PaperlayerView exact revision on arXiv

Source-linked AI summary

Which wind and solar modelling choices matter? Effects on capacity factors, energy droughts and European power system configuration

L. Karkossa, A. Grochowicz, M. J. Koivisto, A. Hahmann, M. VIctoria

arXiv:2610.39859v1eess.SYcs.SY

TL;DR

In renewable-heavy power systems, rare VRE droughts can shape backup needs, yet it remains unclear how renewable-modelling choices affect their representation and system outcomes relative to weather variability. The paper varies these choices across 85 years of reanalysis and evaluates generation, droughts and system design. It reports that stress-event timing is largely robust across configurations, though some choices reorder the most severe period in two of five weather years.

  • Problem

    It remains unclear how renewable-generation modelling choices affect capacity factors, VRE drought characteristics and system outcomes, or how their influence compares with inter-annual weather variability.

    Source: §1 · p. 2

  • Method

    The study varies renewable-generation modelling choices one at a time using 85 years of ERA5 data for 33 European countries, then evaluates selected configurations in a capacity-expansion and dispatch model for five critical weather years.

    Source: §1 · p. 3, §2.8 · p. 5

  • Results

    System-stress timing is largely robust across modelling choices, but bias-corrected, 150 m hub-height and resource-grade configurations reorder the most severe period in 1941/42 and 2010/11.

    Source: S16 Identified system stress event… · p. 44

  • Takeaways & Limitations

    Renewable modelling choices alter electricity delivered during VRE droughts and resulting system design, sometimes by amounts comparable to inter-annual weather variability.

    Source: §4.4 · p. 13

    The one-at-a-time design cannot capture interactions, while country-level modelling, perfect foresight and omitted non-electricity demands constrain the system analysis and likely underestimate reported system stress.

    Source: §4.3 · p. 12

Abstract

from arXiv · show

Europe's future power system will rely heavily on wind and solar power and must remain resilient during prolonged periods of low wind and solar generation, known as variable renewable energy (VRE) droughts. To capture such rare events, adequacy assessments increasingly draw on decades of weather data, but the conversion of weather into power is largely treated as fixed. It remains unclear which modelling choices in this conversion matter for representing these droughts, and whether they matter as much as the weather itself. Here we systematically compare these choices side by side (bias correction, wind speed extrapolation, capacity layouts, turbine and photovoltaic configurations, and wake losses) and trace their effects through VRE droughts to the design of a net-zero European power system, using 85 years of reanalysis data. We find that modelling choices barely change how often VRE droughts occur or when the system is stressed. They do, however, determine how much capacity is needed to withstand that stress. By altering the electricity delivered during droughts, they shift required capacity by up to 480 GW, which is comparable to the spread across the most critical weather years. Modelling exercises should therefore use multiple weather years; bias-correct and regionally validate weather inputs; explicitly represent resource heterogeneity, forward-looking technologies, and wake losses; and judge these choices by their effect on the low-generation tail, not only on the mean.

1 Introduction

Resilient power-system planning depends on how weather data are converted into renewable generation, because rare VRE droughts—not average conditions—can determine backup capacity. The study compares renewable-modelling choices across capacity factors, drought characteristics and system outcomes, and asks how their effects compare with inter-annual weather variability.

Motivation

  • Backup capacity in highly renewable systems is determined by a few meteorological stress events, or VRE droughts, rather than average operating conditions.Adequacy assessments increasingly use reanalysis and climate-model data, so their outcomes depend on how weather data are converted into renewable generation time series.Source: §1 · p. 2

Research gap

  • The conversion of weather into power has received less attention than weather inputs, although small differences in estimated generation can alter adequacy outcomes.Existing studies examine subsets of bias correction, spatial representation, technology, capacity-layout, wake-loss and wind-speed choices, but not jointly across capacity factors, droughts and system outcomes.Source: §1 · p. 2

Motivation

  • Drought characteristics depend both on identification methods and on how weather is converted into power, making their representation relevant to resilient system design.Short climate records poorly capture inter-annual variability and low-generation periods, motivating the growing use of multi-decadal weather data.Source: §1 · p. 2

Research questions and approach

  • The study tests whether renewable-modelling choices affect annual capacity factors, VRE drought characteristics and optimal system design, and how their influence compares with inter-annual weather variability.It uses a one-at-a-time sensitivity analysis of 85 years of ERA5 data for 33 European countries and five critical weather years.Source: §1 · pp. 2–3

2 Methods

The study varies renewable-generation modelling choices one at a time, converts 1941–2025 weather data into country-level generation series, and evaluates capacity factors, droughts and system outcomes. It then uses selected configurations in a capacity-expansion and dispatch model for five critical weather years.

