The use of wind energy resource is an integral part of many nations’ strategies towards realizing the carbon emissions reduction targets set forth in the Paris Agreement, and global installed wind power cumulative capacity has grown on average by 22% per year since 2006. However, assessments of wind energy resource are usually based on today’s climate, rather than taking into account that anthropogenic greenhouse gas emissions continue to modify the global atmospheric circulation. Here, we apply an industry wind turbine power curve to simulations of high and low future emissions scenarios in an ensemble of ten fully coupled global climate models to investigate large-scale changes in wind power across the globe. Our calculations reveal decreases in wind power across the Northern Hemisphere mid-latitudes and increases across the tropics and Southern Hemisphere, with substantial regional variations. The changes across the northern mid-latitudes are robust responses over time in both emissions scenarios, whereas the Southern Hemisphere changes appear critically sensitive to each individual emissions scenario. In addition, we find that established features of climate change can explain these patterns: polar amplification is implicated in the northern mid-latitude decrease in wind power, and enhanced land–sea thermal gradients account for the tropical and southern subtropical increases.
While the rapid growth of wind power capacity across the world1 constitutes a component of a broader strategy to reduce the reliance on fossil fuels and mitigate future climate change2, this deployment occurs in the context of today’s climate or that of the very recent past, subject to interannual variations. Specifically, wind farms are being sited, designed and built under the assumption that the amount of power that can be drawn from the wind today will remain approximately constant into the future, within the range of existing interannual variability3. However, climate, including the global atmospheric circulation, is already changing in response to anthropogenic radiative forcing4 and is projected to continue doing so in the coming decades5. Therefore, the economic and engineering calculus of wind resource management should consider projected changes in surface climate.
Several studies have considered evolution in wind energy resources, generally emphasizing local and regional scales. Observational records6,7,8,9,10,11,12,13,14 and reanalysis datasets12,15,16,17 have assessed past variability, in some locations finding relationships with internal modes of climate variability such as El Niño–Southern Oscillation (ENSO)9,10,11,14,15,16 while other locations are more sensitive to changes in surface roughness12,13 or broader pressure gradients14. Considering future developments, global climate model (GCM) simulations have been analysed to understand impacts on wind resources in North America18,19, Europe20, in China21 on the East Indian Coast22 or to explore wind extremes23. Other studies employ dynamic downscaling of GCM results to focus on particular regions24,25,26,27,28. Some researchers emphasize the need to consider multi-model ensembles due to uncertainties in predictions of interannual variability20. Such analyses are critically important at local and regional levels, but together are not sufficient to provide an overall assessment of how climate change will impact the global wind power capacity and thus the extent to which wind energy production can offset fossil fuel combustion to meet the world’s growing energy demand. To date, a global assessment of likely future changes in the wind energy resource has not been made. This study utilizes an ensemble of ten state-of-the-art GCMs from the Coupled Model Intercomparison Project phase 5 (CMIP5)29 to predict the large-scale geographic distribution of future changes in wind energy resource under two future greenhouse gas emissions scenarios (see Methods) and understand the physical mechanisms for such changes.
Using global climate models to assess wind energy
Three characteristics of GCMs render them imperfect tools for assessing wind energy potential; we demonstrate quantitatively that none of them are sufficiently severe to undermine the scientific validity or societal value of such projections derived therefrom. The first and perhaps the most obvious potential limitation of GCMs is their horizontal resolution. The resolution of the atmospheric component of CMIP5 GCMs varies by model and is non-uniform in space, but is typically on the order of 1–2° or ~100 km. To demonstrate that the magnitude and variability of actual monthly total wind power can be reproduced by coarse-resolution gridded data, we compare the power calculated using wind measurements from a tower located at the US National Renewable Energy Laboratory’s (NREL) National Wind Technology Center (NWTC) in Boulder, CO, USA (Tower M2; 105.23° W, 39.90° N)30 to power calculated using winds estimated by the 2.5°-resolution NCEP/NCAR Reanalysis31. The NWTC Tower M2 is located approximately 4 km from mountains and other complex terrain, and flow there is influenced by the complex terrain32,33. We use a reanalysis product, rather than a CMIP5 GCM, to compare with tower measurements because we are comparing time series (that is, actual, historical monthly to interannual variability) of wind estimates, whereas fully coupled GCMs are not temporally synchronous with the real world at such frequencies. Furthermore, the two datasets being compared are independent because wind measurements made by the NWTC towers are not assimilated into the NCEP/NCAR Reanalysis. Despite their coarse resolution (and averaging two adjacent 2.5° × 2.5° grid cells), the gridded data are able to reproduce (r = 0.94) the magnitude and variability of monthly total power inferred from the point measurements made on the tower (Fig. 1a). By proxy, we conclude that the coarse horizontal resolution of the GCMs does not undermine their utility for assessing wind resource variability.
