycle phase, as shown in Eq. 21. technology. We do not take into account regional differences in pro- duction and transformation processes28. However, the EROI of the CO2,i ðt Þ = EREQ,i ðt Þ  CIðtÞ ð21Þ entire energy system does change with improvements in energy con- version efficiencies, changes in the EROIST of fossil fuels due to a We calculate the carbon intensity of each carrier (c) by dividing declining abundance of these energy resources, and as the mix of the total carbon emissions from energy generation for each carrier by energy technologies change over time. the total amount of energy generated by each carrier, as shown in In our EROI scenarios, high-EROI values assumed for each energy Eq. 21. Changes in the carbon intensities of energy carriers in the technology are based on studies with favourable assumptions mitigation pathways over time are depicted in Supplementary Figs. 8 regarding resource abundance and deployment of the most efficient and 9. low-carbon energy-generating technologies. Low-EROI values, in turn, P assume lower resource abundance and limited technological CO2,i,c ðtÞ i improvement of low-carbon energy technologies. Median-EROI values CIc ðtÞ = P ð22Þ E GEN,i,c ðtÞ represent a balanced, middle-of-the road EROI trajectory. For a i detailed overview of EROI assumptions for different energy technol- ogies, see Supplementary Tables 3–7. Emissions from electricity generation are obtained directly from the original scenario data, whereas emissions from gases, liquids, and Net energy solids are calculated using the carbon intensities of energy conversion To calculate net energy per capita, we divide the difference between technologies (φi ), which are endogenous to the REMIND model (see gross final energy and the total energy requirements of the energy Eq. 23). For an overview of the carbon intensity parameters, see Sup- system (as shown in Eq. 1), by the global population projections in the plementary Table 10. mitigation pathways. CO2,i,c ðtÞ = φi × E GEN,i,c ðtÞ ð23Þ Energy system emissions Estimating the carbon emissions associated with the build-up of the Cumulative energy system emissions are calculated as the sum of energy system and the operation and maintenance of the energy annual emissions from all of the energy generation technologies, over supply during transition is crucial for assessing different mitigation the period from 2020 to 2100. The share of energy emissions that is pathways. If a substantial amount of the remaining carbon budget goes shown in Fig. 2 is calculated by dividing energy system emissions by to decarbonising the energy system, this may significantly affect the the total carbon emissions from energy and industrial processes. projections of energy use and emissions in the end-use energy sectors. We report negative emissions, realised by BECCS technologies, Future energy system emissions depend on changes in the energy separately from (positive) anthropogenic emissions. Negative Nature Communications | (2022)13:6932 12 Article https://doi.org/10.1038/s41467-022-33976-5 emissions realised in energy generation are therefore not counted in and related global greenhouse gas emission pathways, in the con- the calculation of the carbon intensities of the four energy carriers. We text of strengthening the global response to the threat of climate refer to the negative emissions data, as they are reported in the original change, sustainable development, and efforts to eradicate poverty. scenario data. Mitigation pathways report the total amount of (eds. Masson-Delmotte, V. et al.). 616 (Cambridge University Press, sequestered carbon by BECCS technologies (gross negative emis- Cambridge, UK and New York, NY, USA, 2022). sions), but do not separately report the positive emissions from 2. Daly, H. E., Scott, K., Strachan, N. & Barrett, J. Indirect CO2 emission BECCS. Positive emissions from BECCS include the emissions from implications of energy system pathways: linking IO and TIMES land-use change, the emissions from fertilisers, the emissions asso- models for the UK. Environ. Sci. Technol. 