Overview of European energy system model PyPSA-Eur
For our analysis, we use the European sector-coupled high-resolution energy system model PyPSA-Eur82 (derivative of v0.13.0) based on the open-source modelling framework PyPSA83 (Python for Power System Analysis), covering the energy demands of all sectors including electricity, heat, transport, industry, agriculture, as well as non-energy feedstock demands, international shipping, and aviation. An overview of considered supply, consumption, and balancing technologies per carrier is shown in Supplementary Fig. 2.
The model simultaneously optimises spatially explicit investments and the operation of generation, storage, conversion, and transmission assets to minimise total system costs in a single linear optimisation problem, which assumes perfect operational foresight and is solved with Gurobi (v11.0.1)84. To manage computational complexity, no pathways with multiple investment periods are calculated, but overnight scenarios targeting net-zero CO2 emissions. The capacity expansion is based on technology cost and efficiency assumptions for 2040 (see ‘Data availability’), acknowledging that much of the required infrastructure must be constructed well before reaching net-zero emissions. Figures 2 and 5 and Supplementary Fig. 29 feature additional scenarios using technology assumptions for 2030 and 2050.
Existing hydro-electric power plants85 are included, as well as nuclear power plants built after 1990 or currently under construction according to Global Energy Monitor’s Global Nuclear Plant Tracker (52 GW total of 106 GW in current operation)86. While hydroelectricity is assumed to be non-extendable due to geographic constraints, additional nuclear capacities can be expanded where cost-effective. We assume the existing nuclear fleet is operated inflexibly and apply country-specific historical availability factors from 2021 to 202387.
Temporally, the model is solved with an uninterrupted 4 h equivalent resolution for a single year (2190 time steps), using a segmentation clustering approach implemented in the tsam toolbox on all time-varying data88. While weather variations between years are not considered for computational reasons, the chosen weather year 2013 is representative in terms of wind and solar availability and heat demand89. Some demands are associated with a time-varying profiles (e.g., residential/services electricity, electric vehicles, and heating demand) based on travel patterns or ambient weather conditions, while the other exogenous demands are assumed to be time-constant (e.g., kerosene, naphtha, methanol, ammonia, and industry electricity).
Spatially, the model resolves 115 European regions90, covering the European Union, the United Kingdom, Norway, Switzerland, and the Balkan countries without Malta and Cyprus. For computational reasons, only electricity, heat, and hydrogen are modelled at high spatial resolution, while oil, methanol, methane, ammonia, and carbon dioxide are treated as easily transportable without spatial constraints. Of the total final energy and non-energy demand (Supplementary Fig. 5), only some demands are spatially fixed (Supplementary Fig. 4). These include electricity for residential, industry, services, and agriculture; heat; electric vehicles; solid biomass for industry; naphtha/methanol feedstocks; and hydrogen for crude steel and ammonia production unless these industries can relocate.
Most other hydrogen demands are spatially variable. Only a small demand of 5 TWh a−1 in the chemicals industry (excluding liquid feedstocks) remains, which is offset by spatially fixed hydrogen production of around 10 TWh a−1 from chlor-alkali electrolysis for chlorine production. High-temperature industrial heat is supplied by methane, shipping and aviation use carbonaceous fuels, and land transport is fully electrified. In district heating and the power sector, backup hydrogen capacities are endogenously sized and sited just as the production capacities of hydrogen derivatives (Fischer-Tropsch, methane, methanol), which account for more than 80% of the hydrogen consumption. Since the model optimises the siting and operation of these fuel synthesis plants and electrolysers, many demands are spatially variable (e.g., electricity demand for electrolysers or hydrogen demand for methanolisation). Existing hydrogen production capacities from fossil gas reforming are not considered, as they are expected to reach the end of life over the model horizon.
