Saturation wind power potential and its

implications for wind energy

Mark Z. Jacobsona,1 and Cristina L. Archerb,1
a
Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-4020; and bCollege of Earth, Ocean, and Environment,
University of Delaware, Newark, DE 19716

Edited by Michael B. McElroy, Harvard University, Cambridge, MA, and accepted by the Editorial Board August 14, 2012 (received for review May 31, 2012)

Wind turbines convert kinetic to electrical energy, which returns It is well known that the array efficiency of a single wind or
to the atmosphere as heat to regenerate some potential and kinetic water farm containing many turbines decreases with decreasing
energy. As the number of wind turbines increases over large distance between turbines (11, 12). However, what is not known
geographic regions, power extraction first increases linearly, but is the extent to which the efficiency loss operates globally when
then converges to a saturation potential not identified previously realistic meteorology and energy extraction by turbines are
from physical principles or turbine properties. These saturation accounted for. This information is critical for determining the
potentials are >250 terawatts (TW) at 100 m globally, approxi- feasibility of a worldwide renewable energy future. Calculating
mately 80 TW at 100 m over land plus coastal ocean outside the SWPP for large penetrations of wind (≥1 TW) is not cur-
Antarctica, and approximately 380 TW at 10 km in the jet streams. rently possible from data analysis, because penetrations are still
Thus, there is no fundamental barrier to obtaining half (approxi- low (239 gigawatts (GW) installed worldwide at the start of 2012).
mately 5.75 TW) or several times the world’s all-purpose power The most accurate method available to analyze this issue is with a
from wind in a 2030 clean-energy economy. complex 3D atmospheric-ocean-land coupled model.

APPLIED PHYSICAL
Previous global simulations of wind farms have assumed that

SCIENCES
atmospheric modeling ∣ climate feedbacks ∣ renewable energy ∣ wind farm effects on the atmosphere can be represented by chan-
water vapor ∣ clean energy economy ging surface roughness or adding a drag coefficient (2, 13–17).
Roughness parameterizations, though, incorrectly reduce wind

A new method is proposed to determine the maximum theo-

retical wind power potential on Earth, based on the concept
of “saturation”. The saturation wind power potential (SWPP) is the
speeds the most in the bottom model layer, whereas in reality
a surface wind turbine reduces wind speed the most at hub height,
approximately 100 m above ground (Fig. 1). Because roughness
maximum wind power that can be extracted upon increasing the lengths and drag coefficients are approximate, it is also difficult to
number of wind turbines over a large geographic region, indepen- ensure they extract the correct amount of energy from the wind.
dent of societal, environmental, climatic, or economic considera- Calaf, et al. (18) demonstrated the inaccuracy of standard rough-
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

