2014 American Control Conference (ACC)

June 4-6, 2014. Portland, Oregon, USA

Geoengineering: the World’s largest control problem

Douglas G. MacMartin, Ben Kravitz, and David W. Keith

Abstract— Solar geoengineering (or Solar Radiation Manage- with either a spatially-uniform aerosol layer or uniform solar
ment, SRM) refers to any intentional, large-scale manipulation reduction yield reasonable spatial compensation of temper-
of the Earth’s incoming solar radiation to offset some of the ature changes due to increased greenhouse gases [15, 16]
effects of anthropogenic greenhouse gases, reducing the associ-
ated risks from climate changes. Examples of such methods are but poorer compensation of precipitation; regional disparities
injecting aerosols into the stratosphere or increasing marine in outcomes [17] could even lead to conflict. Further, the
cloud reflectivity, both of which would reflect some sunlight climate system is uncertain, requiring design for a system
back to space. There are many serious concerns associated not fully understood, yet experimentation at a scale sufficient
with any such approach, and also many challenges. One often to measure climate response involves risks similar to deploy-
overlooked aspect in geoengineering research is that this is a
control problem, requiring (a) feedback of the observed climate ment: if solar geoengineering is used, we must get it right
state to manage significant uncertainty in both the radiative the first time. These last two issues involving uncertainty and
forcing and the climate’s dynamic response to this forcing, imprecise compensation can be at least partially addressed
and (b) optimization of the distribution of radiative effect to with control theory.
minimize regional disparities as well as side-effects from the
In addition to challenges that span physical, social, polit-
geoengineering implementation. We present recent progress on
control for this challenging problem, building on [1, 2], and ical, and ethical issues, geoengineering is a control problem
discuss open research gaps. This is the first time an explicit that requires choosing the amount, spatial pattern, and tem-
external feedback loop has been implemented in a fully coupled poral pattern of radiative forcing (or equivalently e.g. aerosol
general circulation model of the Earth’s climate. injection latitude and timing), to minimize the combined
I. INTRODUCTION risks from both climate changes (temperature, precipitation,
etc.) and from the implementation of geoengineering (less
Solar geoengineering might reduce risks from climate forcing results in lower risks, such as ozone depletion). These
change [3]–[5], but research is required to understand the choices must be made despite significant model uncertainty,
effectiveness and risks and to develop knowledge so that much of which is unresolvable without large-scale, lengthy,
if it were ever deployed, it could be implemented as ef- and risky global experiments [18, 19].
fectively as possible. The most frequently discussed option Here we describe two key contributions. We first discuss
is injection of sulfate aerosols into the stratosphere to re- global system dynamics and feedback algorithms to achieve
flect a small fraction of incoming sunlight [4, 6]. This is a climate goal in the face of uncertainty (Section II; see
motivated by the global cooling that follows large volcanic also [2]). Next we introduce optimization of the spatial and
eruptions; e.g., after the Pinatubo eruption introduced more seasonal pattern of forcing to minimize regional climate
than 30 Tg of sulfate aerosols into the stratosphere in 1991, change or other metrics while constraining the amount of
global mean temperatures temporarily decreased by roughly solar reduction used (Section III; see also [1]). Finally, we
0.5◦ C [7]. “Marine cloud brightening” [8] could similarly briefly discuss open research questions.
reflect sunlight; other options include space-based [9] (likely
prohibitively expensive) and surface-based [10] methods.
II. FEEDBACK
Approaches to remove CO2 from the atmosphere are also
generally described as “geoengineering”, but these yield only Models of the climate response to greenhouse gases (or
slow changes in climate and pose very different risks from solar reductions) do not all agree; see, e.g., Figure 1, which
solar geoengineering techniques; we do not discuss these. compares 24 fully coupled atmosphere-ocean general circula-
Solar geoengineering could quickly reduce global mean tion models (AOGCMs) from the Coupled Model Intercom-
temperatures at relatively low cost [11], but with a number parison Project, phase 5 (CMIP5). Model discrepancies are
of risks [12,13]. Using geoengineering may lead to a delay in even larger for regional precipitation responses [16]. Further-
reducing CO2 emissions; as a result, geoengineering could more, in addition to uncertainty in the climate response, there
be required for centuries. Any technology will have some is uncertainty in the radiative forcing from any particular
undesired side effects, e.g., stratospheric approaches will lead geoengineering implementation (e.g., how much sunlight is
to changes in ozone and other chemistry [14]. Simulations reflected by stratospheric aerosols depends on particle size,
which depends on coagulation; there may also be some
D. G. MacMartin is in the Department of Computing and Mathemat-
ical Sciences, California Institute of Technology, Pasadena, CA 91125 influence on cirrus clouds that could change the net radiative
Email: [email protected] effect [20]). Open-loop strategies will therefore result in
B. Kravitz is in the Atmospheric Sciences and Global Change Division, significant error in meeting specific climate objectives, and
Pacific Northwest National Laboratory, Richland, WA 99352.
D. W. Keith is in the School of Engineering and Applied Sciences and feedback of the observed climate state must be an essential
Kennedy School of Government, Harvard University, Cambridge, MA 02139 element of any solar geoengineering strategy [30].

