The impact of future UK offshore wind farm distribution and climate change on generation performance and variability

The UK plans to significantly increase offshore wind generation capacity as part of the effort to achieve net zero targets. Current installation is densely located in a few areas, particularly off the east coast of England, and although current siting proposals include new offshore regions, significant volumes of wind generation capacity are yet to be located to meet 2050 installation targets. This paper uses a recent dataset of multi-decadal offshore wind power capacity factor timeseries to assess how UK offshore wind generation is likely to be affected by both the spatial distribution of future wind farms, and by the impacts of near-future (2020–2050) climate change. We determine that a wider geographic spread of offshore capacity results in a much-improved and less-variable UK-aggregated power generation profile, with substantial reductions in periods of low generation and extreme wind power ramping events, without negatively impacting mean or peak generation outputs. The impact of near-term climate change appears to be minor, slightly reducing overall generation and possibly resulting in an underestimation of future installation requirements, but this climate signal is outweighed by the effects of spatial distribution, and even more so by inherent hourly to inter-annual wind speed variability. This study implies that the intermittency of wind generation can be partly mitigated through increasing the spatial diversity of the existing wind farm distribution. Alongside a more in depth investigation of future climate change, and a holistic assessment of relevant geospatial factors such as Levelised Cost of Energy, infrastructure, and environmental constraints, this study could be used for optimisation of future offshore wind siting.


Introduction
Transitioning from non-renewable fossil-sourced energy generation to low emission technologies is a critical foundation of the international effort to decrease anthropogenic carbon emissions [1].Wind energy is a key component of the UK's current climate change policy, with expansion of offshore wind vital to meeting the net zero commitment.The UK Government's target is to increase installed offshore wind capacity from the existing 14 GW [2] to 50 GW by 2030 [3], with the climate change committee projecting up to 140 GW by 2050 [4].Currently, offshore wind farms are largely concentrated in relatively shallow North Sea waters [2], and whilst there are already many new wind farms either confirmed or under construction, important decisions are still to be made about where to place additional wind power capacity to maximise both grid resilience and economic feasibility.Many factors will influence these decisions (e.g.geophysical, financial, infrastructure availability), but a key starting point is the spatial distribution of near-surface wind speeds and their spatiotemporal variability in UK offshore waters.
Variability in UK wind speeds considerably impacts wind power generation [5][6][7] and this has implications for future electricity market design [8][9][10].High, steady wind speeds (∼10-20 ms −1 ) generally benefit wind farm owners via consistent energy generation yields at the peak of offshore wind power capacity curves (i.e. the wind speed range at which turbines produce their nominal power output, and where wind speed fluctuations within this range does not impact output-see [11] for the power curves used in this study), but the intermittency of wind speeds creates several challenges: short-term hourly/daily variability requires significant grid supply management during periods of supply/demand mismatch, whilst longer-term (monthly/annual/ decadal) variability inherently creates uncertainty for revenue forecasts.It is becoming common practice in the energy-meteorology community to use multi-decadal synthetic demand, wind, and solar power data for wind farm site assessments and to model power system operation, with many openaccess datasets available [12][13][14][15].
We are also in a period of rapid climate change [1], which is having consequences for global wind speeds, with multiple studies now writing about the concept (and recent reversal) of global stilling [16][17][18].Future climate model projections generally show decreases in near-surface wind speeds across the whole of Europe, particularly for high end climate scenarios, and for summer [19,20].However, at present there is an emerging consensus that impacts of inter-annual climate variability on regional and global wind power production are large compared to the projected 21st century impacts of climate change [19,[21][22][23].
As well as inter-annual climate variability, the distribution of wind farms has been shown to strongly influence output [24].Drew et al showed that the addition of 50GW of new wind capacity, mostly in the North Sea (at the time, those proposed by 2025) could increase mean capacity factors from 2014 values from 32.7% to 39.7%.
Past academic work on wind farm distributions has focused on optimising the spatial variability of aggregated renewable generation, commonly over a continental scale, including both wind and solar power [25][26][27][28][29], but tending to utilise meteorological reanalysis datasets that recreate the past climate (by combining observational data with numerical weather prediction models), therefore focusing on the historical period rather than considering the potential impacts of climate change.A recent study by the energy consultancy Regen [30] found that diversifying UK offshore wind geographic siting can reduce generation variability and benefit electricity system balancing.Elsewhere, extreme characteristics of wind power production have been highlighted as an operational issue, such as persistent high/low generation [31] and short timescale wind power ramps [32], the impact of which could be exacerbated by planned construction of large offshore wind farms in close proximity [24].
A direct analysis of the relative importance of these three factors-wind variability, wind farm distribution, and climate change-is yet to be performed for the UK, and is greatly needed.The aims of this study are therefore to: 1. Determine how the spatial distribution of future UK offshore wind farms influences potential power generation output, characteristics, and variability.2. Investigate the influence of near-term (2020-2050) climate change on the UK offshore wind portfolio.
The next section outlines the data and methods used to investigate potential power generation under future scenarios of climate change and wind farm distribution.Section 3 describes our findings in terms of the wind power generated under each scenario, and how it varies between years, months, and days (section 3.1), as well as differences between scenarios in the number and magnitude of extreme ramping and persistence events (sections 3.2 and 3.3 respectively).This is followed by a concluding discussion on how our findings impact and relate to wider energy system operation and development.