tions. Once the SWPP is reached, increasing the number of ness parameterizations against large-eddy simulation results
turbines further does not increase the generated power further. and developed a multiple layer roughness parameterization for
At the SWPP, winds still occur because individual turbines can ground-based turbines that is more accurate. Here, however, we
extract no more than 59.3 % of the kinetic energy in the wind use a straightforward approach to calculate the momentum sink
(Betz’s limit). This paper also defines the fixed wind power poten- at any specified hub height, not just near the ground, but also
tial (FWPP), which is the maximum power that can be extracted aloft, each time step, because it precisely determines the time-
by a fixed number of wind turbines at decreasing installed density dependent energy extraction from one or many turbines.
and increasing geographic area. The SWPP is calculated here at Another common omission during modeling has been that of
100 m above ground, the hub height of most modern wind tur- energy conservation during electric power use and turbulence dis-
bines, assuming conventional wind turbines distributed every- sipation. If wind turbines generate 5.75 TW (0.0113 W∕m 2 ), such
where on Earth, including over the oceans (simulation named power ultimately returns to the air as heat following electricity
“global-SWPP”) and, separately, over land only but excluding use. This heat does not depend on the electricity source, thus
Antarctica (“land-SWPP”). The SWPP is also calculated at 10 km it is also released when coal, nuclear, and natural gas produce
above ground in the jet streams assuming airborne wind energy electricity. Such generators, though, produce additional heat due
devices (“jet stream-SWPP”). Capturing jet stream winds pre- to combustion or nuclear reaction and emit global warming
sents greater technological challenges than capturing surface pollutants (10, 19). As such, wind turbines reduce direct heat
winds but is still of interest (1, 2). and pollutant emissions compared with conventional generators.
The main purpose of these simulations is to use a physical However, the electricity use still needs to be accounted for be-
model to determine the theoretical limit of wind energy available cause the heat is a source of some regenerated kinetic energy
at these altitudes, particularly because some recent studies that (via conversion of internal energy to some available potential
accounted for energy extraction by turbines, but not physically, energy to kinetic energy). To date, only ref. 1 has calculated the
have suggested that available wind energy is small (2, 3). Previous heat from electricity returned to the air, but they focused on air-
theoretical estimates of the power in the wind (4–9) are similarly borne rather than ground-based wind turbines.
not based on a physical model of energy extraction so cannot give
estimates of wind potential at the height of turbines. As found
Author contributions: M.Z.J. and C.L.A. designed research; M.Z.J. and C.L.A. performed
here, energy extraction at a given altitude does not deplete energy
research; M.Z.J. contributed new reagents/analytic tools; M.Z.J. and C.L.A. analyzed data;
at all altitudes above or below it; so an estimate of wind potential and M.Z.J. and C.L.A. wrote the paper.
in the whole atmosphere does not answer the practical question The authors declare no conflict of interest.
about wind turbine potential at typical hub heights. This article is a PNAS Direct Submission. M.B.M. is a guest editor invited by the Editorial
More relevant for practical applications, the FWPP of four mil- Board.
lion turbines at 100 m in three different configurations is quan- 1
To whom correspondence may be addressed. E-mail: [email protected] or
tified here to determine if this number is sufficient for powering [email protected].
half the world’s all-purpose power demand in a 2030 clean-energy This article contains supporting information online at www.pnas.org/lookup/suppl/
economy (10). doi:10.1073/pnas.1208993109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1208993109 PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15679–15684

 A B C

Fig. 1. Comparison of heights and magnitudes of globally averaged percent wind speed reduction averaged over one year from (A) the turbine momentum
sink parameterization presented here vs. (B) the Lettau (33) surface roughness parameterizations. In both cases, the world is covered with either 1.146 billion
(blue, world), 324.5 million (green, land), or four million (red, 50% of 2030 power demand) 5-MW, 126-m rotor diameter turbines with hub height of 100 m,
spaced 0.44 km 2 each. A jet stream case is also shown in (A) for 931 million (black) 5-MW turbines with hub height at 10 km. (C) Percent difference in water
vapor mass mixing ratio vertical profiles between the world (Simulation B), land (I), and jet stream (Q) cases and the base case (A). All these simulations were run
at 4° × 5° horizontal resolution. Numbers in parentheses are installed wind power in each case, from Table 1.

Materials and Methods comparable with (C), were then run to scale the coarse-resolution results
Here, the GATOR-GCMOM global model (20, 21) is used to examine satura- to finer resolution.
tion and fixed wind power potentials. The model is modified to treat wind To determine the land-SWPP (land excluding Antarctica) at 100 m, four
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

