In the formal study, projections on wind speeds from the models are weighted against past data belonging to the same models in order to evaluate future changes. As a result, assessing distributional differences of the GCMs is of much more relevance than distributional biases. In this work, the performance of a GCM in the region of study is assessed using the Kolmogorov-Smirnov test (K-S test) – which is applied to ascertain whether two series come from the same distribution [36].

First, since the models provide data with different spatial resolutions, these are mapped into a common grid. In order to preserve the flux integrals, a first-order conservative remapping is the choice [37]. The destination grid is chosen to have a 1.5° × 1.5° (latitude × longitude) resolution, based on the spatial resolution of the models. Second, given the great difference in latitudes between the areas included in the work, the seasonal bias of the models is eliminated – with this step, the K-S test can better evaluate statistical differences [36].

Data on historical (2005-2014) daily-averaged wind speeds from the models are thus compared to their remapped counterparts sourced from the ERA-5 reanalysis data. The Kolmogorov-Smirnov test is applied to the remapped, unbiased GCMs and the remapped, unbiased ERA-5 data. The performance of the GCMs in North America and Europe is measured based on the percentage of mesh points which are deemed to belong to the same distribution as their ERA-5 counterparts (Table 1 and Table 2, respectively) [23].

With the validation performed in Tables 1 and 2, the models that scored the best following the criteria of the validation in the separate two regions are chosen to build the multi-model ensemble. The top five models in each region are selected, and the MME is constructed with an unweighted approach [35]. It is noteworthy that the same five GCMs were selected for the two regions: NorESM2-MM, GFDL-ESM4, GFDL-CM4, EC-Earth3 and CESM2-WACCM.

The wind power density (P), which is used hereinafter to quantify the available wind resources, is obtained following:

$P=\frac{1}{2}\rho U^3$ (1)

In eq. (1), ρ represents the air density (1.225 kg/m³) and U the wind speed. Hence, daily values of wind power density are calculated at each grid point with data on daily wind speeds sourced from the GCMs.

The MME is thus constructed with the above-mentioned five models. Mean values of wind power density are calculated for the period from 2091 to 2100 (long term) from the daily-averaged wind power density at each grid point. Subsequently, these values are compared with their historical (2005-2014) counterparts to assess changes in future wind resources.

In North America, the historical onshore and offshore wind resources show great differences (Figure 1). Offshore, regions north to the 40^th parallel show exhibit the greatest wind resources (600 W/m²). In stark contrast, the mean wind power density in continental regions does not exceed 250 W/m².

Figure 1.

Climate change effects on wind power density in North America. Historical (2005-2014) and long-term (2091-2100) changes according to wind projections following the intermediate scenario (SSP2-4.5) and the augmented emissions scenario (SSP5-8.5)

Notably, projections anticipate significant evolution in mean wind power density (Figure 1) – reaching declines of 40% and increases of 30%. The most substantial changes are associated with augmented emissions, i.e., SSP5-8.5, where a general reduction of approx. 15% in wind power density is predicted across the North American continent. The greatest drops in this scenario, however, are predicted in specific regions of Canada (in the provinces of Quebec and Nunavut) and the US (in Alaska). There are some regional exceptions to this decreasing trend, with growth in certain areas: parts of northeastern Mexico and Texas (over 10%), Hudson Bay (~20%) and parts of Central America (Belize, Guatemala and Honduras, ~30%).

In SSP2-4.5 (Figure 1), a general decrease in wind resources is also anticipated, albeit less pronounced than in SSP5-8.5. South of the Great Lakes is an area with a particularly strong decrease. In this scenario the main regional exception to the general decrease, where growth is anticipated, is Texas – again, less pronounced than in SSP5-8.5. Interestingly, the few other areas with increasing resources in SSP5-8.5 exhibit very muted growth (Hudson Bay) or no growth (Belize, Honduras) in SSP2-4.5.

In Europe, the greatest values of historical mean wind power density occur offshore around the 55^th parallel, off west Ireland and south of Iceland (>700 W/m²), as a result of the prevailing westerlies (Figure 2) [14]. Onshore and in the Mediterranean Sea, values rarely surpass 250 W/m².

Figure 2.

Climate change effects on wind power density in Europe. Historical (2005-2014) and long-term (2091-2100) changes in the intermediate scenario (SSP2-4.5) and the augmented emissions scenario (SSP5-8.5)

Remarkably, the projections in the high-emissions scenario (Figure 2) include a substantial drop in average wind power densities (approximately 15%) over Europe. The most substantial reductions (between 20% and 35%) are predicted in Ireland, Great Britain, Northern Scandinavia and Lapland. In addition, a focus of significant decreases (up to 25%) is found in the Mediterranean Sea and, in particular, in its central regions, including the Ionian Sea [60]. By contrast, substantial growth is anticipated in SW Finland (30-35%).

In SSP2-4.5, spatial variations are more marked (Figure 2). A significant drop in wind energy (10-30%) is anticipated in Ireland, Great Britain, Northern Scandinavia and Lapland, and a substantial increase in SW Finland – similar to, albeit less pronounced, SSP5-8.5. The discrepancies arise in Central and Western Europe, mainly in Germany and France: SSP2-4.5 anticipates slight rises in wind resources (15%), whereas SSP5-8.5 predicts slight reductions. It is interesting to note that previous works have reported significantly different trends in wind resources in these areas depending on the climate-change scenario considered [61], which may indicate a particular sensitivity to the scenarios [23].