2.1 Study design and baseline configuration

  • Hourly PV and onshore/offshore wind series for 33 countries over 1941–2025 are produced with atlite and ERA5, varying one modelling choice at a time against a baseline.The choices cover bias correction, wind-speed extrapolation and hub height, spatial representation and capacity layout, technology parameters, and wake losses.Source: §2.1 · p. 3

2.2 Bias correction

  • Wind speed is corrected against Global Wind Atlas data and solar irradiance against SARAH-3; linear correction rescales cell means, while quantile mapping matches the distribution, including its tails.Solar uses monthly empirical distributions, and wind uses a Weibull-to-Weibull transfer function; neither method distorts diurnal or seasonal cycles.Source: §2.2 · p. 3

2.3 Wind speed at hub height

  • Three wind-speed extrapolation schemes are compared at a common 150 m hub height to isolate the extrapolation method.They are logarithmic extrapolation with hourly ERA5 roughness length, a power law using hourly shear, and interpolation between ERA5 model levels.Source: §2.3 · p. 3

2.4 Spatial representation and resource grades

  • Capacity layouts distribute each country’s capacity across grid cells using cell capacity factors and installable capacity, with the exponent setting preference for higher-resource sites.The tested layouts are uniform (γ = 0), baseline proportional (γ = 1), and quality-weighted (γ = 2); resource grades divide cell-mean capacity factors into three equal quality intervals.Source: §2.4 · p. 3

2.5 Technology parameters

  • The technology variants use horizontal single-axis tracking PV and larger, lower-specific-power wind turbines than the baseline.The wind variants are 5.5 MW onshore and 7 MW offshore, compared with 3 MW and 5.5 MW baseline turbines, respectively.Source: §2.5 · p. 3

2.6 Wake losses

  • Wake losses are modelled with a wake-modified wind power curve for representative farms at 4 MW km−2 onshore and 8 MW km−2 offshore.The curve accounts for near-farm and mesoscale losses and is averaged over wind directions.Source: §2.6 · pp. 3–4

2.7 Capacity factor and VRE drought metrics

  • Droughts are identified separately by country, technology and configuration using a seasonal daily p5 threshold; solar uses daily capacity factors and wind uses a 24-hour rolling mean.Solar events can bridge one-day interruptions; wind periods require at least 24 consecutive hours below threshold, and events separated by no more than 24 hours above it are merged.Source: §2.7 · p. 4, Figure 2 · p. 6

2.8 System analysis

  • Eight selected configurations are compared in a perfect-foresight, net-zero electricity-only capacity-expansion and dispatch model for 33 countries and five critical weather years.The model uses hourly resolution, 2023 transmission capacities and 2050 technology costs; biomass, hydropower and pumped-hydro capacities are fixed, while nuclear capacity has a 2023 lower bound.Source: §2.8 · p. 5

3 Results

Across the modelling choices, wind and solar capacity factors and electricity delivered during droughts can shift substantially, even as drought occurrence and system-stress timing remain largely robust. Those changes propagate into system capacity and the balance of renewables, storage, and backup.

3.1 Average annual country capacity factors

  • Horizontal single-axis tracking raises solar capacity factors by 19–26% across the four countries, while high-resource wind sites raise mean wind capacity factors by 11–26 pp.For wind, 150 m hub heights add 3–11 pp, more than larger turbines or wake losses; annual weather-year deviations can reach −12% to +13%.Source: §3.1 · pp. 6–7
  • Bias correction increases wind capacity factors in complex terrain, but its alignment with historical values differs by country: it narrows the gap in Spain and Italy and widens it in Denmark and Germany.Historical comparisons are uncertain because installed-capacity sources differ and reported generation can be affected by curtailment.Source: §3.1 · pp. 7–8, Figure 4 · p. 7