turbines as an elevated momentum sink, where the kinetic energy extracted 4°x5° simulations (I–L), each with decreasing power density, were run. A
from the wind is determined from a turbine power curve at the instanta- 1.5° × 1.5° simulation (M) was then run to scale results to finer resolution.
neous model wind speed. The treatment of turbines developed is concep- To determine the FWPP of four million 5-MW turbines, the number esti-
tually similar to that in (22, 23) but differs as follows: (i) it assumes each mated to supply half the world’s all-purpose power in a clean energy econ-
wind turbine occupies multiple vertical atmospheric layers rather than one omy in 2030 (10), the turbines were distributed in three configurations: over
layer, (ii) it is applied to numerous wind farms worldwide simultaneously all land 15S–60S and 15N–66.56N, and below 3 km altitude (Simulation N);
over eight land and coastal sites (Table 1, footnote) (O); and over three land
rather than one local farm, (iii) it is applied in a global model where momen-
sites (Table 1, footnote) (P).
tum extraction feeds back to global dynamics rather than a limited-area
Finally, to determine the SWPP of the jet streams (10–70N and 10–70S) at
model with only regional feedbacks, and (iv) it accounts for energy conserva-
10 km, a 4° × 5° simulation (Q) with the maximum power density as in simu-
tion due to both electricity use and turbulent dissipation of kinetic energy. In
lation (B) was run. A 1.5° × 1.5° simulation (R) was also run to scale results
addition, the new treatment allows distributed wind turbines in a grid cell to
with resolution.
extract energy from four points on a staggered Arakawa C grid, thereby
All simulations included 68 vertical sigma-pressure layers up to 0.219 hPa
impacting five cells simultaneously, rather than from the center of the cell, (≈60 km), including 15 layers from 0–1 km and 500-m resolution from
affecting only that cell. 1–21 km. The center of the lowest model layer was 15 m above ground.
The SI Materials and Methods describes the model treatment of wind tur- The rotor of each surface turbine (simulations B–P) intersected five model
bine kinetic energy extraction. Briefly, each turbine is characterized by a layers. That of each jet-stream turbine (simulations Q–R) intersected two
rated power [P t , 5 megawatts (MW)], rotor diameter (D, 126 m), hub height layers. The model was run forward from January 1, 2006 with no data assim-
above the topographical surface (H, 100 m or 10 km), and characteristic ilation. Because this study does not focus on temperature response and due
spacing area (At , m 2 ) each simulation. Each turbine is assumed to intersect to the long computer time required for radiative, cloud, aerosol, and gas
multiple atmospheric layers of a grid column (Fig. S1). Each time step, kinetic processes, only five-year simulations were run. Wind power extraction in all
energy is extracted from each model layer that intersects the turbine rotor. five years was similar and convergent in all simulations.
The kinetic energy reduction is translated into a wind speed reduction. The
resulting shear produces turbulence that is combined with ambient mechan- Results
ical and thermal turbulence. Energy is conserved by converting all electric Fig. 2A shows that, up to about 715 TW (1.4 W∕m 2 ) of installed
power generated by the wind turbines to heat via electricity use at the sur- power, the output from power-extracting wind turbines first in-
face, where it occurs, and by converting kinetic energy lost by natural surface creases linearly. The linearity is demonstrated by comparing the
roughness to turbulence, then heat. initial slope of the “global-SWPP curve” (with power extraction)
Table 1 summarizes the simulations. A control simulation (A) was first run
with the slope of the “global-no power extraction” line. The latter
with turbines at 100 m hub height but no momentum extraction from them.
is the line between zero and the power output from Simulation A,
In this case, the global capacity factor was about 31% based on instantaneous
modeled wind speeds applied to the power curve for a 5-MW turbine with
which is the reference case with turbines but without power
126-m rotor diameter. extraction. At higher penetrations, power output increases with
Fig. S2 compares resulting near-surface wind speeds with data. To deter- diminishing returns until it reaches global saturation (approxi-
mine the global-SWPP at 100 m, five 4° × 5° horizontal resolution sensitivity mately 253 TW, also Fig. S3B for coarse-resolution results) at
simulations (B–F) with momentum extraction, each with decreasing installed about 2,870 TW (5.65 W∕m 2 ) installed. Higher penetrations
power density, were run. Simulations (G) (2.5° × 2.5°) and (H) (1.5° × 1.5°), of wind serve no additional benefit. Thus, for the first 715 TW

15680 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1208993109 Jacobson and Archer

 Table 1. Summary of simulations and power output in year 5
Model Total Number of Total Turbine Annual total
horizontal Turbine number of turbines installed Percent of installed power output
resolution spacing 5-MW turbines over ocean power world for power density year 5
Simulation (degrees) At (m 2 ) (millions) (millions) (TW) spacing (W∕m 2 ) (TW)
Global-SWPP
A) World (control) 4×5 28D 2 1,146 821 5,730 100 11.3 1,750
B) World 4×5 28D 2 1,146 821 5,730 100 11.3 224
C) World 4×5 56D 2 573 411 2,864 100 5.62 228
D) World 4×5 84D 2 382 274 1,909 100 3.75 219
E) World 4×5 112D 2 286 205 1,432 100 2.81 206
F) World 4×5 224D 2 143 103 716 100 1.41 160
G) World 2 × 2.5 56D 2 574 410 2,870 100 5.65 251*
H) World 1.5 × 1.5 56D 2 575 410 2,872 100 5.65 253*
Land-SWPP†
I) Land 4×5 28D 2 299 0 1,495 26.0 11.3 71.2
J) Land 4×5 56D 2 149 0 747.6 26.0 5.62 66.7
K) Land 4×5 112D 2 74.8 0 373.8 26.0 2.81 56.4
L) Land 4×5 224D 2 37.4 0 186.9 26.0 1.41 39.7
M) Land 1.5 × 1.5 28D 2 302 0 1,510 26.3 11.3 72.0*
Land-FWPP
N) Land‡ 4×5 1;470D 2 4 0 20 18.3 0.21 7.50
O) Land+coast 8 sites§ 4×5 56D 2 4 0.004 20.195 0.696 5.62 3.93
P) Land 3 sites¶ 4×5 28D 2 4 0 20.105 0.348 11.3 1.63
Jet stream-SWPP