Wind power potential data
In this study hourly wind power capacity factor timeseries from Bloomfield et al's 2022 dataset [13] are used.This dataset is derived from gridded 100 m wind speed outputs from the ERA5 reanalysis [33], at ∼30 km resolution, from 1950-2020.ERA5 data are initially bias corrected against Global Wind Atlas data [34] to account for sub-gridscale topographic effects.These bias-corrected wind speeds are then scaled to 92 m (average offshore turbine hub height in the UK in 2021) using a wind profile power law [35,36], before being passed through representative onshore and offshore wind farm power curves to provide the wind capacity factor at each ERA5 gridpoint and timestep.
Bloomfield et al's 2022 offshore wind capacity factor data has then been spatially aggregated over the UK Met Office shipping zones [37], averaging all outputs within each of the 16 zones that lie within the UK's exclusive economic zone to a single value at each timestep (see appendix c for zone boundaries).It is not possible to directly verify the wind power values from this dataset as they do not correspond to particular wind farm locations.However, comparison against Bloomfield et al's companion locationweighted dataset with values taken only at locations of operational wind farms (as of April 2021), indicates that fine-scale geophysical processes that affect generation performance (e.g.coastal wind boundary impacts, turbine wake effects) are not fully captured, resulting in higher GB-aggregated mean annual capacity factors than expected.In this study, a correction to account for these processes has been implemented, and is incorporated into the power data generation, as discussed in section 2.4.Climate change-adjusted outputs from [13] are also used to investigate the impact of near-term climate change (i.e. the mean change from 1980-2010-2020-2050 under the representative concentration pathway 8.5 (RCP8.5)low mitigation scenario).These are derived by interpolating 10 m wind speed outputs from five climate models with similarly high spatial and temporal resolution (see [13] for details of the selected models) to ERA5's ∼30 km resolution, and calibrating the reanalysis data to represent a future climate via a seasonal quantile-based deltacorrection.This correction accounts for the potential difference in 10 m wind speed at each gridpoint in a future climate.This seasonal quantile adjustment allows for analysis of the future change in seasonal wind speed, but the conservation of the reanalysis data weather sequence inevitably ignores any changes (e.g.extreme weather event freqencies) as a result of climate change [38] as the same underlying ERA5 weather is still present.The merit of this approach is that users are able to see the impacts of near-term mean climate change on notable extreme events from the recent past (e.g.what would the 1963 cold-wave look like in a near-future average climate?).The ERA5 timeseries will be referred to as the 'historical' climate conditions, and the climate-change adjusted timeseries will be referred to as 'future' climate conditions.

Wind farm location data
Existing and proposed offshore UK wind farm locations as of August 2022 [39,40] are spatially allocated to their corresponding shipping zone (see appendix c for shipping zone locations), with the capacity of any wind farm located partially outside of the 16 zones included as part of the nearest zone (e.g.all 'Celtic Sea' projects are considered to be within the 'Lundy' shipping zone).
The development status of each located wind farm project is determined as per The Crown Estate's July 2022 Project Listings [41] ('Fully Commisioned' and 'Committed' projects considered "Existing; 'Under Development' , 'Pre-Planning' and 'Future' projects considered 'Proposed').Figure 1 shows details of these wind farm locations.