3.2 Periods of low VRE generation

  • Modelling choices change drought severity more than occurrence: frequency and duration deviate by only about 2.4% on average, while delivered shares shift in opposing directions.Bias correction and higher hub heights increase delivered shares; Resource Grade C reduces Germany’s by up to 7 pp, and wake losses deepen shortfalls.Source: §3.2 · p. 8, Figure 7 · p. 10
  • Bias correction and higher hub heights increase drought-period delivered shares, while wake losses deepen shortfalls; solar bias correction raises delivered shares and resource grades and tracking reduce them.For onshore wind, linear correction increases the delivered share more than quantile mapping in complex terrain despite their similar capacity-factor distributions.Source: §3.2 · pp. 8–9

3.3 Different renewable modelling assumptions change the system configuration

  • 295–480 GW of modelling-choice shifts in total system capacity are comparable to the 430 GW spread across five baseline weather years.Backup capacities, such as H2 turbines, are more strongly driven by weather year than by modelling assumptions.Source: §3.3 · p. 9, Figure 8 · p. 10
  • Bias correction and technology refinements reshape system capacity: bias correction raises onshore wind and reduces total capacity by 9–15%, while neglecting wake losses underestimates total capacity by 6–8%.With bias correction, onshore wind replaces solar PV and offshore wind; neglecting wakes shifts capacity toward solar PV, onshore wind, and batteries.Source: §3.3 · p. 9, Figure 8 · p. 10

3.4 System stress events require adequate resource representation

  • Peak net load falls by 8–54 GW under bias correction, horizontal single-axis tracking, and 150 m hub heights, but reduced net load can still coincide with greater backup needs.Bias correction displaces solar and reduces battery fleets; lower battery discharge can lead to average H2-turbine capacity increases of 7% under linear correction and 16% under quantile mapping.Source: §3.4 · p. 11, Figure 9 · p. 11

4 Discussion & Conclusions

Renewable modelling choices barely shift drought frequency or stress timing, but they alter drought severity and required capacity; their effects can rival differences between weather years. The study recommends evaluating choices against low-generation periods and accounting for key modelling assumptions, while noting important limits to its system analysis.

4.1 Comparison with previous work

  • Modelling choices change installed capacity by 295–480 GW, comparable to the 430 GW spread across five critical weather years.The paper concludes that converting weather into generation should be treated as a source of planning uncertainty comparable to the choice of weather years.Source: §4.1 · p. 12
  • Drought frequency and duration change by only 2–3%, and major stress events largely retain their timing, yet changes in drought-delivered electricity shift the capacity needed to cover them.Higher mean capacity factors can coincide with lower delivered shares during droughts: HSAT deepens solar shortfalls in low irradiance, while concentrating wind capacity at high-yield sites can reduce spatial diversity.Source: §4.1 · p. 12

4.2 Implications for energy system modellers

  • Use multiple weather years, since inter-annual variability rivals the most influential modelling choices and exceeds most others examined.Single-year assessments or representative days risk understating uncertainty and missing system-relevant stress periods.Source: §4.2 · p. 12
  • Bias-correct wind inputs and validate them regionally; correction method matters little for wind, while solar quantile mapping lowers winter output and raises backup needs.A single wind-correction dataset without regional validation may perform worse than no correction; the study also evaluates how bias-corrected wind speeds improve agreement with historical duration curves.Source: §4.2 · p. 12, §4.1 · p. 12
  • Represent resource heterogeneity: three resource grades per country reduce total installed capacity by 9–12% versus a single country-level time series.Resource quality matters more than within-country capacity layout, especially for onshore wind; several grades are a practical alternative when high spatial resolution is infeasible.Source: §4.2 · p. 12
  • Include forward-looking technologies and wake losses: higher hub heights help during droughts, HSAT has opposing mean-yield and drought effects, and offshore wakes increase installed capacity by 6–8%.Offshore wake losses reduce mean capacity factors by 8–13% and deepen drought shortfalls; the analysis includes only within-farm losses, so effects may be conservative. Larger turbines have a smaller impact, onshore wake effects are modest, and logarithmic versus power-law extrapolation or model-level winds differ little at country scale when 100 m winds are available.Source: §4.2 · p. 12