APPLIED PHYSICAL
Q)Jet stream∥ 4×5 28D 2 931 668 4,653 81.0 11.3 375

SCIENCES
R) Jet stream∥ 1.5 × 1.5 28D 2 941 673 4,707 81.9 11.3 378*
Earth’s surface area is 510.6 million km 2 . Hub heights were 100 m above ground level except in the jet stream cases (10 km). D ¼ 126 m is turbine rotor
diameter.
*The 1.5° × 1.5° simulations were run for only six months and the 2° × 2.5° simulations, for one year, due to their enormous computing requirements. The
ratio of the power generation averaged over the months or year to that from the same time for the corresponding 4° × 5° resolution simulation was
multiplied by the last-year annual-average result from the 4° × 5° simulation to estimate the 1.5° × 1.5° and 2° × 2.5° power generation averaged over the
last year for conditions from that simulation.
†
Land in these cases included all land north of 60S (outside of Antarctica) but did not include coastal ocean.
‡
Land in this case included all land 15S–60S and 15N–66.56083N (Arctic Circle), and below 3 km altitude but did not include coastal ocean.
§
All turbines in this case were distributed among 19 windy cells in 8 locations: the Great Plains (4 cells), offshore East Coast (3), the North Sea (2), the Sahara
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

Desert (3), the Gobi Desert (2), the Yellow Sea (1), Australia (2), and Patagonia (2).
¶
All turbines in this case were distributed among nine windy cells in three locations: the Great Plains (4 cells), the Sahara Desert (3), and the Gobi Desert (2).
∥
The jet stream winds considered were 10S–70S and 10N–70N, and at 10 km altitude. Top and bottom turbine heights were 10.063 km and 9.937 km,
respectively.

installed, output increases roughly linearly proportionally to (Simulation G), which were about 10% higher than at 4° × 5°
turbine installation, but thereafter, it increases with diminishing (Simulation C). Even higher resolution may increase power out-
returns until saturation. put more (24).
The global SWPP obtained in Fig. 2A is likely a lower bound The 253 TW of worldwide extractable power at 100 m is
because power extraction increases, while still converging, upon approximately 22.5 times that of a previous simple-equation
grid refinement. For example, 1.5° × 1.5° resolution results estimate of extractable power in the bottom 200 m of the atmo-
(Simulation H) were about 1% higher than those at 2° × 2.5° sphere worldwide (11.25 TW) calculated by ref. 3 as the estimated

A B

Fig. 2. (A) Convergence to global-SWPP (Simulations B–F) and land-SWPP outside Antarctica (Simulations I–L and N) scaled by higher-resolution results from
Simulations H and M, respectively. As such, values represent 4° × 5° results from Table 1 scaled by 1.5° × 1.5° results, from the table. Also shown is the straight
line between 0 and 1,750 TW power output (at 5,730 TW, or 11.3 W∕m 2, installed power) from the global turbine, no-power-extraction case (Simulation A).
The highest installed turbine density in the land-SWPP case was also 11.3 W∕m 2 . (B) Wind power potential at three installed power densities of four million
wind turbines. Also shown is 50% of the world all-purpose power demand in 2030 upon conversion to wind, water, and sunlight (WWS) and electricity/
electrolytic hydrogen.