A mathematical description of PyPSA-Eur can be found in Supplementary Note 1, adapted from Neumann et al.14
Gas and electricity network modelling
Networks are considered for electricity, methane, and hydrogen transport. Existing gas pipelines taken from SciGRID_gas91, can be repurposed to hydrogen in addition to new hydrogen pipelines14. Data on the gas transmission network is further supplemented by the locations of fossil gas extraction sites and gas storage facilities based on SciGRID_gas91, as well as investment costs and capacities of LNG terminals in operation or under construction from Global Energy Monitor’s Europe Gas Tracker92. Geological potentials for hydrogen storage are taken from Caglayan et al.93, restricting where this low-cost storage option is available. In modelling gas and hydrogen flows, we incorporate electricity demands for compression of 1% and 2% per 1000km of the transported energy, respectively94. Existing high-voltage grid data is taken from OpenStreetMap95. For HVDC transmission lines, we assume 2% static losses at the substations and additional losses of 3% per 1000 km. The losses of high-voltage AC transmission lines are estimated using the piecewise linear approximation from Neumann et al.96, in addition to applying linearised power flow equations97. Up to a maximum capacity increase of 30%, we consider dynamic line rating (DLR), leveraging the cooling effect of wind and low ambient temperatures to exploit existing transmission assets fully98. To approximate N − 1 resilience, transmission lines may only be used up to 70% of their rated dynamic capacity99. To prevent excessive expansion of single connections, power transmission reinforcements between two regions are limited to 15 GW, while an upper limit of 50.7 GW is placed on hydrogen pipelines, which corresponds to three 48-inch pipelines94.
Wind and solar potentials
Renewable potentials and time series for wind and solar electricity generation are calculated with atlite100, considering land eligibility constraints like nature reserves, excluded land use types, topography, bathymetry, and distance criteria to settlements. Given low onshore wind expansion in many European countries in recent years101, a deployment density of 1.5 MW km−2 is assumed for eligible land for onshore wind expansion102. For reference, this assumption leads to an onshore wind potential for Germany of 244 GW. The temporal renewable generation potential for the available area is then assessed based on reanalysis weather data, ERA5103, and satellite observations for solar irradiation, SARAH-3104, in combination with standard solar panel and wind turbine models provided by atlite.
Biomass potentials
Biomass potentials are restricted to residues from agriculture and forestry, as well as waste and manure, based on the regional medium potentials specified for 2050 in the JRC-ENSPRESO database105. Continued use of energy crops or biomass imports are not considered. The finite sustainable biomass resource can be employed for low-temperature heat provision in industrial applications, biomass boilers, and CHPs, and (electro-)biofuel production for use in aviation, shipping, and the chemicals industry. In addition, we allow biogas upgrading, including capturing the CO2 contained in biogas, which unlocks all considered uses of regular methane (Supplementary Fig. 2). The total assumed bioenergy potentials are 1372 TWh, which splits into 358 TWh/a for biogas and 1014 TWh/a for solid biomass. The total carbon content corresponds to 605 \({{{{\rm{Mt}}}}}_{{{\mbox{CO}}}_{2}}\) a−1, which is not fully available as a feedstock for fuel synthesis or sequestration for negative emissions due to imperfect capture rates of up to 90%. Biogenic CO2 can be captured from biogas upgrading, biomass CHPs and biomass-based low-temperature heat provision in industrial use, if the added cost of carbon capture is economically viable.
Carbon management
The carbon management features of the model trace the carbon cycles through various conversion stages: industrial emissions, biomass and gas combustion, carbon capture in numerous applications, direct air capture, intermediate storage, electrofuels, recycling, landfill or long-term sequestration. The overall annual sequestration of CO2 is limited to 200 \({{{{\rm{Mt}}}}}_{{{\mbox{CO}}}_{2}}\) a−1, similar to the 250 \({{{{\rm{Mt}}}}}_{{{\mbox{CO}}}_{2}}\) a−1 highlighted in the European Commission’s carbon management strategy70. This number allows for sequestering the industry’s unabated fossil emissions (e.g., in the cement industry) while minimising reliance on carbon removal technologies. A carbon dioxide network topology is not co-optimised since CO2 is not spatially resolved. This means that the location of biogenic or industrial point sources of CO2 is not a siting factor that this model version considers for PtX processes, implicitly assuming that the CO2 would be transported there at low cost75,106.
Transport sector fuel assumptions
While the shipping sector is assumed to use methanol as fuel, given its high technology-readiness level compared to hydrogen or ammonia107, land-based transport, including heavy-duty vehicles, is fully electrified in the presented scenarios108. Aviation can use green kerosene derived from Fischer-Tropsch fuels or methanol, owing to the lower technology readiness levels of fuel cell or battery-electric aircraft107. Alternative uses for methanol and Fischer-Tropsch fuels extend beyond transport, including power-to-methanol73, diesel for agriculture machinery and as feedstock for high-value chemicals.