Wind farm distribution scenarios
To investigate the potential operation of a 2050 energy system heavily reliant on offshore wind power generation, the total installed offshore wind capacity used throughout this study is 140 GW, aligning with the climate change committee's upper estimate for 2050 [4].
We create three wind farm distribution scenarios to explore the impact of alternative options for spatially allocating the 140 GW of future wind capacity in each of the 16 zones: It is assumed that all existing operational wind farms remain or are replaced in the established and proposed scenarios, and that the build-out rate for all scenarios is instant (i.e. the total installed capacity is available and fixed from the beginning of each time series).Installed capacities for each shipping zone by scenario are provided in appendix A.
Note that the distributed scenario has an equal amount of capacity installed in each zone, but as the zones vary in area, it does not have an equal density of installation, with the smaller near-shore 'Dover' shipping zone equalling an area 13x smaller (and therefore a 13x higher installation density) than the large 'Fair Isle' zone.It is not suggested that the Distributed scenario placement is feasible, but is used here to represent a hypothetical extreme example of high wind farm spatial distribution beyond that of the proposed scenario, extrapolating the trend identified between the established and proposed scenarios of increasing geographic dispersion.

Generating wind power timeseries for each shipping zone and scenario
A 70 year, hourly wind power capacity factor time series (P t ) is created for each of the climate datasets (derived from ERA5 and the 5 climate projections), for each wind farm distribution scenario, for each shipping zone.This is created for each climate dataset (m), by multiplying the unitless hourly (t) capacity factor data (CF t ) by the installed capacity in GW (IC) in that zone (z) and wind farm distribution scenario (s).
Comparing the ERA5-derived capacity factor data for each shipping zone against measured recent UK output [42] highlights a small discrepancy.These capacity factor values (determined as a national aggregate of power output from actual operational offshore wind capacity) consistently indicates values ∼6% lower than the particular ERA5-derived modelled values at the same locations throughout the timeseries, reflecting the absence of wind farm wake effects in the climate data and non-weather driven wind power curtailment.Applying a uniform 6% reduction to the capacity factor data at the operational windfarm locations reduces the 70 year ERA5derived mean of 48.32% to 45.42%, aligning closely to the highest observed average annual offshore wind capacity factor (45.69% in 2020).This 6% correction is applied to all of the shipping zones uniformly as an initial representation of these impacts; other possible methods for correcting this dataset are discussed in the final section.
This correction (δ = 0.94) is applied as part of the power profile generation, such that: Statistical significance testing of each scenario and climate model's timeseries is assessed using independent two-sample t-tests with a p-value threshold of 0.1.

Electricity grid extreme event metrics
To investigate the impact of the selected climate and distribution scenarios on extreme events for the electricity grid, several metrics are calculated.
The volatility of nationally-aggregated power generation is indicated via 'ramping events' , defined as the change in magnitude of wind power output between two points in time (R(t)), such that: The magnitude, frequency, and direction of hourly, 6 h, 12 h, and daily ramps are determined for the entire timeseries-each ramp event could therefore appear multiple times for multi-hour events.Substantial grid-aggregated ramp events may result in cascading impacts throughout the energy system, requiring different operational procedures from system operators depending on the scale and direction of ramp (e.g.curtailing generation during 'up'-ramps, or dispatching peaking plant during 'down'-ramps).
Long periods of persistently high or low generation are also identified as a function of capacity factor, with periods with less than 5% and 10% of the total UK capacity factor identified as 'persistent low events' , and periods with more than 80% and 90% being identified as 'persistent high events' , as in [6,24].The length of each persistent event below/above these low/high thresholds are also determined and aggregated to find the mean number of events per timeseries year.

Results
The potential for offshore wind power generation is examined under three wind farm distribution scenarios, and under historical and future climate conditions.Section 3.1 describes how the potential for power generation varies between years, months, and days under each condition.We then explore how spatial distribution and future climate change affects the incidence of extreme generation events across the full time series, with sudden changes of nationally-aggregated generation (ramping) examined in section 3.2, followed by analysis of persistent periods of extreme generation in section 3.3.