Jacobson and Archer PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15681
 total power in that region (100 TW) multiplied by the fraction hydrogen (10). Fig. 2B shows that the power output of four
that could interact with wind turbine rotors (<0.3), the fraction million turbines increases with decreasing wind turbine spacing.
in the range of turbine cut-in and cut-out speeds (0.75), and the When turbines are packed at an installed density of 11.3 W∕m 2
fraction converted from kinetic to electrical energy (0.5). These into three sites worldwide, the power output is too low (approxi-
factors were all accounted for in time and space in the simulations mately 1.6 TW—Table 1 and Fig. S3F) to match power demand. At
here. The large difference highlights the importance of using eight locations (5.6 W∕m 2 installed), the output improves to
physical calculations. approximately 4 TW (Fig. S3E) but is still lower than needed.
The SWPP over land outside Antarctica here was approxi- However, when turbines are spread over land outside the tropics,
mately 72 TW (Fig. 2A and Fig. S3C). Based on the high-resolu- away from the poles, and in all regions below 3 km altitude
tion global-SWPP calculations here, another approximately 8 TW (0.11 W∕m 2 installed), the output jumps to approximately 7.5 TW
was available offshore at depths <200 m, giving a land plus coast- (Fig. S3D), much more than needed. The crossover point is at
al SWPP estimate of 80 TW. Like with the global case, the land- an installed density of approximately 2.9 W∕m 2 . It is not necessary
SWPP curve (Fig. 2A) shows a linear portion at low turbine to spread turbines evenly across such land. In fact, individual farms
penetrations. Beyond approximately 185 TW of installed power, can have installed densities of 5.6–11.3 W∕m 2 so long as reason-
diminishing returns set in. However, the full land-SWPP was not able spreading between farms occurs and the average installed
obtained until approximately 1,500 TW (11.3 W∕m 2 ) of installed density within and between farms is ≤2.9 W∕m 2 (or ≤3.1 W∕m 2
power. The result here suggests that bottom-up approaches for accounting for higher model resolution).
calculating wind power potentials over land are justified for
<185 TW installed power. Discussion
The land-SWPP is not much lower than the 125 TW of onshore It is well known that spreading wind turbines in a farm increases
power from a study (25) that assumed a fixed percentage energy farm array efficiency by decreasing interference of one turbine
loss due to turbine interference but not increasing competition with the next (11, 12). The results here suggest that staggering
for wind with increasing turbine penetration. Results from (25) farms themselves, geographically, improves the overall power
fall near the linear “global-no extraction” curve in Fig. 2A, just output. In other words, the power potential of a fixed number of
above the land-SWPP. turbines (FWPP) increases with increased spreading of farms.
Another study (26) estimated the world land plus coastal wind The addition of surface wind turbines reduced horizontal
potential based on world sounding and surface data as 72 TW. wind speeds in their wake the most and below and above the
Similarly, ref. 25 estimated a land potential of 78 TW for capacity wake centerline to a lesser extent (Fig. 1A). The reduction in
factors of 20% or higher. Both studies accounted only for loca- wake wind speed reduced shearing stress below and increased it
tions with mean-annual wind speeds before extraction >7 m∕s above the wake centerline, consistent with large-eddy-simulation
and did not account for increasing competition. These two off- results (18). Greater shearing stress above the wake increased
setting factors caused their results to be similar to the land-SWPP subgrid-scale turbulent kinetic energy (TKE) there, increasing
(72 TW) and the land plus near-shore estimate (approximately the downward transport of horizontal momentum from above
80 TW) found here. to the turbines. Downward transport of horizontal momentum
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