Technical constraints of synthetic fuel production
We consider potential flexibility restrictions in the synthesis processes to obtain more realistic operational patterns of green electrofuel synthesis plants. We apply a minimum part load of 20% for methanolisation and 50% for methanation and Fischer-Tropsch synthesis109,110,111,112. The assumed lower operational flexibility is a potential disadvantage of Fischer-Tropsch over methanol synthesis, where theses fuels compete. These ‘green’ options then compete with ‘blue’ and ‘grey’ options, such as steam methane reforming of fossil gas with or without carbon capture for hydrogen (Supplementary Fig. 2). Some carriers also feature a biogenic production route (e.g., methane and oil).
Heating sector modelling and PtX waste heat
Heating supply technologies like heat pumps, electric boilers, gas boilers, and combined heat and power (CHP) plants are endogenously optimised separately for decentral use and central district heating. District heating shares of demand are exogenously set to a maximum of 60% of the total urban heat demand with sufficiently high population density. Besides the options for long-duration thermal energy storage, district heating networks can further be supplemented with waste heat from various power-to-X processes: electrolysis, methanation, ammonia synthesis, and Fischer-Tropsch fuel synthesis. Because the thermal discharge from the methanol synthesis is primarily used to distillate the methanol-water output mix73, its waste heat potential is not considered for district heat. Here, we assume a utilisable share of waste heat of 25%, considering that within the 115 regions, only a fraction of fuel synthesis plants might be connected to district heating systems. In further sensitivity analyses, we explore the effect of no or full waste heat utilisation.
Backup heat and power options
The model includes a variety of options for providing backup power and heating in periods of low renewable generation and high demand (Supplementary Fig. 2). Backup power options include hydrogen, gas and methanol turbines. Backup heat options include gas boilers and resistive heaters. For combined backup heat and power, we consider biomass, hydrogen, and gas CHPs. Furthermore, flexible demands like electric vehicles, heat pumps and fuel synthesis units, as well as batteries and thermal storage in district heating, can be utilised to reduce the need for backup capacities.
Industry relocation modelling for crude steel and ammonia production
Unless indicated otherwise, all scenarios also allow the model to relocate the crude steel and ammonia industry within Europe endogenously. This allows the best sites within Europe to compete with outsourced production abroad. While this captures some of the most energy-intensive industry sectors, other sectors, like concrete and alumina production, are not considered for relocation.
Without relocation of crude steel and ammonia production allowed, the production volumes of primary crude steel, by direct iron reduction (DRI) and electric arc furnace (EAF), and ammonia for fertilisers, by Haber-Bosch synthesis, are spatially fixed. This results in exogenous hydrogen demand per region. Total production volumes are based on current levels113,114. For the spatial distribution, we use data on the existing integrated steelworks listed in Global Energy Monitor’s Global Steel Plant Tracker115 and manually collected data on the location and size of ammonia plants in Europe.
With the relocation of crude steel and ammonia production allowed, the model endogenously chooses the regional production volumes of primary crude steel, HBI, and ammonia, subject to the availability of cheap hydrogen. Thereby, the regional capacities and operation of Haber-Bosch, DRI, and EAF plants are co-optimised with the rest of the system, similar to the siting of Fischer-Tropsch or methanolisation plants. For DRI and EAF, investment costs and specific requirements for fuels and iron ore are taken from the Steel Sector Transition Strategy Model (ST-STSM) of the Mission Possible Partnership116,117. and assume steel can be stored and transported without constraints within Europe.
For both cases, we assume a rise in the steel recycling rate from 40% today to 70% in our carbon-neutral scenarios118. We assume that the electric arc furnaces for secondary steel remain, in proportion, at current locations and do not relocate.
A limitation of the relocation modelling of crude steel and ammonia production is that it only considers the cost of energy in the siting of these industries. Other factors, such as impacts on regional economies and local jobs, integration with other production processes, or availability of other existing infrastructure, are not considered, largely due to a lack of data. The resulting relocation patterns should therefore be interpreted with caution, as they might underestimate total relocation costs and frictions. We allow domestic relocation, nevertheless, in most scenarios, as it would be inconsistent to allow crude steel and ammonia imports from abroad while preventing relocation within Europe.
Import supply chain modelling with TRACE
The European energy system model is extended with data from the TRACE model (derivative of v1.1) used in Hampp et al.54 to assess the unit costs of different vectors for importing green energy and material to entry points in Europe from various world regions. For consistency with the European model, the techno-economic assumptions were aligned, using the same values for 2040 (plus 2030 / 2050 in Fig. 2 and 5 and Supplementary Fig. 29 and a uniform weighted average cost of capital (WACC) of 7%119. As possible import vectors, we consider electricity by transmission lines, hydrogen as a gas by pipelines and as a liquid by ship, methane as a liquid by ship, liquid ammonia, crude steel and HBI, methanol and Fischer-Tropsch fuels by ship. Liquid organic hydrogen carriers (LOHC) are not considered as export vectors due to their lower technology readiness level (TRL) compared to other vectors1.