Variability in wind power generation: from inter-annual to daily
Figure 2 illustrates the variability in wind power generation from the three spatial distribution scenarios (see figure 1), under historical and future climate conditions.The turbine power curves used in this study [11] result in high capacity factors (>75%) for wind speeds between 10 and 23 ms −1 .Low capacity factors therefore indicate wind speeds being too low or high for nominal power outputs, and when aggregated over time, could indicate periods of high variability as wind speeds alternate between and within these toohigh or too-low ranges.
Regardless of climate model or spatial distribution, the profiles reflect climatological UK offshore wind conditions [7], with mean UK-aggregated capacity factors consistently higher and more variable in winter months (historical Oct-Mar mean standard deviation, σ, = 0.036) than in summer (historical April-September mean σ = 0.024) (figure 2(a)).The overall inter-annual variation in the UK is substantial, with the mean annual capacity factor varying by an average of 5.7% from year to year.Capacity factors are higher on average in shipping zone regions that are further offshore, compared to sheltered coastal regions with lower wind speeds (figure 2(b)).This appears as a latitudinal gradient, though Laurilia et al [43] show that for near-surface wind speeds in the region, land-sea gradients and jet stream proximity has a more dominant influence.The seasonal wind power generation cycle is illustrated by the interquartile ranges of monthly capacity factor values shown by the boxes in figure 2(c).From the winter peak in December and January, there is a reduction in capacity factors through Spring, to a minimum in June and July, followed by an increase in Autumn.The mean daily generation (shown in figure 2(e)) also follows this seasonal cycle, but the range of daily values is much larger, throughout the year.There are days where mean capacity factors approach 0% and exceed 90% in all months, highlighting the large variability of wind power at shorter temporal scales.The exposed northern zones (e.g.'Fair Isle' , 'Hebrides' , 'Forties') have less variable monthly (figure 2  the regions within the English Channel and Irish Sea, which generally have lower CF values, but increased variability.This may explain the results of Drew et al's study [24], where the addition of future (at the time) capacity to the exposed northern regions results in higher and less variable nationally-aggregated capacity factors.The east-coast regions of present-day wind-farm concentration are shown to have relatively moderate capacity factor magnitudes and variability, but it is noted that, as the ∼30 km gridded capacity factor data is aggregated over entire shipping zones, the higher wind speeds in regions far from existing wind farm locations (i.e.further offshore) result in shipping-zone-aggregated capacity factors slightly higher than those observed for the current operational, near-shore fleet.
Examining the impact of climate change, we note that mean annual capacity factor under all future climate models is slightly reduced compared to historical (figure 2(a) shows reductions of 2.3%, (p = 0.003) averaged across all models, individual models shown in appendix b), with the climate change impact distinctly significant only during the summer months (summer = − 3.6% (p = 0.0003); winter = − 1.5% (p = 0.1559)) (figure 2(a)), as seen in Bloomfield et al's original study [35], as well as others [19,20].Inter-model variation is low (see appendix b) however, a detailed analysis of the individual future climate models is not undertaken in this study.
Turning to the spatial scenarios, figure 2(a) shows that the seasonal mean capacity factor (and therefore, power generation) in a given year results in no statistically significant change as a result of the dispersion of wind farm concentration from the eastern Established regions, either to the northeastern/south-western proposed areas (+0.5% on average (p = 0.57)) or to the broad distributed scenario (−0.6% on average (p = 0.47)).The distribution of wind farms has a bigger effect on the variability of generated power from year to year.The interannual variability in the proposed and distributed scenarios is reduced compared to the established scenario (historical proposed mean σ = −4.8%;distributed σ = −9.4%).This reduction in interannual variability as a result of spatial scenario is more notable in summer than winter, with the standard deviation of the summer mean capacity values being 16.5% smaller under the historical distributed scenario than the historical established value, compared to a reduction of only 11.4% in winter.The reduction in output variability via distribution is more substantial at shorter timescales of months and days (figure 2(c) and (e)) (historic proposed mean daily σ = −6.8%;distributed σ = − 15.5%), and notably, increasing wind farm distribution results in disproportionally fewer days of low generation than high (figure 2(e)), with the minimum daily average bottom decile of power increasing by 27.9%, compared to a reduction in top decile output by 5.5% (historical, distributed change relative to established).
The analysis of variablity from interannual and daily therefore indicates an important role for distribution in mitigating the intermittency of wind power, particularly at higher temporal scales.To explore this further we now consider the hourly timescale, starting with an example of an event that posed challenges for the UK energy system: the '2020 Cold Snap' (figure 3).In late December, low temperatures were associated with anomalously high electricity demand.At the same time, variations in wind lead to pronounced variations in capacity factor of offshore wind.We estimate what the capacity factor would have been under the three distribution scenarios.Periods of high generation are retained in the proposed and distributed scenarios, whilst periods of low generation are significantly diminished, with the minimum capacity factor increasing from 0.11 (established) to 0.16 (proposed) and 0.30 (distributed).This results in a profile 3x less variable than the established scenario (distributed σ = −29.9%relative to established; proposed σ = −11%), and the magnitude of the sudden capacity factor ramps early and half-way through the period are dampened.
This case study provides an example of how increasing the distribution of wind farms can result in substantial reductions in generation variability and periods of low generation without negatively affecting mean or peak generation; however, it is just one example.In the next sections we analyse ramping events and persistence throughout the dataset to investigate whether and to what extent distribution can reduce extreme generation events throughout the timeseries.