If only 50% of land-based wind were in economically viable to a turbine wake was also increased in the model by subgrid-scale
locations (3), the feasible wind potential on land (not counting thermal turbulence and grid-scale gravity waves when they were
near shore) here would be approximately 36 TW, 36 times the present. Lesser shearing stress below the wake decreased TKE
single-equation estimate from ref. 3. and downward momentum fluxes near the surface, as in ref. 18.
The SWPP at 10 km in the jet streams (approximately 378 TW— Evaporation rates are proportional to both surface wind speed
Table 1 and Fig. S3G) was approximately 150% that at 100 m and surface shearing stress, and both decreased in all surface
despite fewer turbines in the jet stream case. The maximum jet turbine simulations, reducing evaporation and water vapor (e.g.,
stream power availability was approximately 50 times that of a Fig. 1C). These calculations were all made with the model resol-
7.5 TW estimate from ref. 2, who used a global model with only ving the bottom kilometer with 15 vertical layers, including five
10 vertical layers to minimize computer time (vs. 68 layers here) layers intersecting turbine rotors.
and an elevated drag coefficient rather than extracting momen- Drag from blade rotation also creates turbulence in the form
tum based on a turbine power curve. The difference again high- of small-scale vortices that can enhance mixing. This mechanism
lights the importance of calculating the SWPP from physical has been suggested by ref. 27 to explain why wind turbines de-
principles. SWPPs are extractable energy potentials, not just crease downwind surface temperatures during the day, when the
available energy potentials, thus include losses and efficiencies. lapse rate is generally unstable, and slightly increase them at
Airborne jet-stream turbines would require energy to ascend night, when the lapse rate is generally stable but winds at hub
and descend and may not operate all year. This analysis does not height are stronger. However, blade-generated turbulence under
quantify such losses, only extractable energy. neutral conditions is observed to be transported and dissipate
The extractable power globally at 100 m and, separately, at downwind in a spiral motion (28), with greater turbulence inten-
10 km in the jet streams, are both independently less than the sity above the turbine centerline than below (28, 29). While such
total extractable power in the wind at all altitudes, estimated turbulence reduces mean wind speeds in the wake, it also in-
broadly as 450–3,800 TW (4–9). These previous studies, though, creases the downward transport of faster winds from aloft into the
did not consider extraction at a single altitude, such as the height wake. Blade-generated turbulence is transported vertically due to
of modern wind turbines nor did they use a 3D model to make shear turbulence generated by the velocity deficit in the wake and
their estimates. Extraction of power at each 100 m and at 10 km ambient turbulence rather than on its own (28). As such, blade-
does not give the same dissipation as complete extraction of generated turbulence decreases substantially between its peak
kinetic energy from the atmosphere, as seen in Fig. 1; instead, above the centerline and surface and little may get to the surface,
each results in wind reduction over a vertical segment of the at- as indicated by at least some measurements and high-resolution
mosphere, decreasing with distance from the height of extraction. modeling (figure 1 of ref. 29). This result may differ under very
Simulations N-P examine whether approximately four million unstable conditions. Even when blade-generated turbulence
5-MW turbines (20 TW installed) can provide at least 5.75 TW reaches the ground, it may largely be offset by reduced shearing
of delivered power, enough to supply 50% of all-purpose end- stress below the turbine caused by reduced wind speed in the
use power demand in 2030 for a world energy infrastructure wake, resulting in little net surface turbulence, consistent with the
converted to wind, water, and sunlight (WWS) and electricity/ aforementioned measurements (29). Both the reduced wind

15682 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1208993109 Jacobson and Archer

 speed and small turbulence change near the surface due to tur- Global warming increases temperatures at the poles more than
bines contributed in the model to reducing surface evaporation. lower latitudes. The temperature gradient reduction could re-
Uncertainties in the treatment of turbulence still exist due to both duce global near-surface wind resources in the future although
the coarse horizontal resolution of the model and the simplifica- ocean wind resources over the last 25 years have increased in the
tion of no turbine-rotor generated turbulence. global average according to multiple datasets (32). Higher water
Reduced evaporation reduced evaporative cooling of the sur- vapor due to future warming will also likely offset reduced water
face, first warming the surface. However, because evaporated vapor due to wind turbines.
water vapor normally recondenses in the atmosphere to form Jet-stream turbines reduced mean wind speeds at altitudes
clouds, releasing latent heat there, the reduction in water vapor above and below them, but increased boundary-layer wind speeds
reduced latent heat release in the air, cooling the air due as a (Fig. 1C). Like in the surface case, turbines decreased zonal wind
result of this process. Because water vapor contributes to air pres- speeds substantially (Fig. S5A), but increased meridional wind
sure, reducing water vapor also reduced globally averaged air speeds (Fig. S5B), moving air pole-ward at 10 km but equator-
pressure by approximately 0.3 and approximately 0.1 hPa in the ward near the surface in both hemispheres, following the respec-
global (Simulation B) and land (I) cases, respectively. Because tive pressure gradients (Fig. S5C). Lower surface pressure in the
water vapor is a greenhouse gas, reducing it increased thermal- tropics through midlatitudes caused air to rise, expand, and cool
IR radiation escape to space, cooling the surface further. How- adiabatically, decreasing temperatures at all altitudes (Fig. S5D)
ever, less water also reduced cloudiness, increasing solar radia- and increasing both cloud liquid below 5 km (Fig. S5E) and cloud
tion to the surface during the day but increasing outgoing ice above that. Enhanced cloudiness increased precipitation,
thermal-IR at night, thus causing a slight warming at night, as ob- and both, together with net divergence, decreased water vapor in
served (27, 30). The net effect of all five changes (air cooling due the tropics and subtropics and increased it toward the poles
to lower atmospheric latent heat release, ground warming due to (Fig. S5F). Compressional heating over the poles increased
lower surface water evaporation, air and ground cooling due to a temperatures there, but the net effect of jet stream turbines was
reduced water vapor greenhouse effect, ground warming due to surface cooling by >1 K (Fig. S5F), as cold air advection from the
reduced daytime cloudiness, and ground cooling due to reduced