Our selection of 53 potential exporting regions broadly comprises countries with favourable wind and solar resources and large enough potential for substantial exports above 500 TWh a−1 in addition to domestic consumption. We exclude some countries due to political instability (e.g., Sudan, Somalia, Yemen), using a Fragile States Index120 value of 100 as a threshold, or due to severe imposed sanctions (e.g., Russia, Iran, Iraq), following the EU Sanctions Map121. Landlocked countries without access to seaports or realistic pipeline connections are excluded. For landlocked regions within pipeline reach, we only exclude shipborne vectors. Some large countries are split into multiple subregions for a more differentiated view (e.g., USA, Argentina, Brazil, and China). The resulting regions are marked in Fig. 1A.
To determine the levelised cost of energy for exports, the methodology first assesses the regional potentials for solar, onshore, and offshore wind energy. These potentials and time series are calculated using atlite100, applying similar land eligibility constraints as in PyPSA-Eur (but using other datasets with global coverage) and applying the same wind turbine and solar panel models to ERA5103 weather data for 2013 in eligible regions. Since TRACE evaluates whole regions without further transmission network resolution, the renewable potentials and profiles within a region are split into different resource classes to reduce smoothing effects. We consider 30 classes each for onshore wind and solar, and 10 for offshore wind, where applicable. Based on these calculations, levelised cost of electricity (LCOE) curves can be determined for each region. A selection of LCOE curves is shown in Supplementary Fig. 22.
In the next step, potentials are reduced by the projected future local energy demand, starting with the lowest LCOE resource classes. With this approach, domestic consumption is prioritised and supplied by the regions’ best renewable resources, even though we do not model the energy transition in exporting regions in detail. To create the demand projections, we use the GEGIS122 tool, which utilises machine learning on historical time series, weather data, and macro-economic factors to create artificial electricity demand time series based on population and gross domestic product (GDP) growth scenarios following the SSP2 scenario of the Shared Socioeconomic Pathways123. From these time series, we take the annual total and increase it by a factor of two to account for further electrification of other sectors, which the GEGIS tool does not consider.
The remaining wind and solar electricity supply can then be used to produce the specific energy or material vector according to the flow chart of conversion pathways shown in Supplementary Fig. 1. Considered technologies include water electrolysis for H2, direct air capture (DAC) for CO2, synthesis of methane, methanol, ammonia or Fischer-Tropsch fuels from H2 with CO2 or N2, and H2 direct iron reduction (DRI) for sponge iron with subsequent processing to green steel in electric arc furnaces (EAF) from iron ore priced at 97.7 € t−1116. Other CO2 sources than DAC are not considered in the exporting regions. Furthermore, while batteries and hydrogen storage in steel tanks are considered, underground hydrogen storage is excluded due to the uncertain availability of salt caverns in many potential exporting regions124,125. We also assume that the energy supply chains dedicated to exports will be islanded from the rest of the local energy system, i.e., that curtailed electricity or waste heat could not be used locally.
For each vector, an annual reference export demand of 500 TWhLHV or 100 Mt of crude steel and HBI is assumed, mirroring large-scale energy and material infrastructures and export volumes, corresponding to approximately 40% of current European LNG imports126 and 66% of European steel production127. Transport distances are calculated between the exporting regions and the twelve representative European import locations using the searoute Python tool128 for shipborne vectors or crow-fly distances for pipeline or HVDC connections, and modified by a mode-specific detour factor. The chosen representative import locations are based on large ports and LNG terminals in the United Kingdom, the Netherlands, Poland, Greece, Italy, Spain, and Portugal, as well as pipeline entry points in Slovakia, Greece, Italy, and Spain. All energy supply chains are assumed to consume their energy vector as fuel for transport to Europe, except for HBI and crude steel, which use externally bought green methanol as shipping fuel. The capital costs of the ships and pipelines are also included, following the metholodogy of Hampp et al.54.