49, 10701–10709 (2015). ciated with the construction and operation of the BECCS energy 3. McDowall, W., Solano Rodriguez, B., Usubiaga, A. & Acosta Fer- facilities, and the carbon emitted along the biomass supply chain73. nández, J. Is the optimal decarbonization pathway influenced by Mitigation pathways use different reporting methodologies for indirect emissions? Incorporating indirect life-cycle carbon dioxide the carbon removal by BECCS that, in some cases, combine gross emissions into a European TIMES model. J. Clean. Prod. 170, carbon removal with removals in the land-use sector74. This makes it 260–268 (2018). difficult to include the total positive emissions associated with BECCS 4. Scott, K., Daly, H., Barrett, J. & Strachan, N. National climate policy in our energy system calculations. Here, we limit our analysis of energy implications of mitigating embodied energy system emissions. system emissions to the emissions associated with the biomass supply Clim. Change 136, 325–338 (2016). and emissions from BECCS facilities, though emissions from land-use- 5. Sgouridis, S., Csala, D. & Bardi, U. The sower’s way: quantifying the change and fertiliser use may be an even bigger source of energy narrowing net-energy pathways to a global energy transition. system emissions47,48. As a result, our estimates may considerably Environ. Res. Lett. 11, 094009 (2016). underestimate the energy system emissions in low-carbon energy 6. Carbajales-Dale, M., Barnhart, C. J., Brandt, A. R. & Benson, S. M. A transitions that assume a large-scale use of bioenergy, such as the S2 better currency for investing in a sustainable future. Nat. Clim. and S5 mitigations pathways. Chang. 4, 524–527 (2014). 7. Heinberg, R. & Fridley, D. Our Renewable Future: Laying the Path for Multiple regression panel data analysis One Hundred PercClean Energy. (Island Press, 2016). To quantify the factors driving energy system emissions, we selected 8. Di Felice, L. J., Ripa, M. & Giampietro, M. Deep decarbonisation from three independent variables: final energy use, the share of energy from a biophysical perspective: GHG emissions of a renewable electricity conventional fossil fuels, and the share of energy for the energy sys- transformation in the EU. Sustainability 10, 41–57 (2018). tem. Panel OLS multiple regression analysis was used to estimate the 9. Trainer, T. Estimating the EROI of whole systems for 100% renew- contribution of each of these factors. The estimated model is as fol- able electricity supply capable of dealing with intermittency. lows: Energy Policy 119, 648–653 (2018). 10. Capellán-Pérez, I., de Castro, C. & Miguel González, L. J. Dynamic CO2,k ðt Þ = γðt Þ + β1 x 1,k ðt Þ + β2 x 2,k ðt Þ + β3 x 3,k ðt Þ + εk ðtÞ ð24Þ Energy Return on Energy Investment (EROI) and material require- ments in scenarios of global transition to renewable energies. where CO2,k ðt Þ is the energy system emissions for pathway k in year t, β Energy Strateg. Rev. 26, 100399 (2019). gives the coefficients for the three independent variables x, and γ is the 11. Jackson, A. & Jackson, T. Modelling energy transition risk: the time-specific term that controls for unobserved heterogeneity over impact of declining energy return on investment (EROI). Ecol. Econ. time, and εk is the error term. Time fixed-effects were included in the 185, 107023 (2021). model, given that we are interested in how the relationship between 12. King, L. C. & Van Den Bergh, J. C. J. M. Implications of net energy- independent variables and energy system emissions varies between return-on-investment for a low-carbon energy transition. Nat. different pathways. Robust standard errors controlling for hetero- Energy 3, 334–340 (2018). skedasticity and autocorrelation were estimated after testing for their 13. Sers, M. R. & Victor, P. A. The Energy-missions trap. Ecol. Econ. 151, presence in the balanced panel dataset. 10–21 (2018). 14. Brockway, P. E., Owen, A., Brand-Correa, L. I. & Hardt, L. Estimation Data availability of global final stage energy-return-on-investment for fossil fuels Energy, capacity, emissions, and population data for energy transition with comparison to renewable energy sources. Nat. Energy 4, pathways were obtained from the IAMC 1.5 °C Scenario Explorer 612–621 (2019). repository (release 2.0)75, available at: https://data.ene.iiasa.ac.at/ 15. Diesendorf, M. & Wiedmann, T. Implications of trends in energy iamc-1.5c-explorer/. Data for the LED pathway31 were obtained return on energy invested (EROI) for transitioning to renewable from the LED database: https://db1.ene.iiasa.ac.at/LEDDB/. Historical electricity. Ecol. Econ. 176, 106726 (2020). data on gross