APPLIED PHYSICAL
Poles prevailed near the surface. Interestingly, the higher bound-
nighttime cloudiness) was a globally averaged surface-air tem- ary-layer wind speeds (Fig. 1C and Fig. S5A) increased evapora-

SCIENCES
perature decrease in 15 out of the 16 surface-turbine simulations. tion there, but enhanced condensation of that vapor decreased
This result is expected because water vapor is known to cause net column vapor at low latitudes (Fig. S5F).
warming of the atmosphere, so reducing it should cause cooling In sum, increasing the number of wind turbines worldwide
(31). Temperature results, though, are still uncertain, particularly allows energy extraction relatively proportional to the number
due to the uncertainty of clouds and the transient nature of the of turbines until saturation is reached. Saturation occurs when
simulations and could change over longer simulations because sources of kinetic energy at nearby altitudes and creation of
full temperature responses take decades to realize. A certain ben- kinetic from potential energy are exhausted. At saturation, each
efit of the slower winds, though, is the reduction in wind-driven additional turbine still extracts energy, but that extraction reduces
soil dust; sea spray; and spore, pollen, and bacteria emissions, energy available to other turbines, so the average extraction
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

reducing human exposure to small particles that penetrate deep among all turbines decreases to maintain a constant SWPP. The
into the lungs. results here suggest that saturation of wind power availability will
Globally distributed turbines decreased zonal winds; however, not limit a clean-energy economy. However, spreading wind
they increased meridional winds in the pole-ward direction in farms out worldwide in high-wind locations will increase wind
both hemispheres (Fig. S4 A and B). The pole-ward transport farm efficiency and reduce the number of farms needed com-
of air increased the pressure gradient between the poles and pared with packing wind farms side-by-side. The careful siting
Equator by approximately 15–25 hPa, supporting the contention of wind farms will minimize costs and the overall impacts of a
that the atmosphere responded to the increased dissipation of global wind infrastructure on the environment.
kinetic energy by increasing some of its available potential energy
via enhanced pole-to-Equator pressure gradients. Reduced water ACKNOWLEDGMENTS. Funding sources include National Science Foundation,
vapor partial pressure at low latitudes contributed slightly to the U.S. Environmental Protection Agency, and National Aeronautics and Space
enhanced pressure gradient. Administration high-end computing.