Extreme generation events-ramping
Figure 4 visualises the magnitude and frequency of intradaily variability via 'ramp events' (see definition in section 2.5), for the UK aggregating across all shipping regions, for each distribution scenario, and for historical and future climate.Each column shows the number of descending/ascending ramps, where the total potential power generation decreases/increases over a period as a proportion of the total future installed capacity across the UK (e.g. a 70 GW ramp is equivalent to ∆CF = 0.5, as we have 140 GW installed in each scenario).Though the annual mean is indicated in the results, significance testing confirms that the effects of scenario or climate on ramping are not sensitive to the choice of dates used throughout the dataset.
The proposed (green) and distributed (blue) ramping curves are steeper than the established (orange) at all timescales, with relatively fewer large ramps compensated by a relative increase in smaller ramps.The mean absolute hourly ramp of the distributed scenario is decreased relative to the established scenario by 27.97% (−27.4% 6 h, −22.4% 24 h).For extreme ramps, the 95th percentile ramp decreases from established to distributed by 33.5% h (−32.6% six-hourly, −24.4% 24 h), and the mean size of the largest hourly ramp in a year decreases by 33.4% (−29.3% 6 h, −12.1% 24 h).The larger 'up' ramps (figures 4(a), (c) and (e) subplots) are slightly more numerous than the 'down' ramps (figures 4(b), (d) and (f) subplots), possibly indicating that weather fronts associated with high-generation enter offshore regions more rapidly than they exit (this would be an interesting topic for future research).
These improvements via spatial distribution slightly decrease in scale for longer timescales, but it should be noted that longer ramps (e.g. over 24 h) correlate with amplitude, and so even though the % reduction via increased distribution is lower at longer timescales, these scale of these ramps are likely to have a more significant impact on grid management.
The impact of future climate is minor, with the future data (averaged across all models) following a marginally lower profile below the ERA5 data-the mean hourly ramp of the Distributed scenario under the future data is 1.18% smaller than the historic data (−1% 6 h, −1.1% 24 h), and the largest hourly ramp across the 70 year distributed profile is reduced by 1.6% (−0.4% 6 h, +0.1% 24 h).
As in section 3.1, explicit benefits result from increased spatial distribution in both ramping event frequency and intensity, with only modest impacts resulting from near-term climate change.The proposed scenario shows consistent improvements over the established scenario, with performance metrics ∼40%-50% of those from the distributed scenario.This indicates that the current planned future wind farm siting will reduce the frequency and intensity of these sudden generation ramps, but it is possible to provide a greater benefit by reducing siting density further.