final energy from different energy carriers were taken 16. Steffen, B., Hischier, D. & Schmidt, T. S. Historical and projected from the online IEA Data and Statistics database, accessible at: https:// improvements in net energy performance of power generation www.iea.org/data-and-statistics. The data for energy system emis- technologies. Energy Environ. Sci. 11, 3524–3530 (2018). sions, energy requirements for the energy system, net energy, and 17. Louwen, A., Van Sark, W. G. J. H. M., Faaij, A. P. C. & Schropp, R. E. I. EROI generated in this study have been made available in the Re-assessment of net energy production and greenhouse gas OSF online data repository: https://osf.io/v5nqg/?view_only= emissions avoidance after 40 years of photovoltaics development. d28f6be45dc44dec884b0afa59098b76. Nat. Commun. 7, 1–9 (2016). 18. Berrill, P., Arvesen, A., Scholz, Y., Gils, H. C. & Hertwich, E. G. Code availability Environmental impacts of high penetration renewable energy sce- The Octave code used to calculate energy system emissions and narios for Europe. Environ. Res. Lett. 11, 014012 (2016). energy for the energy system will be made available by the corre- 19. Babacan, O. et al. Assessing the feasibility of carbon dioxide miti- sponding author upon reasonable request. gation options in terms of energy usage. Nat. Energy 5, 720–728 (2020). References 20. Raugei, M., Kamran, M. & Hutchinson, A. A Prospective net energy 1. IPCC, 2018: Global warming of 1.5 °C. in An IPCC Special Report on and environmental life-cycle assessment of the UK electricity grid. the impacts of global warming of 1.5 °C above pre-industrial levels Energies 13, 2207 (2020). Nature Communications | (2022)13:6932 13 Article https://doi.org/10.1038/s41467-022-33976-5 21. Pehl, M. et al. Understanding future emissions from low-carbon 46. Damon Matthews, H. et al. An integrated approach to quantifying power systems by integration of life-cycle assessment and inte- uncertainties in the remaining carbon budget. Commun. Earth grated energy modelling. Nat. Energy 2, 939–945 (2017). Environ. 2, 1–11 (2021). 22. Murphy, D. J., Hall, C. A. S., Dale, M. & Cleveland, C. Order from 47. Creutzig, F. et al. The mutual dependence of negative emission chaos: a preliminary protocol for determining the EROI of fuels. technologies and energy systems. Energy Environ. Sci. 12, Sustainability 3, 1888–1907 (2011). 1805–1817 (2019). 23. Murphy, D. J. & Hall, C. A. S. Year in review-EROI or energy return on 48. Harper, A. B. et al. Land-use emissions play a critical role in land- (energy) invested. Ann. N. Y. Acad. Sci. 1185, 102–118 (2010). based mitigation for Paris climate targets. Nat. Commun. 9, 24. Castro, C. de & Capellán-Pérez, I. Standard, point of use, and 2938 (2018). extended energy return. Energies 13, 3036 (2020). 49. Heck, V., Gerten, D., Lucht, W. & Popp, A. Biomass-based negative 25. Arvesen, A., Luderer, G., Pehl, M., Bodirsky, B. L. & Hertwich, E. G. emissions difficult to reconcile with planetary boundaries. Nat. Deriving life cycle assessment coefficients for application in inte- Clim. Chang. 8, 151–155 (2018). grated assessment modelling. Environ. Model. Softw. 99, 50. Cullen, J. M., Allwood, J. M. & Borgstein, E. H. Reducing energy 111–125 (2018). demand: what are the practical limits? Environ. Sci. Technol. 45, 26. Palmer, G. & Floyd, J. An exploration of divergence in EPBT and EROI 1711–1718 (2011). for solar photovoltaics. Biophys. Econ. Resour. Qual. 2, 1–20 (2017). 51. Jacobson, M. Z., Delucchi, M. A., Cameron, M. A. & Mathiesen, B. V. 27. Palmer, G. Renewables rise above fossil fuels. Nat. Energy 4, Matching demand with supply at low cost in 139 countries among 538–539 (2019). 20 world regions with 100% intermittent wind, water, and sunlight 28. Raugei, M. Net energy analysis must not compare apples and (WWS) for all purposes. Renew. Energy 123, 236–248 (2018). oranges. Nat. Energy 4, 86–88 (2019). 52. Oswald, Y., Owen, A. & Steinberger, J. K. Large inequality in inter- 29. Dale, M., Krumdieck, S. & Bodger, P. A dynamic function for energy national and intranational energy footprints between income return on investment. Sustainability 3, 1972–1985 (2011). groups and across consumption categories. Nat. Energy 5, 30. Rogelj, J. et al. Scenarios towards limiting global mean temperature 231–239 (2020). increase below 1.5 °C. Nat. Clim. Chang. 