RESEARCH PREVIEWevidence: modelled · physics simulation, validated at border grain
Half the power takes the long way round
Electricity follows impedance, not contracts. When Germany sells France 100 MW, only about half crosses their shared border; the rest loops through Switzerland, the Benelux and Austria, occupying other countries' wires. This page runs a DC power-flow model (2,741 substations, real line reactances) for any zone-to-zone exchange in Central-Western Europe and shows which borders physically carry it. Every number's model was tested against a year of observed cross-border flows it never trained on.
+0.88
median correlation vs observed flows (out-of-sample, 2025)
+0.99
median correlation vs JAO flow-based market PTDFs (11 borders)
2,741
substations in the model (open OSM network)
9
zones · 72 exchange pairs · loop flows included
100 MW sold from→
bars = MW physically carried per border per 100 MW exchanged · teal = the direct border · amber = loop flows through third countries · r = that border's own out-of-sample validation correlation · borders under 1 MW are dropped, never zero-filled · Luxembourg borders are unvalidated intra-market-zone meterings
The corridors operators pay to relieve
Europe's two most constraint-stressed grids share one geometry: wind in the north, load in the south. The model's north→south corridor transfer (the summed flow on every network branch crossing the cut) is recomputed for each hour of 2025, in Germany (latitude 51.2°N) and in Great Britain (the B6 Scotland–England boundary). It is the one congestion signal here validated against money, in both countries independently: the hours when these modelled corridors run hot are the hours the operators actually paid, German TSOs to redispatch power stations, NESO to curtail Scottish wind and constrain southern gas on.
GERMANY · north→south across 51.2°N
GREAT BRITAIN · Scotland→England across B6 (55.3°N)
bars = 90th-percentile modelled north→south corridor transfer per month, labels in GW (amber = Germany, teal = Great Britain) · this is the post-intervention state of each grid: in heavy hours the operators have already relieved part of the pressure shown · a transfer level, never an overload claim: no line, substation or N-1 statements are served at any grain
Corridor pressure & receipt. Base-case DC flows, one solve per hour of 2025 per corridor, summed signed across the branches straddling each cut; ratings-free by construction (a transfer level, not a violation statement). Validated against what the operators actually paid, independently in both grids: Germany Spearman +0.57 vs observed current-driven redispatch MW with the redispatch reversed back into the injections (+0.54 on actuals, 93% of top-decile hours saw above-median redispatch). Great Britain Spearman +0.82 vs observed SO-flagged Balancing Mechanism constraint energy: the settlement system's own accepted volumes joined to the system operator's own constraint flag (+0.57 on actuals, 100% of top-decile hours). GB injections are Elexon fuel-half-hourly actuals with the ten interconnectors at their landing buses; ENTSO-E's GB feed carries ~1% of real volumes post-Brexit and is not used. Branch-grain congestion indices (substrate ratings, JAO-rated base case, N-1) validated weak or negative on this substrate and are deliberately not served: corridor grain only. No operator records are served; the statistics are facts about this model. Source: GET /v2/insights/congestion-corridor.
Method & receipt. Zonal PTDF from a DC power flow on the PyPSA-Eur prebuilt OSM network (Zenodo, ODbL: real per-line reactances; HVDC excluded as controlled devices): 100 MW injected in the selling zone on generator-capacity weights, withdrawn in the buying zone on night-lights load weights; each border's sensitivity scaled by a calibration chosen out-of-sample against observed 2025 ENTSO-E physical flows. Phase-shifter-controlled borders carry an empirical response term: a fit, not a device model, labelled as such. This is a screening model: DC linearisation, 2025-vintage network, no intra-zonal or line-level claims, never an injection study or capacity allocation. First metric on the platform in the modelled evidence class, the register's red badge. Second, independent validation: the model's raw sensitivities agree with the official JAO Core flow-based market-coupling PTDFs, the sensitivities the day-ahead market actually clears on, at median per-border r +0.99 across 11 shared borders and 28 zone-pair exchanges (no JAO data is served; the statistic is a fact about this model). Source: GET /v2/insights/corridor-sensitivity · methodology · sources.