The second potential limitation of GCMs in the context of wind energy is the fact that wind turbines integrate momentum from wind across the rotor disk34, typically 40–120 m above the surface, whereas the vertical grid structure of GCMs is such that wind information is available only at a height of 10 m and on standard pressure levels (for exmple, 1,000 mbar, 925 mbar, 850 mbar, and so on). To assess the uncertainty introduced by this vertical offset, we first calculate monthly total power based on wind speed measured at a height of 80 m on the M2 tower, and compare that estimate with that estimated using 10-m tower measurements of wind speed extrapolated upward to 80 m using a power law. Although instantaneous wind profiles rarely follow the power law with the same coefficient throughout the diurnal cycle35,36, the power law applies on average, and the monthly total power at 80 m is predicted extremely well (r = 0.98) by the extrapolated 10-m winds (Fig. 1b).
The third potential limitation of GCMs is the provision of monthly mean fields, whereas winds fluctuate at much higher frequencies. To quantify the possible effect of averaging out the high-frequency wind fluctuations, we compare time series of monthly total power based on high-frequency (hourly) wind speed measurements (from the M2 tower) with that based on the same hourly wind speed measurements but averaged across the calendar month prior to computing power. In this case, a systematic bias occurs, such that using monthly averaged wind speed results in an underestimation of monthly total power by 114 kW (48% of average production) on average (Fig. 1c). However, the temporal variability of this offset is relatively small (σ = 43 kW) and the temporal correlation remains very high (r = 0.94). A broader perspective on the potential impact of this approximation is provided by further comparing the global wind power climatology obtained using 6-hourly and monthly mean wind speed fields, both from reanalysis (Supplementary Figs. 3 and 4). Offsets are similar in magnitude and stability to that estimated from the tower measurements, thus temporal correlations between monthly power computed with and without explicitly resolving the high-frequency wind speed variations are extremely high across the globe.
The offset due to using monthly averaged wind speed dominates the exercise wherein all three limitations are considered simultaneously (Fig. 1d) and thus should be borne in mind as a persistent low bias when interpreting absolute wind power values derived from monthly mean GCM output (or any monthly mean data, including measurements). The correlation between monthly total power derived from hourly wind measurements at 80 m on the M2 tower (ideal case) versus monthly mean wind estimates at 10 m from the gridded reanalysis data (most equivalent to a GCM) is 0.92 with a mean offset of 93 ± 44 kW (~39% of average production).
Global wind power projections from CMIP5 simulations
The global distribution of present-day annual mean wind power is well reproduced by the CMIP5 multi-model median of historical simulations (Fig. 2) including the correct location of maxima within the American continents (central US and Patagonia), Africa (Sahara, horn and southern tip), Europe (northwestern and far eastern), Asia (Tibetan Plateau and Inner Mongolia), and Australasia (southwestern Australia and New Zealand)37,38. The remainder of this Article focuses on how and why this distribution is projected to change in response to future greenhouse gas forcing.
Projected changes in wind power over the coming century reveal strong interhemispheric asymmetry (Fig. 3 and Supplementary Fig. 6). Specifically, reductions in wind power are generally projected broadly across northern mid-latitude regions, while increases are projected across tropical and southern subtropical regions. Particularly notable and robust (in terms of inter-model agreement) reductions are found over the US, the British Isles, northern Middle East, central and northern Asia, and far eastern Asia including the Korean Peninsula and Japan. Increases are found throughout Mexico and Central America, eastern Brazil, southeastern Africa including Madagascar, the southeastern Arabian Peninsula including Yemen and Oman, India, Southeast Asia, and northeastern Australia. Most of the aforementioned increases are only robust under RCP8.5 and for later time periods; see below for further discussion of dependence on emissions and interhemispheric asymmetry.