Importantly, the results highlight that models based on the latest climate scenarios anticipate bigger changes than models participating in previous phases of the CMIP [38]. In North America [62], only slight changes (~5%) were anticipated in the US, and no significant effects [22]. In another study in the USA [63], a widespread decrease in wind speeds of 5% was predicted in the East Coast, while small increases were anticipated in the central regions of the West Coast. In Europe [59], changes anticipated by the GCMs in the CMIP5 [23], the previous phase of the CMIP, are also of lesser magnitude than those depicted in Figure 2.

The year is divided into DJF (from December to February), MAM (from March to May), JJA (from June to August) and SON (from September to November). Wind power density data from wind projections in North America (Figure 4 and Figure 5) and Europe (Figure 7 and Figure 8) in all four periods are weighted against historical data from the same MMEs (Figure 3 and Figure 6, respectively).

Historical data show the largest values of mean wind power density take place in SON and, particularly, DJF (surpassing 900 W/m²), corresponding with the occurrence of extratropical cyclones in the Northern Hemisphere, in both North America and Europe (Figure 3 and Figure 6, respectively). Conversely, the lowest values occur in JJA, in summer in the Northern Hemisphere.

The intra-annual variability is evaluated according to the following rationale. If wind resources increase in the seasons with already the greatest values of wind power density, a greater intra-annual variability will result – and similarly if the resources drop in seasons with the lowest values. On the contrary, wind resources dropping in seasons with the greatest resources would result in reduced intra-annual variability – and similarly if the resources grow in seasons with the lowest values. As a result, based on the historical wind resources, increases/decreases in wind projections in DJF/JJA result in an augmented intra-annual variability.

Figure 3.

North America, historical (2005-2014) seasonal wind power density

In North America, changes anticipated by wind projections present a high seasonal component. In SSP2-4.5 (Figure 4), a notable increase (reaching 40%) is anticipated over Hudson Bay during DJF and, to a lesser degree, in MAM, whereas negligible changes are projected during the rest of the year. Another increase is predicted in wide areas in the Pacific Ocean (10%), between the 30^th and 50^th parallels. Being DJF the season with the most energetic winds, a growth in intra-annual variability may be expected in these regions.

Figure 4.

North America, long-term (2091-2100) evolution of seasonal wind power density. Climate change scenario SSP2-4.5

Figure 5.

North America, long-term (2091-2100) evolution of seasonal wind power density. Climate change scenario SSP5-8.5

Changes in this scenario that are consistent throughout the year include the general decline anticipated in the US mainland and Canada – in the regions of Quebec and Nunavut (40% drop). Regardless of the season, a growth in wind power density is predicted in northern Mexico and Texas.

Projections according to SSP5-8.5 (Figure 5) anticipate changes with a much more pronounced seasonal character. Notably, a widespread decrease in wind resources (over 50%) is predicted in JJA in the US and Canada, particularly prominent, again, in the provinces of Quebec and Nunavut. In the same period, remarkable growth (surpassing 60%) is predicted to occur in Central America. In JJA and SON, important decreases are predicted across the Pacific, including Hawaii. Finally, the strong growth of wind resources predicted in Hudson Bay (Figure 1) is concentrated in the DJF and MAM periods.

In Europe, the seasonality of changes is less marked than in North America. However, the SSP2-4.5 anticipates increases (10-15%) in the Iberian Peninsula in DJF and JJA, but decreases in MAM and SON (Figure 7). Seasonality can also be found in the Central Mediterranean, with important decreases (25%) during JJA and SON, but negligible changes in other periods.

Nonetheless, changes present projected during the whole year are predicted in multiple regions in SSP2-4.5. This is the case, most notably, of West Finland (growth >40%), Central and Western Europe (20% increases) and Ireland, the UK and the northernmost parts of Norway and Finland (decreases in the range of 10-30%).

Similarly to North America, SSP5-8.5 anticipates in Europe much stronger seasonal changes than SSP2-4.5 (Figure 8). The scenario with augmented emissions predicts decreases in wind power density in latitudes in the range of 45°N and 60°N in the Atlantic Ocean and Central Europe in JJA (up to 35%). This remarkable increase, however, is of a much lesser magnitude for the rest of the year. In stark contrast, the Iberian and Balkan Peninsulas present the opposite trend – a widespread growth in wind resources exclusively in JJA of approx. 15%. Changes that appear during the entire year are exclusively located in Central Mediterranean (35% drop) and SW Finland (40% increase).

Figure 6.

Europe, historical (2005-2014) seasonal wind power density

The results reveal a clear picture of an overall drop in wind resources across North America and Europe, two of the primary markets for wind energy. This decline aligns with the generalised drop in wind energy across the Northern Hemisphere as anticipated by previous climate scenarios [64]. This phenomenon may be attributed to polar amplification induced by climate change, which might result in diminished temperature gradients between the Polar Regions and mid-latitudes of the Northern Hemisphere. Nonetheless, anticipated changes are greatly amplified under the novel climate scenarios.

Despite this general trend, there is no doubt that the impact on wind resources differs notably between North America and Europe, and between regions in each continent. These differences have implications for the wind energy industry, given the bearing of wind energy resources on the LCOE. Therefore, future projects should consider the regional tendencies in order for the great benefits of wind energy [65] to be realised. Regional downscaling of climate projections could shed light on the local dynamics that contribute to the spatial variability of the resource.

Importantly, a continuous assessment of wind resources is crucial to determining whether their evolution aligns with climate projections [66]. The results indicate changes between past-present conditions (2005-2014) and the long-term future (2091-2100); it is noteworthy that this timescale (approx. 85 years) is considerable compared to the average lifetime of an energy project. Hence, a continuous assessment of wind resources may facilitate the adaptation of the wind energy industry to the evolving nature of these resources.

Figure 7.

Europe, long-term (2091-2100) evolution of seasonal wind power density. Climate change scenario SSP2-4.5