Extreme generation events-persistence
low generation is evidently undesirable, but extended periods of high generation also have several repurcussions for stable grid operation.Wind turbines inherently lack rotational inertia, which could increase the susceptibility of a highly-renewable energy system to voltage frequency fluctuations, and sufficient energy storage will be required at periods where generation outweighs demand to avoid power curtailment.Figure 5 shows the periods of persistent low/high generation for each distribution and climate scenario, averaged over the UK (a)-(d), and for each region (e)-(f).Sensitivity tests were conducted to confirm that results are not dependent on the timeseries data used throughout the dataset.
The proposed and distributed scenarios substantially mitigate the occurrence of persistent low (figures 5(a) and (b)) and high (figures 5(c) and (d)) wind events at the UK-wide scale, relative to the established scenario, both in terms of frequency (figures 5(a)-(d) subplots) and intensity (as shown by both scenarios' frequency distribution curve lying beneath the established's).Increased distribution reduces the lengths of the 3 longest extreme low capacity events (<5%) in the full 70 year time series from 69, 62, and 61 h (established ERA5), to 56, 43, and 36 h (proposed ERA5), and to 34, 32, and 22 h (distributed ERA5), with the future climate models slightly increasing the lengths of these three longest events to 70, 64, and 63 h (established), 58, 46, and 42 h (proposed), and 46, 32, and 24 h (distributed, note these results are averaged over the five Ramp-event frequency distribution curves from three wind farm distribution scenarios, under historical (solid) and future (dashed) climate conditions.Main panels show the number of periods per year where the change in total power generated across all shipping zones is above a particular magnitude (ramp event, measured as proportion of total installed capacity factor, ∆CF) for a period of (a), (b) 1 h, (c), (d) 6 h, and (e), (f) 24 h.Panels (a), (c), and (e) show ramp-events where total generated power is reduced (ramp down).Panels (b), (d), and (f) show ramp-events where total generated power is increased (ramp up).Inset panels show the mean annual count of ramp-events above a threshold of (a), (b) 5%, (c), (d), 20%, and (e), (f) 50% of the total installed capacity, indicated on the frequency distribution as a dashed black line.The coloured distribution scenarios show ERA5 data, with future climate shown for distributed scenario comparison only, illustrated by the average of the 5 climate models.future climate projections).The most notable difference between scenarios is found during the more extreme low (figure 5(a)) and high (figure 5(d)) persistent events, with the Proposed scenario spending on average 65% and 47% fewer hours in each respective extreme state annually, increasing to 87% and 71% under the distributed scenario.
The impacts of near-term climate change slightly increases the low-wind event frequency, and decreases the high-wind event frequency, reflecting the slight average reduction in near-surface wind speeds identified in the climate models.The extreme high wind events (i.e.those where capacity factors are >90% total installed capacity, figure 5(d)) are most strongly impacted, with total annual hours in each state reducing by 30%-50%, flattening each scenario's event frequency profile almost as much as the change between distribution scenarios.This implies that the climate change impact does not affect the wind speed distribution equally, with periods of high wind generation reduced moreso than periods of low generation.Further investigation analysing climate model impacts across high and low wind speed periods (such as in Carvalho et al [19]) is recommended.Under a distributed future climate scenario, this could result in a considerable reduction in periods (relative to the historical established case) where extensive volumes of power are being generated nationally-depending on demand conditions, this could either result in reduced wind power curtailment, or a requirement to increase residual (nonwind) power generation.
Perhaps counterintuitively, we find these improvements by spatially distributing into nonestablished regions that spend more time in high and low persistent states (e.g.Dover, Portland, Forties, Hebrides) (figures 5(e) and (f)).This is due to the fact that capacity factor between shipping zones inversely correlates with distance (see Pearson correlation matrix in Bloomfield et al [35])-i.e.spreading out wind farm locations results in a lower impact of localised wind events on the nationally aggregated generation, as persistent periods in one region are less likely to coincide with those in distant regions.Increased spatial distribution therefore helps mitigate the negative outcomes of localised weather events by homogenising the national generation profile, helping ensure a more consistent and managable power supply.