1. Archer CL, Caldeira K (2009) Global assessment of high-altitude wind power. Energies 13. Keith DW, et al. (2004) The influence of large-scale wind power on global climate. Proc
2:307–319. Natl Acad Sci USA 101:16115–16120.
2. Miller LM, Gans F, Kleidon A (2011) Jet stream wind power as a renewable energy 14. Kirk-Davidoff DB, Keith D (2008) On the climate impact of surface roughness anoma-
resources: Little power, big impacts. Earth Syst Dynam 2:201–212. lies. J Atmos Sci 85:2215–2234.
3. De Castro C, Mediavilla M, Miguel LJ, Frechoso F (2011) Global wind power potential 15. Barrie D, Kirk-Davidoff DB (2010) Weather response to a large wind turbine array.
and technological limits. Energy Policy 39:6677–6682. Atmos Chem Phys 10:769–775.
4. Lorenz EN (1967) The Nature and Theory of the General Circulation of the Atmosphere 16. Wang C, Prinn RJ (2010) Potential climatic impacts and reliability of very large-scale
(WMO, Geneva) p 161. wind farms. Atmos Chem Phys 10:2053–2061.
5. Gustavson MR (1979) Limits to wind power utilization. Science 204:13–17. 17. Miller LM, Gans F, Kleidon A (2011) Estimating maximum global land surface wind
6. Peixoto JP, Oort AH (1992) Physics of Climate (American Institute of Physics, New York)
power extractability and associated climatic consequences. Earth Syst Dynam 2:1–12.
p 520.
18. Calaf M, Meneyeau C, Meyers J (2010) Large eddy simulation study of fully developed
7. Sorensen B (2004) Renewable Energy: Its Physics, Engineering, Use, Environmental Im-
wind-turbine array boundary layers. Phys Fluids 22:015110.
pacts, Economy, and Planning Aspects (Elsevier Academic Press, London), 3rd Ed, p 86.
19. Sta Maria MRV, Jacobson MZ (2009) Investigating the effect of large wind farms on
8. Li L, Ingersoll AP, Jiang X, Feldman D, Yung YL (2007) Lorenz energy cycle of the global
energy in the atmosphere. Energies 2:816–836.
atmosphere based on reanalysis datasets. Geophys Res Lett 34:L16813.
20. Jacobson MZ, Wilkerson JT, Naiman AD, Lele SK (2011) The effects of aircraft on cli-
9. Stacey FD, Davis PM (2008) Physics of the Earth (Cambridge Univ Press, Cambridge),
4th Ed, p 531. mate and pollution. Part I: Numerical methods for treating the subgrid evolution of
10. Jacobson MZ, Delucchi MA (2011) Providing all global energy with wind, water, and discrete size- and composition-resolved contrails from all commercial flights world-
solar power, part I: Technologies, energy resources, quantities and areas of infrastruc- wide. J Comp Phys 230:5115–5132.
ture, and materials. Energy Policy 39:1154–1169. 21. Jacobson MZ, Ten Hoeve JE (2012) Effects of urban surfaces and white roofs on global
11. Milborrow DJ (1980) The performance of arrays of wind turbines. J Wind Eng Ind and regional climate. J Climate 25:1028–1044.
Aerod 5:403–430. 22. Baidya Roy S, Pacala SW, Walko RI (2004) Can large wind farms affect local meteor-
12. Li Y, Calisal SM (2010) Estimating power output from a tidal current turbine farm with ology. J Geophys Res 109:D19101.
first-order approximation of hydrodynamic interaction between turbines. Int J Green 23. Baidya Roy S (2011) Simulating impacts of wind farms on local hydrometeorology.
Energy 7:153–163. J Wind Eng Ind Aerod 99:491–498.

Jacobson and Archer PNAS ∣ September 25, 2012 ∣ vol. 109 ∣ no. 39 ∣ 15683
 24. Pryor SC, Nikulin G, Jones C (2012) Assessing climate change impacts on the near- 29. Crespo A, Hernandez J (1996) Turbulence characteristics in wind-turbine wakes. J Wind
term stability of the wind energy resources over the United States. J Geophys Res Eng Ind Aerod 61:71–85.
117:D03117. 30. Baidya Roy S, Traiteur JJ (2010) Impacts of wind farms on surface air temperatures.
25. Lu X, McElroy M, Kiviluoma J (2009) Global potential for wind-generated electricity. Proc Natl Acad Sci USA 107:17899–17904.
Proc Natl Acad Sci USA 106:10933–10938.
31. Kiehl JT, Trenberth KE (1997) Earth’s annual global mean energy budget. Bull Am
26. Archer CL, Jacobson MZ (2005) Evaluation of global wind power. J Geophys Res
Meteorol Soc 78:197–208.
110:D12110.
27. Zhou L, et al. (2012) Impacts of wind farms on land surface temperatures. Nature 32. Wentz FJ, Ricciardulli L (2011) Comment on “Global trends in wind speed and wave
Climate Change 2:539–543. height”. Science 334:905.
28. Vermeer LJ, Sorensen JN, Crespo A (2003) Wind turbine wake aerodynamics. Progr 33. Lettau H (1969) Note on aerodynamic roughness-parameter estimation on the basis of
Aero Sci 39:467–510. roughness-element description. J Appl Meteorol 8:828–832.
Downloaded from https://www.pnas.org by 161.53.40.100 on April 25, 2024 from IP address 161.53.40.100.

15684 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1208993109 Jacobson and Archer