4.3 Limitations and outlook

  • The one-at-a-time, country-level, perfect-foresight electricity-only analysis omits interactions, local balancing, and coupled demand; technology capacity benefits are upper bounds, and generic configurations limit capacity-factor validation.Future work could test interactions, higher spatial resolution, sector coupling, and climate projections; historical validation is also constrained by inconsistent capacity and generation data and unreported curtailment.Source: §4.3 · pp. 12–13

4.4 Concluding remarks

  • Assess renewable resources by what they deliver during the scarcest periods, not only by average output; annual-generation-based siting and support schemes may undervalue scarcity-period performance.As multi-decadal weather records become standard in adequacy assessments such as ERAA, the conversion of weather into generation warrants scrutiny comparable to the choice of weather data.Source: §4.4 · p. 13

Supplementary Material

The supplementary material defines abbreviations and groups symbols used for time, space, drought events, bias correction, wind profiles, and capacity-factor analysis.

Abbreviations

  • The abbreviations include VRE for variable renewable energy, PV for photovoltaics, CF for capacity factor, and ERA5 for the fifth-generation ECMWF atmospheric reanalysis.Other listed abbreviations include HSAT for horizontal single-axis tracking and ERAA for European Resource Adequacy Assessment.Source: Abbreviations · p. 18

Symbols

  • The symbol glossary covers grid-cell and time indices, drought-event measures, bias-correction variables, wind speed and vertical profiles, and spatial capacity representation.It also defines seasonal capacity-factor references, delivered electricity shares, layout weights, and eligible area.Source: Symbols · pp. 18–19

S1 Literature gap

The literature review identifies a gap in jointly evaluating renewable modelling choices across generation, drought, and system outcomes. The supplementary material sets out the baseline assumptions and land eligibility constraints.

S1 Literature gap

  • Earlier studies compared alternative renewable modelling choices within single frameworks, but none evaluated them jointly across capacity factors, VRE droughts, and system outcomes.Figure S1 uses circles for wind and solar choices, squares for evaluated metrics, and filled versus grey symbols for varied versus fixed or unconsidered choices; its literature selection is illustrative, not exhaustive.Source: S1 Literature gap · p. 20, Figure S1 · p. 20

S2 Baseline configuration

  • The baseline uses uncorrected ERA5, static south-facing 35°-tilt silicon PV, specified onshore and offshore turbines, logarithmic wind extrapolation, proportional capacity layout, and no wake losses.The onshore turbine is 3 MW with an 80 m hub height; the offshore turbine is 5.5 MW with a 120 m hub height.Source: S2 Baseline configuration · p. 20
  • Eligible land excludes protected areas and follows CORINE land cover; offshore sites are limited to water shallower than 60 m, with technology-specific installation-density caps.The stated caps are 5.1 MW km−2 for utility-scale solar PV, 4.43 MW km−2 with HSAT, 3 MW km−2 for onshore wind, and 2 MW km−2 for offshore wind; these are averages over eligible areas, not individual farm densities.Source: S2 Baseline configuration · pp. 20–21

S3 Bias correction

Wind and solar inputs are bias-corrected against higher-resolution reference datasets, using either uniform scaling or distribution-based mapping. The correction methods differ in how they adjust resource distributions, while their temporal profiles are largely preserved.

S3.1 Reference datasets

  • Solar correction uses SARAH-3 data from 2011–2020; wind correction uses GWA data from 2008–2017, both compared with ERA5.Solar direct and diffuse irradiance are corrected separately; wind correction is applied to 100 m wind speed.Source: S3.1 Reference datasets · p. 21

S3.2 Linear correction

  • Linear correction applies a grid-cell factor based on the reference-to-ERA5 mean ratio uniformly across hours and weather years.Wind correction maps vary regionally, with strong upward corrections in complex terrain and limited GWA v2 coverage.Source: S3.2 Linear correction · p. 21, Figure S2 · p. 21