8, 325–332 (2018). 53. Warszawski, L. et al. All options, not silver bullets, needed to limit 31. Grubler, A. et al. A low energy demand scenario for meeting the 1.5 global warming to 1.5 °C: a scenario appraisal. Environ. Res. Lett. 16, °C target and sustainable development goals without negative 064037 (2021). emission technologies. Nat. Energy 3, 515–527 (2018). 54. Ackerman, F., Decanio, S. J., Howarth, R. B. & Sheeran, K. Limitations 32. Peters, G. P. Beyond carbon budgets. Nat. Geosci. 11, of integrated assessment models of climate change. Clim. Change 378–380 (2018). 95, 297–315 (2009). 33. Davis, S. J. et al. Net-zero emissions energy systems. Science 360, 55. Kaya, A., Csala, D. & Sgouridis, S. Constant elasticity of substitution eaas9793 (2018). functions for energy modeling in general equilibrium integrated 34. Luderer, G. et al. Residual fossil CO2 emissions in 1.5-2 °C pathways. assessment models: a critical review and recommendations. Clim. Nat. Clim. Chang. 8, 626–633 (2018). Change 145, 27–40 (2017). 35. Hertwich, E. G. et al. Integrated life-cycle assessment of electricity- 56. Pietzcker, R. C. et al. System integration of wind and solar power in supply scenarios confirms global environmental benefit of low- integrated assessment models: a cross-model evaluation of new carbon technologies. Proc. Natl Acad. Sci. USA 112, approaches. Energy Econ. 64, 583–599 (2017). 6277–6282 (2015). 57. Palmer, G. A biophysical perspective of IPCC integrated energy 36. Fajardy, M. & Mac Dowell, N. Can BECCS deliver sustainable and modelling. Energies 11, 1–17 (2018). resource efficient negative emissions? Energy Environ. Sci. 10, 58. van Vuuren, D. P. et al. Energy, land-use and greenhouse gas 1389–1426 (2017). emissions trajectories under a green growth paradigm. Glob. 37. Fajardy, M. & Mac Dowell, N. The energy return on investment of Environ. Chang. 42, 237–250 (2017). BECCS: Is BECCS a threat to energy security? Energy Environ. Sci. 59. Fricko, O. et al. The marker quantification of the Shared Socio- 11, 1581–1594 (2018). economic Pathway 2: a middle-of-the-road scenario for the 21st 38. Fabre, A. Evolution of EROIs of electricity until 2050: estimation and century. Glob. Environ. Chang. 42, 251–267 (2017). implications on prices. Ecol. Econ. 164, 1–12 (2019). 60. Kriegler, E. et al. Fossil-fueled development (SSP5): an energy and 39. Millward-Hopkins, J., Steinberger, J. K., Rao, N. D. & Oswald, Y. resource intensive scenario for the 21st century. Glob. Environ. Providing decent living with minimum energy: a global scenario. Chang. 42, 297–315 (2017). Glob. Environ. Chang. 65, 102168 (2020). 61. Calvin, K. et al. The SSP4: a world of deepening inequality. Glob. 40. Kikstra, J. S., Mastrucci, A., Min, J., Riahi, K. & Rao, N. D. Decent living Environ. Chang. 42, 284–296 (2017). gaps and energy needs around the world. Environ. Res. Lett. 16, 62. Hall, C. A. S., Balogh, S. & Murphy, D. J. R. What is the minimum EROI 095006 (2021). that a sustainable society must have? Energies 2, 25–47 (2009). 41. Vogel, J., Steinberger, J. K., O’Neill, D. W., Lamb, W. F. & Krishna- 63. Raugei, M. & Leccisi, E. A comprehensive assessment of the energy kumar, J. Socio-economic conditions for satisfying human needs at performance of the full range of electricity generation technologies low energy use: an international analysis of social provisioning. deployed in the United Kingdom. Energy Policy 90, 46–59 Glob. Environ. Chang. 69, 102287 (2021). (2016). 42. Creutzig, F. et al. Towards demand-side solutions for mitigating 64. Hall, C. A. S., Lambert, J. G. & Balogh, S. B. EROI of different fuels climate change. Nat. Clim. Chang. 8, 268–271 (2018). and the implications for society. Energy Policy 64, 141–152 (2014). 43. Keyßer, L. T. & Lenzen, M. 1.5 °C degrowth scenarios suggest the 65. Sgouridis, S., Carbajales-Dale, M., Csala, D., Chiesa, M. & Bardi, U. need for new mitigation pathways. Nat. Commun. 