Discussion and conclusions
To meet the UK government's net zero target, an unprecedented expansion of offshore wind installation is required, both in scale and geographical domain.Whilst the inherent variability of wind speeds cannot be controlled, the subsequent variability and characteristics of wind power supplied to the grid can be managed via spatial design of the energy system, with our study indicating that reducing the spatial density of offshore wind installations can: • Increase annual, monthly, and daily minimum output generation, with minimal impact on maximum or mean generation (figure 2); • Smooth the UK-aggregated generation profile by reducing variability across multiple timescales (figure 2), resulting in more consistent power generation, and; • Reduce the frequency and intensity of extreme ramping events (figure 4) • Greatly reduce the frequency, intensity, and lengths of extreme and persistent wind events (figure 5).
The scale of these benefits are stronger the more wind farms are spatially distributed, with the proposed scenario (extrapolating the UK's current siting plans) showing consistent improvements above the current established spatial approach, offering encouraging prospects for the future UK energy system if these strategic siting trends continue.
Whilst capacity/siting studies on wind power variability are well established in the literature [16,17,21,22,44,45], to the authors knowledge, this is the first study that attempts to compare the relative impact of changing spatial distribution to the potential impacts of climate change.The distribution benefits above largely outweigh the identified near-term climate change impacts, which show a slight reduction in mean power generation, reflecting the general wind speed stilling within future climate models [20].There are a few more notable shifts beneath this small mean change (seasonal generation patterns, extreme persistently high wind events), but these signals are mostly dwarfed by the large natural interannual wind speed variability.In fact, the decrease in summer output identified in the near future simulations (especially in the later years of the timeseries) could reflect multi-decadal variability rather than evidence of a global stilling effect [16,17].However, as the future climate data used is constructed using historical weather profiles, it is possible that future variability in the region may be significantly altered by climate change effects on the frequency and intensity of extreme wind events, especially beyond the near-term-focus within this study.Large climate model ensembles are increasingly being utilised to expand the sample of meteorological event variability, both for contemporary hindcasts [46,47], and for full-bias-corrected future climate projections [48][49][50].Further research on how climate change is likely to affect regional wind power is highly recommended, but our results clearly highlight that a future energy system must consider the impacts of future climate change-designing wind energy systems using historical climate data will likely result in a slight underestimation of installation required, which is already currently not yet at the pace required to meet UK netzero targets [51].
This study can help to inform the UK's future offshore wind fleet design, both for optimising the resilience and operation of the future net-zero energy system, and for guiding efficient investment decisions for wind farm operators and investors.The studied scenario's increased penetration into previously under/un-utilised regions beyond the historically established eastern shipping zones provides several benefits to power generation, as these new offshore areas are typically more exposed, have higher average wind speeds and lower wind speed variability.
However, this study does not consider many wider extrinsic energy system and geographic factors of these regions that are relevant for grid management, investment decisions, and policy makers.While this study utilises a dataset comprising previouslyaggregated wind-speed time-series to ease computational constraints, there are inherent biases that arise from the choice of dataset used.The wind speed power law conversion has been shown to be deficient in representing interpolated wind speeds and shear effects [20,52], and the ERA5 reanalysis dataset is outperformed at representing hourly ramps by higher-resolution models [53,54].The spatial aggregation of offshore wind speeds to the 16 spatial profiles inherently simplifies the geography of these regions, masking geophysical impacts (e.g.bathymetry, wind boundary layer, coastal, and wake effects), and ignores any ecology, policy, or infrastructure constraints (e.g.shipping lanes, marine protected areas, electricity grid constraint) that may restrict feasible wind farm installation in these regions.Timely connection of large new generation assets is particularly reliant on the local status of the transmission system, which is likely to be a significant constraint under a more distributed scenario with significant volumes of new wind connections in coastal regions without sufficient network capability.These regions might also be particularly rural, with inadequate port infrastructure or community support for development.Wake effects within and between wind farms in particular can drastically affect generation performance both spatially and temporally, with Borgers et al [55] showing capacity factors reducing by up to a quarter as a result of increased inter-farm wake effects from higher turbine installation densities in UK waters, and Rosencrans et al [56] identifying wake lengths propagating up to 55 km downstream of large installations, resulting in similar reductions in capacity factor values in the stable summer months.Whilst these deficiencies do not invalidate the relative national-scale scenario comparisons within this study, future sitespecific distribution exercises should consider specific turbine heights, inter/intra-farm installation densities, alternate reanalysis models, and relevant constraints in the proposed wind farm region of study.
It is also suggested that future research includes further stakeholder engagement to determine and incorporate crucial siting factors for alternative scenario exploration.These might include using the datasets and scenarios from this study within levelised-cost-of-energy system models, environmental spatial constraints, consideration of electricity transmission network reinforcements, and on-shore demand capacity.