S3.3 Quantile mapping

  • Quantile mapping matches values at each percentile to the reference distribution, including its tails; solar uses monthly distributions, while wind uses a Weibull-based transfer.For wind, GWA's static Weibull parameters are combined into a cell-level reference distribution.Source: S3.3 Quantile mapping · pp. 22–23
  • Quantile mapping leaves solar diurnal and seasonal profiles largely intact, while wind configurations mainly shift capacity-factor levels without changing the seasonal cycle.In Denmark, quantile mapping substantially lowers winter midday solar capacity factors.Source: S3.3 Quantile mapping · p. 23, Figure S5 · p. 23, Figure S6 · p. 24

S4 Estimation of wind speed at hub height

The study estimates hub-height wind speed using alternative extrapolation methods and direct interpolation of ERA5 model levels. The order of operations in model-level interpolation can affect calculated speeds, but the difference is negligible at the close vertical spacing used.

S4 Estimation of wind speed at hub height

  • The baseline extrapolates 100 m ERA5 wind speed with the hourly roughness length under neutral stratification; a power-law alternative uses 10 m and 100 m speeds.Both methods estimate wind speed at the target turbine hub height.Source: S4 Estimation of wind speed at hub… · p. 27
  • Model-level interpolation reconstructs hourly layer heights and interpolates wind speed between the two levels bracketing hub height.The height reconstruction uses layer temperature, humidity, pressure, and the hypsometric equation.Source: S4 Estimation of wind speed at hub… · p. 27
  • Taking wind-speed magnitudes at each model level before interpolation avoids vector cancellation, but the resulting difference is negligible because ERA5 levels are closely spaced.Wind speed at each level is the magnitude of its eastward and northward components.Source: S4 Estimation of wind speed at hub… · pp. 27–28

S5 Resource grades

Resource grades preserve within-country variation in wind and solar quality by grouping grid cells into three tiers and representing each tier separately in the system model.

S5 Resource grades

  • Each country’s grid cells are divided into three equal intervals of mean capacity factor, forming Grade A, B, and C from best to worst resource.Grades are defined from the 1941–2025 mean and held consistent across years; the split is by resource quality, not area or capacity.Source: S5 Resource grades · p. 28, Figure S9 · p. 28

S6 Technology parameters and wake losses

The technology and wake-loss configurations alter renewable capacity factors, with particularly large effects for offshore wind and winter solar. The reported offshore sensitivities fall between those of solar PV and onshore wind.

S6.1 Solar tracking

  • HSAT panels track the sun east to west, broadening generation into morning and evening hours while reducing low-sun-season midday peaks.Wider row spacing for tracking lowers maximum installation density from 5.1 to 4.43 MW km−2.Source: S6.1 Solar tracking · p. 28, Figure S10 · p. 29

S6.2 Wind turbines

S7 Winter capacity factors

  • Winter solar quantile mapping reduces capacity factor more than linear correction, by up to 1.2 pp in Denmark, or 28 % relative to the low winter level.Winter onshore wind bias correction increases capacity factor by up to 9.8 pp in Italy.Source: S7 Winter capacity factors · p. 29, Figure S11 · p. 30

S8 Offshore wind annual and winter capacity factors

  • Offshore wake losses reduce mean capacity factor by 2.9–4.7 pp (8–13 %), twice the onshore reduction, while the larger 7 MW turbine changes it by only 0.3–0.5 pp.The offshore wake effect reflects the denser plant layout assumed in the model.Source: S8 Offshore wind annual and winter… · p. 30
  • Offshore bias correction raises mean capacity factor by 2.5–5.7 pp (5–21 %); spatial layout shifts it by up to 3.9 pp in Spain and Italy.The layout effect is less than 0.5 pp in Denmark and Germany, and the overall offshore spread is smaller than onshore wind's.Source: S8 Offshore wind annual and winter… · p. 30, Figure S13 · p. 31

S9 Wind speed bias correction factors and capacity factor distributions

Wind bias correction can reshape wind-speed distributions, but much of that variation is dampened when wind is aggregated into country-level capacity factors. Across wind and solar droughts, modelling choices usually alter event frequency and duration little, while changing the electricity delivered during droughts; these effects vary by technology, country, and season.