Changes in wind power vary by region. In the central US, the area-averaged power is projected to decline by 8% (14%) by 2050 (2100) under the RCP4.5 scenario or by 10% (18%) by 2050 (2100) under the RCP8.5 scenario (Fig. 4). An extremely large increase in wind power is projected for eastern Brazil under the RCP8.5 scenario—21% (42%) by 2050 (2100), but those increases are only a quarter as large under the RCP4.5 scenario. Similarly, northeastern Australia is also projected to experience an extremely large increase in wind power by 2100 (41%), but only under the RCP8.5 scenario. Several other key regions around the globe have substantial projected changes in wind energy resource, as shown in Fig. 4. Beyond the robustness of projections in terms of inter-model agreement, the significance of the future changes can be assessed relative to the amplitude of local internal variability (Supplementary Fig. 5). At the near-term time horizon (2020–2040), most of the robust changes are between 50–100% of the baseline variability, with the exception of eastern Europe and northern Asia, where on average the projected changes exceed one standard deviation of the simulated interannual variability. Near the end of the century, nearly all projected changes due to anthropogenic forcing are well in excess of the level of internal variability—in many cases several times greater than baseline interannual variability.
Some of the projected regional changes in wind power noted above have a strong seasonal dependence, while others are manifest throughout the year (Supplementary Fig. 7). For example, the projected decreases in the central US, Japan and the horn of Africa occur primarily during boreal winter, the projected increases in eastern Brazil, Madagascar and the southern Arabian Peninsula are much stronger during austral winter, and the projected increases in southern mainland Africa and Southeast Asia are relatively constant throughout the year.
When integrated zonally (along lines of latitude; Fig. 3), a clear and asymmetric dependence on emissions scenario emerges. The total decrease in power across the northern mid-latitudes (30–65° N) is largely insensitive to emissions scenario. In contrast, the total increase across the southern tropical/subtropical band (10–30° S) only emerges under the RCP8.5 scenario. Thus, in a globally integrated sense, the decrease in the Northern Hemisphere is partially balanced by an increase in the Southern Hemisphere, but only under the high greenhouse gas emissions scenario.
The regional changes in continental wind power described above are linked to large-scale dynamical and thermodynamic mechanisms governing the overall response of the atmospheric general circulation to radiatively forced changes in surface temperature (Fig. 5). However, the mechanisms explaining the changes in the two hemispheres are distinct. The relatively uniform decrease in power across the Northern Hemisphere mid-latitudes on land is consistent with a general reduction in surface wind speed in that latitude band including the ocean (Fig. 5a,b). This general reduction is caused by a strongly reduced meridional temperature gradient due to polar amplification of global warming (Fig. 5d,e), which reduces mid-latitude baroclinicity and hence storm track intensity. It is worth noting, however, that there is a relatively large inter-model spread in the projected response of Northern Hemisphere storm tracks39,40,41, which cannot be fully explained by uncertainties in the projected change in the meridional temperature gradient42. Other factors, such as tropical processes and changes in subtropical static stability may also play a role43. While the reduction of westerly surface flow is strongest over the ocean and is thus not entirely zonally symmetric, it extends readily across the continents within approximately the same latitude band without exception (note that background mean surface wind speeds are also higher over the oceans, so the fractional change in wind speed is roughly zonally symmetric). In contrast, the increase in power in the Southern Hemisphere tropics and subtropics is driven by land–sea warming gradients (or ‘land amplification’). The regions where surface wind speed increases within the latitude band 0–30° S (Fig. 5f) are not only constrained by the continents, but by their associated pressure gradient amplifications (Fig. 5g), which are clearly driven by warming amplifications (Fig. 5h) via thermal low intensification over southern Africa, Australia and northern South America
Summary and essential research needs
Results presented herein reveal strong interhemispheric asymmetry of future changes in potential wind power, including decreases across the Northern Hemisphere mid-latitudes and increases across the tropics and subtropics of the Southern Hemisphere. The global distribution of projected regional changes in surface wind speed over land can be mechanistically linked to well-documented and understood large-scale changes in surface temperature and atmospheric circulation. This clearly explicable relationship lends further confidence in the projections of wind power provided here. While the relatively coarse spatial and temporal resolution of the models on which these projections are made may not allow high-fidelity predictions of wind resource that would be required for localized wind farm siting, especially in regions of complex terrain, this investigation does highlight areas around the globe where more refined investigations with downscaling can lend critical insight into the likely impacts of climate change on wind resources. Consideration of future build-out of wind energy resources should also consider climate change impacts on offshore wind energy, as well as the impacts of wind farms on local and regional climate44,45,46,47,48.
