A wind-modeling choice can move Europe’s power plan by 480 GW
A modeling choice about how wind and sunlight become electricity can shift the capacity Europe needs to build by as much as 480 gigawatts. That is roughly the same scale as the difference between the most challenging weather years in the study. The striking part is what does not move much: the assumptions barely change how often renewable-energy droughts happen, or when the electricity system comes under stress. They change how severe the shortfall looks—and, in turn, how much infrastructure a planner decides is needed (Karkossa et al., 2026).
This distinction matters because a power system is not designed for an average Tuesday. It must also survive the rare stretches when wind and solar output sag together, sometimes across large parts of a continent. If a model slightly overstates what turbines or panels can deliver during those hours, the system can look more secure on paper than it would be in practice. If it understates output, it can point planners toward a more expensive build-out than necessary.
The study’s central message is not that one modeling recipe is always right. It is that decisions about the weather-to-power conversion deserve the same scrutiny as the weather record itself—especially at the low end of the generation distribution, where adequacy is tested.
The Science
Karkossa and colleagues at the Technical University of Denmark asked a practical planning question: when analysts turn weather data into hourly wind and solar generation, which assumptions actually change the result? They used 85 years of ERA5 weather reanalysis, from 1941 through 2025, covering 33 European countries. A reanalysis combines observations with a weather model to reconstruct a consistent picture of past atmospheric conditions. It gives planners many more weather years than a short record of measured power plants could provide.
The team used that long record to generate hourly profiles for solar photovoltaic power, onshore wind, and offshore wind. Then, one choice at a time, they changed how the conversion was done. The variations included correcting weather data against other sources, estimating wind at turbine height, distributing capacity across different sites, choosing different turbine and solar technologies, and including the energy lost when one turbine disrupts the wind reaching another. The baseline and variants were processed with the atlite modeling toolbox.
A capacity factor is the electricity a technology actually produces over a period divided by what it would have produced running at full power throughout. It is useful for comparing resource and technology performance, but a high annual average does not guarantee reliable output during a difficult week. The researchers therefore examined three layers: average capacity factors, low-generation “droughts,” and the capacity choices produced by an electricity-system optimization model.
For droughts, they set a low-output threshold using the fifth percentile for each day of the year: in plain terms, the level that generation falls below on roughly five percent of comparable days. They tracked event frequency, duration, and how much of a seasonally typical amount of electricity was delivered during those periods. They then carried selected renewable profiles into PyPSA-Eur, a model that chooses generation, storage, and dispatch for a net-zero European electricity system. The system analysis covered five previously identified critical weather years, while the full renewable-resource analysis used all 85 years.
The wind correction map hints at why these choices are not merely technical housekeeping. ERA5’s wind estimates do not differ from reference estimates by the same amount everywhere. In the study’s illustration, correction factors vary across Europe, with particularly large adjustments in some locations. A correction that looks modest at continental scale can matter greatly in the places where a country has concentrated its best sites.
What They Found
First, the modeling choice with the biggest effect on average solar capacity factors was the panel configuration. Horizontal single-axis tracking, which turns panels to follow the Sun through the day, raised annual solar capacity factors by 19 to 26 percent across the four focus countries. It also shifted production toward morning and evening. That temporal change matters to the system: a megawatt-hour delivered at a different hour can be more or less useful even if annual totals look similar.
Spatial variation mattered for wind. Restricting the analysis to a country’s highest-resource sites raised onshore wind capacity factors by 11 to 26 percentage points compared with the country-wide baseline. That is a reminder that a national average can hide a lot: some countries have a small area of exceptional wind, while much larger areas have weaker conditions. The map of resource grades shows the contrast between top-, middle-, and lower-quality areas across Europe.
Other wind assumptions also moved annual averages. Raising turbine hub height to 150 metres increased onshore wind capacity factors by 3 to 11 percentage points. A larger, lower-specific-power turbine raised them by 2.1 to 5.3 points; including wake losses lowered them by 0.8 to 2.2 points. The exact effect varied by country and technology. The method used to extrapolate wind speeds to hub height was less consequential than the height change itself.
Bias correction was especially important in complex terrain. Correcting ERA5 wind speeds against the Global Wind Atlas increased modeled onshore wind capacity factors by about 8 percentage points in Spain and Italy. That brought values closer to historical estimates in those countries. In Denmark and Germany, however, the same correction pushed modeled capacity factors farther from reported historical figures. Historical comparisons are not a perfect test—installed fleets, curtailment, and reported capacity all complicate the match—but the contrast underlines that correction should be regional, not treated as a universal uplift.
The weather year itself remained a substantial source of variation. For solar, individual years deviated from the 85-year mean by as much as minus 12 percent or plus 13 percent. For onshore wind, annual capacity factors varied by up to 6.5 percentage points. In other words, a model’s assumptions and the particular weather years selected can both shape what “normal” looks like. In several cases, weather-year variation exceeded the effects of layout choices or bias correction.
But the sharpest result concerns drought behavior. The number and duration of low-wind events barely changed across the modeling variants: their frequency and duration differed from baseline by about 2.4 percent on average. Put differently, the broad calendar of stress was robust. The modeling choices did not usually make the system’s difficult periods appear out of nowhere or disappear.
The amount of electricity delivered during those periods did change. Bias correction and higher hub heights generally increased the share delivered during wind droughts, reducing the modeled shortfall. Wake losses consistently made the shortfall worse. Resource grades had no single direction across countries: the best-resource category improved drought delivery in Denmark and Italy, while low-resource Grade C produced the largest single reduction, lowering the delivered share by as much as 7 percentage points in Germany. Solar showed similarly mixed effects: bias correction raised drought delivery, while tracking and the resource-grade variants reduced it under the study’s drought measure.