S9 Wind speed bias correction factors and capacity factor distributions

  • Wind bias correction shifts country-level capacity-factor distributions only slightly at the tails, despite distinct quantile-specific wind-speed corrections.In Spain and Italy it brings the distribution closer to historical observations, while Denmark shows the largest deviation.Source: S9 Wind speed bias correction factors… · pp. 31–32

S10 Bias correction versus calibrated power-curve smoothing

  • Calibrated power-curve smoothing combines possible weather bias, capacity layout, turbine-fleet heterogeneity, and hourly conversion effects, obscuring which mechanism drives agreement with historical curves.Evaluating assumptions separately tests how much wind-speed bias correction alone improves that agreement.Source: S10 Bias correction versus calibrated… · p. 32

S11 Onshore wind energy droughts

  • Onshore drought frequency and duration deviate from baseline by about 2.4% on average, while Resource Grade C has the largest effect on Europe’s load-weighted delivered share.Effects on delivered share vary across countries and seasons; wake losses reduce it by less than 1 percentage point in all cases.Source: S11 Onshore wind energy droughts · p. 32

S12 Solar energy droughts

  • Solar droughts average 15.4 events per year and 1.46 days, and Resource Grade C reduces the drought-period delivered share by 9.6 percentage points on average across the four countries.Solar frequency and duration deviate by about 3% on average, but Resource Grade C produces country-season changes up to +30%/−19%; HSAT reduces the delivered share by up to about 5 percentage points.Source: S12 Solar energy droughts · p. 35, Figure S17 · p. 35, Figure S18 · p. 36

S13 Offshore wind energy droughts

  • Offshore-wind drought frequency and duration deviate by about 2% on average, while Resource Grade C reduces delivered share by 3.9 percentage points on average across the four countries.Wake losses reduce offshore delivered share by up to about 2.4 percentage points in Spain, more than the less-than-1-point onshore effect.Source: S13 Offshore wind energy droughts · p. 36, Figure S19 · p. 37, Figure S20 · p. 38

S14 European electricity shortfall during energy droughts and sensitivity to the drought threshold

  • The directions and ranking of modelling-choice effects on load-weighted drought-period delivered shares remain consistent when thresholds change from p5 to p10 or p15.The comparison covers onshore wind, solar PV, and offshore wind; the other event-identification rules are unchanged.Source: S14 European electricity shortfall… · p. 39

S14.1 Main threshold (p5)

The supplementary analysis compares load-weighted European electricity delivered during p5 droughts and repeats the comparison with p10 and p15 thresholds. It also examines how modelling choices affect annual supply and the timing of severe system stress periods.

S14.1 Main threshold (p5)

  • Load-weighted European drought-period delivered shares are compared with each technology’s baseline for onshore wind, solar PV, and offshore wind.The p5 figures use demand weighting for onshore wind and include all countries for solar PV and countries with offshore wind for offshore wind.Source: Figure S21 · p. 39, Figure S22 · p. 40, Figure S23 · p. 40

S14.2 Threshold sensitivity (p10 and p15)

S15 Changes in the supply of the European energy system

  • 5.6%: total energy supply varies by up to 5.6% (−238 to +85 TWh) across configurations despite identical demand of about 3,300 TWh per weather year.Differences stem mainly from batteries: bias-corrected systems average 358 TWh (linear) and 312 TWh (quantile mapping) of battery discharge, versus 507 TWh in the baseline.Source: S15 Changes in the supply of the… · p. 44, Figure S27 · p. 44

S16 Identified system stress event overview

  • The timing of severe stress periods is largely robust across configurations in three of five critical weather years, but some choices shift the most severe period in 1941/42 and 2010/11.The original baseline period still causes high stress in those two years.Source: S16 Identified system stress event… · p. 44, Figure S28 · p. 45

S17 Battery discharge and H2-turbine energy during stress events

During stress events, bias-corrected systems use less battery discharge and generally more H2-turbine energy than the baseline, with a different pattern in 2010/11.

S17 Battery discharge and H2-turbine energy during stress events

  • 30–37%: bias-corrected systems have smaller battery fleets, discharge less during stress events, and use more H2-turbine energy than the baseline except in 2010/11.Quantile mapping has the lowest battery discharge and, except in 2010/11, the most H2-turbine energy.Source: S17 Battery discharge and H2-turbine… · p. 51, Figure S34 · p. 51
Loading 2610.39859v1…