Global climate models
This study utilizes an ensemble of ten state-of-the-art global climate models (GCMs) from the Coupled Model Intercomparison Project phase 5 (CMIP5) to calculate the large-scale geographic distribution of future changes in wind energy resource under two future greenhouse gas emissions scenarios, or Representative Concentration Pathways (RCPs). Details of the CMIP5 experimental design are given in ref. 29, and the names and relevant details of the CMIP5 GCMs used in this study are provided in Supplementary Table 1. The GCMs were selected based on if they provided all of the required output variables (listed below) for both of the RCPs (2006–2100) as well as the historical experiment (1850–2006), where the availability of 10-m wind speed was frequently a limiting factor. Nonetheless, the GCMs included in this study represent a broad cross-section of the full CMIP5 ensemble, both in terms of national origin of GCM development and various GCM characteristics such as resolution (Supplementary Table 1).
Wind power calculation
As wind power is not a standard output of any GCM, a method for an offline diagnostic calculation thereof was developed here. The calculation requires four monthly mean fields: 10-m wind speed (W 10), surface pressure (p s), surface temperature (T s) and surface specific humidity (q s). On acquiring all necessary GCM output fields for the historical, RCP4.5 and RCP8.5 simulations, all fields were linearly regridded onto a common 1° by 1° grid and an ocean mask was applied.
Wind power, P, is a function of wind speed, W, at rotor height (taken here as 100 m), thus wind speed as provided at the nominal GCM output level of 10 m (W 10) must first be extrapolated to rotor height (W 100). Furthermore, wind speed must be corrected for air density (as greater density of air imparts a larger force on turbine blades), and density must be corrected for humidity (a relatively minor adjustment). All calculations and corrections are transient in both space and time, and therefore account for the full influence of climate change on wind power (for example, changes in wind speed and air density that may arise due to global warming). The calculations and corrections are made as follows.
Next, dry air density, ρ, is calculated using the ideal gas law
where R = 287.058 J kg–1 K–1. Density is then corrected for humidity50
Next, wind speed is scaled for density51
Finally, wind power is calculated using an industry wind turbine power curve
where f is a function that models the General Electric (GE) 1.5 MW sle turbine (Supplementary Fig. 8). The power curve function f is implemented numerically using the following algorithm.
The polynomial function describing the ramp-up stage is modelled by fitting a curve via least-squares linear regression to the wind speed–power relationship given in the industry table for wind speeds between cut-in and rated power. Graphically, the solution to the full algorithm is virtually indistinguishable from the values given by the industry table (Supplementary Fig. 8).
All data used in this study are freely available online. Global climate model output data are available from the Earth System Grid Federation (http://cmip-pcmdi.llnl.gov/cmip5/). Reanalysis data are available from the US National Oceanic and Atmospheric Administration (https://www.esrl.noaa.gov/psd/data/gridded/). Observations from the M2 meteorological tower in Boulder, CO, USA are available from the US National Renewable Energy Laboratory (https://midcdmz.nrel.gov/nwtc_m2/).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We acknowledge the WCRP Working Group on Coupled Modelling and US DOE/PCMDI for CMIP, and thank the climate modelling groups (listed in Supplementary Table 1) for producing and making available their model output. We also acknowledge the WHOI CMIP5 Community Storage Server, Woods Hole Oceanographic Institution, Woods Hole, MA, USA (cmip5.whoi.edu). We express appreciation to the US DOE’s National Renewable Energy Laboratory for sustained observations from the M2 meteorological tower, and to NOAA/OAR/ESRL/PSD in Boulder, CO, USA for archiving NCEP/NCAR Reanalysis fields.
Supplementary Tables and Figures