Converted PyPSA-Eur models — pypsa_eur

Unsolved energyRt models converted from PyPSA-Eur networks, spanning the range this package is built for: the European system at several spatial resolutions, a small Belgian model that runs in seconds with its copperplate reference, a one-year horizon series, and a vintaged 2025-2050 pathway model. All are built on 2025 weather and measured ENTSO-E 2025 load, and are loaded lazily, so they cost nothing until touched.

Usage

pypsa_eur_nuts3

pypsa_eur_41

pypsa_eur_5

pypsa_eur_5cp

pypsa_eur_41v

pypsa_eur_41_2025

pypsa_eur_41_2030

pypsa_eur_41_2035

pypsa_eur_41_2040

pypsa_eur_41_2045

pypsa_eur_41_2050

Format

An energyRt model object.

An object of class model of length 1.

Source

Converted from PyPSA-Eur v2026.02.0 (commit d6383eb, PyPSA 1.1.0) by convert_pypsa().

Licence: ODbL-1.0, required by the share-alike terms of the OpenStreetMap-derived inputs, and independent of this package's Apache-2.0 code licence. Attribution is required for OpenStreetMap contributors and the European Environment Agency among others.

The World Database on Protected Areas, which may not be redistributed, is not present – see system.file("LICENSE.note", package = "reneuro") for how that was verified and what it costs (renewable land availability in Moldova and Ukraine is overstated as a result). The same file records that the OPSD and ENTSO-E demand series carry unclear upstream terms.

The build scripts live in two places. This package's data-raw/ holds pypsa_eur_41.R, pypsa_eur_nuts3.R, pypsa_eur_41_years.R and pypsa_eur_41v.R; pypsa_eur_5 and pypsa_eur_5cp are built by the converter workspace's reneuro.dev/data-raw/make_models.R, from the network also bundled as reneuro_example("BE_base_s_5_elec.nc"). All shipped models were rebuilt 2026-09-03 with the corrected calendar chronology: the converter used to emit the timetable hour-major, which chained the storage balance across days at the same wall-clock hour – storage could not carry noon into evening. The Belgian pair now also carries the current defaults (flat corridors, sanitised calendar names), and be_solved's objective gate moved accordingly (verified +0.386% above PyPSA's own solve of the same network, the documented transport-relaxation class of difference).

Details

pypsa_eur_5 is PyPSA-Eur's own Belgium example: five regions, one week (168 hourly snapshots). It is small enough to solve in seconds, which makes it the model every example and test uses.

pypsa_eur_5cp is the same network with transmission = "copperplate" – the same corridors, but unbounded and lossless. It is the reference point for what the network representation costs: a copperplate cannot be more expensive than any bounded version of itself, so it is the floor.

pypsa_eur_41 is PyPSA-Eur's own k-means clustering at 41 nodes over the full year, so it is directly comparable to a PyPSA-Eur run at that size. Its corridors carry a single flat loss rate: the six-tranche version it replaced produced 224,640 generated constraints on a 288-slice calendar, which no open solver generates in reasonable time. It is also the first model built with the fuel price separated – varom holds the true VOM and each fuel's supply carries its price – so a gas-price sensitivity is a change to one number rather than a re-conversion.

pypsa_eur_nuts3 is the NUTS3 network over the full year (8,760 hourly snapshots) – the finest resolution the data supports, intended as the starting point for aggregation and study-area work in R rather than as a model to solve whole. NUTS3 yields 1,035 AC nodes rather than 1,477 because 442 regions contain no substation and merge into neighbours, retaining their demand and generation in the region they join. Trade uses one flat loss rate per corridor and a transport formulation without Kirchhoff's voltage law. Its hourly weather profiles are seven eighths of its size, so they ship as separate wx_nuts3_* objects and are put back with attach_weather() – a model without them cannot be interpolated, because its technologies reference them by name. It carries nuts_gs, so energyRt::aggregate_model_regions(pypsa_eur_nuts3, level = "nuts2") needs no second argument; the coarser NUTS levels (289 regions at NUTS2, 106 at NUTS1, 36 at NUTS0) are derived that way rather than shipped.

pypsa_eur_41_2025 ... pypsa_eur_41_2050 are the one-year horizon series: the 41-node system rebuilt per planning horizon in five-year steps. Weather, load and network are identical to pypsa_eur_41; what changes with the horizon is the cost vintage (technology-data v0.14.0 for that year) and the existing fleet, aged by two rules kept separable in provenance – announced DateOut schedules (114 GW, mostly coal and lignite phase-outs) and assumed carrier-lifetime retirement, which reproduces what PyPSA's myopic mode does between horizons (its overnight mode does no ageing at all). attr(m, "reneuro_provenance")$retirement holds the per-carrier split, so the assumed half can be identified and discounted. The series ships without weather: all six share pypsa_eur_41's profiles, so attach them before interpolating – attach_weather(pypsa_eur_41_2030, from = pypsa_eur_41).

pypsa_eur_41v is the vintaged multi-year model: one model over the 2025–2050 horizon (five-year milestones) where each carrier is a single technology with a STOCK vintage carrying the existing fleet's milestone-by-milestone surviving-stock series (announced retirements plus assumed lifetime aging, summed over PyPSA's build-year bins – the bins themselves are parameter-flat under the 2025 cost-table clamp, so they are aggregated away) and, for investable carriers, open paths: where the assumption tables move a real parameter across the horizon (gas efficiency), one NEW<year> vintage window per assumption year with that table's efficiency, lifetime and annuity frozen – year-keyed efficiency inside a single open path would let standing capacity retrofit itself for free – and where only cost varies (VRE), a single NEW path with year-keyed investment costs, so build-year economics stay endogenous. Carriers with no existing fleet (solar-hsat, offwind DC and floating) stay un-vintaged and investable. Demand is flat at 2025; like the horizon series it ships without weather – attach_weather(pypsa_eur_41v, from = pypsa_eur_41). energyRt::getVariants() and energyRt::variantSummary() list the vintages; attr(m, "reneuro_provenance") records the retirement-rule split and the aggregation.

The continental models are built on 2025 weather and load (ENTSO-E measured demand; the europe-2025-sarah3-era5 cutout). All were converted with cost_source = "network", taking costs from the network rather than from a separate cost assumption, so they agree with the PyPSA solve they can be compared against.

Examples

# Provenance travels with the object.
str(attr(pypsa_eur_5, "reneuro_provenance"), max.level = 1)
#> List of 9
#>  $ clone       : chr "pypsa-eur-v2026"
#>  $ commit      : chr "d6383ebf602767b1adbb676fe8a16e37a6e9f932"
#>  $ describe    : chr "v2026.02.0"
#>  $ dirty       : logi TRUE
#>  $ built_on    : chr "2026-09-03"
#>  $ reneuro     : chr "0.3.0.9000"
#>  $ object      : chr "pypsa_eur_5"
#>  $ source      : chr "base_s_5_elec.nc"
#>  $ convert_args:List of 1