The difference between event timing and event severity is crucial. A grid planner may already know that a cold, still winter week is dangerous. But if the model says wind turbines will generate a little more energy during that week, the estimated need for firm generation, storage, transmission, or other flexibility can fall. A model can therefore agree on when the system is vulnerable while disagreeing substantially about how much backup it needs.
In the system optimization, the modeled capacity shift reached as much as 480 gigawatts across renewable-generation assumptions. The spread in total capacity across the five critical weather years was 430 gigawatts. These are system-wide planning differences, not a forecast that Europe will necessarily build a particular amount. They show that the choice of weather-to-power representation can rival the influence of which difficult weather year a model is asked to withstand.
Changes in onshore wind capacity factor
Reported range of changes in annual onshore wind capacity factor across countries for selected modeling choices.
| Label | Value |
|---|---|
| Hub height raised to 150 m | 3 percentage points |
| Larger, lower-specific-power turbine | 2.1 percentage points |
| Wake losses included | -2.2 percentage points |
System capacity: modeling choices vs. weather years
The study reports that renewable modeling choices shift modeled capacity by up to 480 GW, while total capacity varies by 430 GW across the five critical weather years.
| Label | Value |
|---|---|
| Modeling-choice shift | 480 GW |
| Spread across five critical weather years | 430 GW |
Why This Changes Things
A common way to check a renewable model is to ask whether it reproduces a plausible annual mean. This paper shows why that check is not enough. In an energy system with a large share of weather-dependent generation, the low-output tail can matter more to reliability than the average. A technology model that gets annual production roughly right may still misrepresent the hours when supply is scarce.
This is not an argument to optimize every detail or to treat the largest modeled capacity as the safest answer. Different choices can move results in opposite directions. Some improve generation during droughts; others deepen the shortfall. The relevant question is not whether a method raises or lowers average output, but whether it represents the combination of location, season, and extreme conditions that determines system adequacy.
That finding also changes how planners should think about “more data.” Long weather records are essential because rare droughts may be absent from a short observational period. Yet 85 years of weather cannot rescue an inaccurate conversion from wind speed and sunlight into electricity. Conversely, an elegant turbine or PV model cannot compensate for feeding it a weather record that misses the conditions that matter. The weather series and the power-conversion model are parts of one chain.
For public policy, the issue is tangible. Decisions about storage, interconnectors, dispatchable low-carbon generation, and the siting of renewables all depend on estimates of the shortfall that must be covered. If system models understate renewable output during the worst periods, they risk overconfidence in adequacy. If they overstate the shortfall, they may favor unnecessarily costly combinations of infrastructure. This study does not tell policymakers which portfolio to choose; it shows that seemingly secondary modeling assumptions can alter the scale of the problem those portfolios are meant to solve.
The results are also a case for representing the geography of resource quality explicitly. A country’s highest-quality sites are not interchangeable with its average land area. Treating them as a uniform national resource can blur both the amount of generation available and its behavior during low-output events. The resource-grade approach is one way to expose this heterogeneity, although it is not a full siting or land-use analysis.
And technology assumptions should look forward, not only backward. A future fleet may use taller turbines, larger rotors, tracking solar, and different siting patterns than today’s installed capacity. Modeling today’s fleet can be useful for validating weather inputs, but it may not represent the system planners are actually trying to build. At the same time, forward-looking technology assumptions need to be connected to realistic costs and deployment constraints; higher modeled capacity factors do not make a technology costless or universally deployable.
The comparison is especially useful because the authors evaluated choices side by side and traced them from generation profiles into a system model. Many sensitivity studies change just one piece of the chain or stop at annual capacity factors. Here, the authors show how choices that seem small at the resource-modeling stage can propagate into a continent-scale capacity plan.
What’s Next
The authors’ recommendations follow directly from the results: use multiple weather years; correct and validate weather inputs regionally; represent resource heterogeneity, plausible future technologies, and wake losses; and evaluate choices by their effects on low-generation periods, not only on annual means. The same logic applies beyond Europe. Any region planning a power system around wind and solar needs to know whether its modeling pipeline captures the tail events that set the system’s reliability challenge.
There are important limits. The system optimization is an electricity-only, perfect-foresight exercise: it can see the modeled year ahead and chooses a least-cost configuration within the constraints supplied. Real systems face uncertainty, imperfect forecasts, political and supply-chain limits, and sectors beyond electricity. The study also changes one assumption at a time, which helps isolate effects but does not reveal every interaction among choices. For example, a different turbine, a different layout, and a different wake representation could combine in ways that are not simply additive.
The system analysis uses five critical weather years rather than optimizing across all 85 years at once, and the results depend on its cost assumptions, transmission limits, eligible land, and available technologies. Historical capacity-factor data used as context are themselves uncertain, and modeled profiles do not necessarily match the locations or hardware of the historical fleet. These caveats do not erase the comparison; they define what it can and cannot establish.
The next step is to test these choices in broader scenario ensembles: multiple climate models, alternative future climates, different technology costs, and system designs that represent electrified heat, transport, and industry. It will also be useful to examine how combinations of assumptions affect the same extreme weeks, rather than varying each in isolation. Such work could reveal whether some modeling choices matter most in particular regions or only when several are combined.
The paper’s most useful lesson is therefore methodological and practical at once. Europe’s power-system stress may arrive on much the same days whichever reasonable wind and solar model a researcher chooses. But the amount of energy expected from those resources on those days can shift dramatically—and the capacity plan shifts with it. Getting the low-generation tail right is not a technical footnote. It is part of deciding how much resilience the energy transition needs.