For each combination of carrier, exporter, and importer, a linear capacity expansion optimisation is performed to determine cost-optimal investments and the operation of generation, conversion, storage, and transport capacities for all intermediary products to deliver 500 TWh a−1 (or 100 Mt a−1 for materials) of the final carrier to Europe. Dividing the total annual system costs by the targeted annual export volume yields the levelised cost of energy or material as seen by the European entry point. To match the multi-hourly resolution used for the European model, the TRACE model was configured to use a 3-hourly resolution for 2013, resulting in similar balancing requirements. Considering the reference export volume of 500 TWh a−1 (or 100 Mt a−1 for materials), the resulting levelised cost curves of imports for different import vectors and exporting regions are presented for the respective lowest-cost entry point to Europe in Supplementary Figs. 7–10. The curves show the varying cost composition of the country-carrier pairs. In this step, each import vector combination of carrier, exporter, and importer is optimised separately. Further constraints, like constraints on total export volumes per country, are imposed in the coupling to the European model.
A mathematical description of TRACE can be found in Section S3 in Hampp et al.54
Coupling of import options to European model
The resulting levelised unit cost for each combination of carrier, exporter, and reference importer is then used as an exogenous input to the European model. For each candidate entry point in the 115 European model regions, we match the closest reference import location from TRACE and add the corresponding import cost curve as a supply option (Supplementary Figs. 7–10). Moreover, we limit energy exports from any one exporting region to Europe for the sum of all carriers to 500 TWh a−1. This is to both prevent a single country from dominating the import mix and be consistent with the target export volume assumed in TRACE. Beyond that, the decision about the origin, destination, vector, volume, and timing of imports is largely endogenous to PyPSA-Eur.
However, imports may be further restricted by the expansion of domestic import infrastructure. For each vector, we identify locations where the respective carrier may enter the European energy system by considering where LNG terminals and cross-continental pipelines are located (Fig. 1b). For hydrogen imports by pipeline, imports must be near-constant, varying between 90–100% of peak imports, aligning with the high pipeline utilisation rates observed in the TRACE model. For methane imports by ship, existing LNG terminals reported in Global Energy Monitor’s Europe Gas Tracker92 can be used. For hydrogen by ship, new terminals can be built in regions where LNG terminals exist. To ensure regional diversity in potential gas and hydrogen imports and avoid vulnerable singular import locations, we allow the expansion beyond the reported capacities only up to a factor of 2.5, taking the median value of reported investment costs for LNG terminals129. A premium of 20% is added for hydrogen import terminals due to the lack of practical experience with them. For electricity, the capacity and operational patterns of the HVDC links can be endogenously optimised. Imports for carbonaceous fuels, ammonia, HBI, and steel are not spatially allocated to specific ports, given their low transport costs relative to value. Port capacities are assumed unconstrained since these commodities, particularly carbonaceous fuels, are comparable to the large fossil oil volumes currently handled at European ports.
Further conversion of imported fuels is also possible once they have arrived in Europe, e.g., hydrogen could be used to synthesise carbon-based fuels, ammonia could be cracked to hydrogen, methane could be reformed to hydrogen, and methane or methanol could be combusted for power generation. However, conversion losses can make it less attractive economically to use a high-value hydrogen derivative merely as a transport and storage vessel, only to reconvert it back to hydrogen or electricity.
The supply chain of electricity imports is endogenously optimised with the rest of the European system rather than using a constant levelised cost of electricity for each export region. This is because, owing to the greater challenge of storing electricity, the hourly variability of wind and solar electricity leads to higher price variability than hydrogen and its derivatives, and the intake needs to be more closely coordinated with the European power grid. The endogenous optimisation comprises wind and solar capacities, batteries and hydrogen storage in steel tanks, and the size and operation of HVDC link connections to Europe based on the renewable capacity factor time series as illustrated in Fig. 1b. Europe’s connection options with exporting regions are confined to the 4% nearest regions, with additional ultra-long distance connection options to Ireland, Cornwall, and Brittany following the vision of the Xlinks project between Morocco and the United Kingdom30. Connections through Russia or Belarus are excluded. In addition to excluded entry points, some connections from Central Asia are affected by additional detours beyond the regularly applied detour factor of 125% of the as-the-crow-flies distance. Similar to intra-European HVDC transmission, a 3% loss per 1000km and a 2% converter station loss are assumed.
Finally, we note that all mass-energy conversion is based on the lower heating value (LHV). To present energy and material imports in a common unit, the embodied energy in steel is approximated with the 2.1 kWh kg−1 released in iron oxide reduction, i.e., energy released by combustion130. All currency values are given in €2020.