Figure 1 .
Figure 1.Proportion of total offshore wind capacity installed in each zone for the three distribution scenarios (blue shading), including location of existing (orange) and proposed (green) UK offshore wind farms.

1 .
Established-high-density, continuing to place wind farms only in regions with existing installations, proportional to the present distribution reflecting the current east-coast-centric allocation (figure 1(a), where orange shapes represent existing wind farm locations); 2. Proposed-medium-density, allocating new wind farm capacity to regions in the proportions indicated in The Crown Estate's Project Listings (figure 1(b), where green shapes show planned wind farm locations); and, 3. Distributed-low-density, maximising distribution of all wind farms equally resulting in 8750 MW (6.25% of total 140 GW) installed in each of the 16 zones (figure 1(c), where farms are spaced equally throughout the entirity of the shipping zones).

Figure 2 .
Figure 2. Wind power generation potential (capacity factor) from three distribution scenarios, under historical and future climate conditions.Top row shows timeseries of summer (April-September) and winter (October-March) mean capacity factors averaged over all shipping zones for each distribution scenario (a) and the average winter (October-March) capacity factor (ERA5 data) for each shipping zone (b).Panels (c) and (e) show the range of monthly (c) and daily (e) capacity factors for each distribution scenario and month.Panels (d) and (f) show the standard deviation of monthly (d) and daily (f) capacity factors (ERA5 data) for each shipping zone.Boxplots boxes indicate median, 25th percentile, and 75th percentile values, with whiskers extending to 1.5x the interquartile range, and outliers shown in circles.In (a), (c), (e) the distribution scenarios show ERA5 data, with future climate shown for distributed scenario comparison only, illustrated by the average of the 5 models.Individual model ranges are similar and are shown in appendix B. Shipping zone locations are shown in appendix C.
(d)) and daily (figure 2(f)) mean capacity factors in comparison to

Figure 3 .
Figure 3. Wind power generation potential (capacity factor) for the 23/12/20-31/12/20 period, from three distribution scenario, under historic climate conditions.Timeseries shows mean capacity factors averaged over all shipping zones for each distribution scenario.

Figure 4 .
Figure 4. Ramp-event frequency distribution curves from three wind farm distribution scenarios, under historical (solid) and future (dashed) climate conditions.Main panels show the number of periods per year where the change in total power generated across all shipping zones is above a particular magnitude (ramp event, measured as proportion of total installed capacity factor, ∆CF) for a period of (a), (b) 1 h, (c), (d) 6 h, and (e), (f) 24 h.Panels (a), (c), and (e) show ramp-events where total generated power is reduced (ramp down).Panels (b), (d), and (f) show ramp-events where total generated power is increased (ramp up).Inset panels show the mean annual count of ramp-events above a threshold of (a), (b) 5%, (c), (d), 20%, and (e), (f) 50% of the total installed capacity, indicated on the frequency distribution as a dashed black line.The coloured distribution scenarios show ERA5 data, with future climate shown for distributed scenario comparison only, illustrated by the average of the 5 climate models.

Figure 5 .
Figure 5. High and low persistent events from three distribution scenarios, under historic and future climate conditions.Panels (a)-(d) show frequency and length of periods where total power generated across all shipping zones is consistently above/below a particular threshold (measured as proportion of total installed capacity factor CF). Frequency distribution curves for low persistence events are shown in panels (a) (periods where CF < 5% total) and (b), (CF < 10% total), and for high persistence events in panels (c) (periods where CF > 80% total) and (d) (CF > 90% total).24 h periods are indicated with light grey vertical lines.Inset panels show the mean annual total number of hours spent in each persistent state.The coloured distribution scenarios show ERA5 data, with future climate shown for distributed scenario comparison only in panels (a)-(d), illustrated by the average of the 5 models.The maps show the hours spent in extreme persistent states per year for each of the 16 regions, with panel (e) showing extreme low persistence (<5%) and panel (f) showing extreme high persistence (>90%).