12, 1–16 (2021). Comparative net energy analysis of renewable electricity and car- 44. Garcia-Olivares, A., Ballabrera-Poy, J., Garcia-Ladona, E. & Turiel, A. bon capture and storage. Nat. Energy 4, 456–465 (2019). A global renewable mix with proven technologies and common 66. Luderer, G. et al. Description of the REMIND Model (Version 1.6). materials. Energy Policy 41, 561–574 (2012). Model Documentation Document. https://doi.org/10.2139/ssrn. 45. Sprecher, B. & Kleijn, R. Tackling material constraints on the expo- 2697070 (2015). nential growth of the energy transition. One Earth 4, 67. U.S. Bureau of Labor Statistics. Producer Price Index by Industry: 335–338 (2021). Turbine and Power Transmission Equipment Manufacturing. Nature Communications | (2022)13:6932 14 Article https://doi.org/10.1038/s41467-022-33976-5 Producer Price Index. https://fred.stlouisfed.org/series/ the results, under the supervision of G.K. and D.W.O. G.K. had the initial PCU33363336. (2019). idea for the project, discussed and refined the results with A.S., and 68. Bauer, N. et al. REMIND: The equations, Potsdam Institute of Climate contributed to the writing of the article. D.W.O. contributed to the ana- Impact Research Publication. (2011). lysis and discussion of the results, the preparation of visuals, and the 69. Meili, C., Niels, J. & Wenzel, P. Life cycle inventories of long-distance writing of the article. transport of crude oil. (ESU-services Ltd., Schaffhausen, 2018). 70. Raugei, M., Leccisi, E., Fthenakis, V., Escobar Moragas, R. & Simsek, Competing interests Y. Net energy analysis and life cycle energy assessment of elec- The authors declare no competing interests. tricity supply in Chile: present status and future scenarios. Energy 162, 659–668 (2018). Additional information 71. Junginger, H. M. et al. The future of biomass and bioenergy Supplementary information The online version contains deployment and trade: a synthesis of 15 years IEA Bioenergy Task 40 supplementary material available at on sustainable bioenergy trade. Biofuels, Bioprod. Bioref. 13, https://doi.org/10.1038/s41467-022-33976-5. 247–266 (2019). 72. Ecoinvent. Ecoinvent Centre for Life Cycle inventories. Ecoinvent Correspondence and requests for materials should be addressed to LCI database version 3.2. https://www.ecoinvent.org/. (2021). Aljoša Slameršak. 73. Hanssen, S. V. et al. The climate change mitigation potential of bioenergy with carbon capture and storage. Nat. Clim. Chang. 10, Peer review information Nature Communications thanks the other 1023–1029 (2020). anonymous reviewer(s) for their contribution to the peer review of this 74. Vaughan, N. E. et al. Evaluating the use of biomass energy with work. Peer review reports are available. carbon capture and storage in low emission scenarios. Environ. Res. Lett. 13, 044014 (2018). Reprints and permission information is available at 75. Huppmann, D. et al. IAMC 1.5 °C Scenario Explorer and Data hosted http://www.nature.com/reprints by IIASA. Integrated Assessment Modeling Consortium & Interna- tional Institute for Applied Systems Analysis. (2019). https://doi. Publisher’s note Springer Nature remains neutral with regard to jur- org/10.5281/zenodo.3363345. isdictional claims in published maps and institutional affiliations. 76. Riahi, K. et al. The Shared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions implications: an Open Access This article is licensed under a Creative Commons overview. Glob. Environ. Chang. 42, 153–168 (2017). Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as Acknowledgements long as you give appropriate credit to the original author(s) and the We are grateful to Paul Brockway, Piers Forster, Eric Galbraith, Julia source, provide a link to the Creative Commons license, and indicate if Steinberger, and Jeroen van den Bergh for their helpful comments. A.S. changes were made. The images or other third party material in this acknowledges financial support from la Caixa Foundation (ID article are included in the article’s Creative Commons license, unless 100010434, LCF/BQ/IN17/11620039). A.S. and G.K. also acknowledge indicated otherwise in a credit line to the material. If material is not support by the María de Maeztu Unit of Excellence (CEX2019-000940-M) included in the article’s Creative Commons license and your intended grant from the Spanish Ministry of Science and Innovation. use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright Author contributions holder. To view a copy of this license, visit http://creativecommons.org/ A.S., G.K., and D.W.O. all contributed to designing the project and licenses/by/4.0/. writing up the results. A.S. led the project and the writing of the article, developed the model, collected and analysed the data, and produced © The Author(s) 2022 Nature Communications | (2022)13:6932 15