978-1-4799-3274-0/$31.00 ©2014 AACC 2401

 Amplitude ( C per % solar)
ACCESS1−0
7 CNRM−CM5 0
10 Direct calculation
CSIRO−Mk3−6−0
CanESM2
Fit to eq (3)
6 EC−EARTH
Global mean response ( C for 4xCO2)

FGOALS−g2

o
FGOALS−s2
5 GFDL−CM3
GFDL−ESM2G
GFDL−ESM2M −1
10
o

GISS−E2−H 512 256 128 64 32 16 8 4 2 1 0.5

4
GISS−E2−R Period (years)
HadGEM2−ES
IPSL−CM5A−LR 0
3 IPSL−CM5A−MR
IPSL−CM5B−LR

Phase (deg)
MIROC−ESM −20
2 MIROC5
MPI−ESM−LR
MPI−ESM−MR −40
1 MPI−ESM−P
MRI−CGCM3
NorESM1−M −60
0 bcc−csm1−1 512 256 128 64 32 16 8 4 2 1 0.5
0 50 100 150 200
inmcm4 Period (years)
Time (years) HadCM3L

Fig. 2. Frequency response G(s) of global mean temperature in response

Fig. 1. Model uncertainty: The step response to an instantaneous quadru- to 1% perturbations in solar forcing calculated for HadCM3L (from [21]),
pling of CO2 is shown for 24 different CMIP5 climate models; the solid and least-squares fit to a box-diffusion model (from [2]).
line corresponds to the box-diffusion model in Fig. 2.
forcing, the best fit to the calculated frequency response gives
A. Dynamic Model λ = 1.2 W m−2 K−1 , τ = β 2 /(λ2 κ) = 13 years, and C =
A box diffusion model represents the dynamic behavior 3.2 × 106 J m−2 K−1 .
of global mean temperature well enough to design the The step response in Fig. 1 illustrates the robustness
control algorithm. This model consists of heat diffusion into challenge. The steady-state response varies by more than a
an effectively semi-infinite reservoir (deep ocean), with a factor of two, and the dynamics also change. Some models
surface layer of fixed heat capacity (land, atmosphere, and are better fit with a much larger heat capacity [25], which
surface ocean). A semi-infinite diffusion model matches the would increase the high frequency phase lag in Fig. 2.
global mean temperature response in the HadCM3L GCM
over a wide range of frequencies [21]; and also fits the B. Feedback Overview
transient response of most CMIP5 models [22]. Including The block diagram in Fig. 3 illustrates the coupled human-
a surface layer better predicts the short time-scale response. climate feedback system. The dynamic system characterized
For radiative forcing F (t), the surface temperature T (t) and by G(s) describes the global mean temperature response to
deep ocean temperature Td (z, t) satisfy imposed radiative forcing F , including forcing associated
with anthropogenic climate change Fd , solar geoengineering
dT ∂Td
= F − λT + β C (1) Fs , and perturbations w responsible for natural variability
dt ∂z z=0 (which has a 1/f power spectrum, so w is approximately
∂Td ∂ 2 Td white). The solar geoengineering forcing includes the best
=κ 2 , (2)
∂t ∂z estimate F̂ of the radiative forcing required to maintain
with boundary condition Td (0, t) = T (0, t) (taking the top T = Tref in the presence of Fd , as well as the component
of the deep ocean as z = 0). The parameter λ describes Fc that corrects for errors in this estimate based on feedback
the natural climate feedback (the change in radiation due to of the observed climate state.
a change in surface temperature), C = cρH is the surface If the dynamics G(s) and radiative forcing Fd were
layer heat capacity per unit area, κ the thermal diffusivity, known, then F̂ could be chosen so that T only differs
and β = cρκ for density ρ and specific heat capacity c. from Tref by natural variability; in reality, using the best
Taking Laplace transforms [2,23,24], the transfer function estimate of F̂ will yield some error that could be corrected
from radiative forcing perturbation F (s) to temperature with feedback. We define Fr = Fd + F̂ as the residual
anomaly T (s) is radiative forcing. The temperature error relative to the desired
  temperature, Te = T − Tref , can then be expressed as
T (s) 1
G(s) = = (3) G(s)
F (s) λ + β(s/κ)1/2 + Cs Te = (Fr + w) = Gfb (s)(Fr + w) (4)
1 + G(s)K(s)
Equation (3) is compared with the calculated HadCM3L
frequency response in Fig. 2; the latter was computed by where Gfb (s) = G(s)S(s). The static equivalent is well
introducing 1% sinusoidal variations in solar forcing [21]. known in climate literature [26, 27]. However, recognizing
In this model, a 2.3% solar reduction would offset the that this predicts the transient as well as steady-state response
3.7 Wm−2 radiative forcing from a doubling of atmospheric is novel in climate analysis, as is the possibility for the
CO2 [1]. Using this to convert solar reduction into radiative feedback K(s) to be chosen as opposed to being a property

2402
 w Noise added to
1
simulate variability 10 PI control, N=1

Magnitude, |GK|
Global mean
temperature PI control, N=2
+ ?
+
Fd - e - G(s) -T Magnitude = 1
0
GHG forcing
+ 6
10
Fs
SRM forcing Climate system −1
10
Geoengineering 1/100 1/50 1/20 1/10 1/5 1/2
Frequency (years)−1

F̂ - e ? −100
−K(s)  
+

Phase of GK (degrees)
+ +
Tref
−
e
Estimated forcing −120
to yield T = Tref
Fc Te Target
Feedback Error temperature −140 o
Phase margin: 53
correction o
−160 Phase margin: 26
Fig. 3. Block diagram of geoengineering feedback (from [2]), assuming −180
radiative forcing from SRM (Fs ) and other sources (Fd ) simply add. The
“noise” w simulates natural variability. The climate system is represented −200
1/100 1/50 1/20 1/10 1/5 1/2
by transfer function G(s), to predict temperature anomaly T in response
Frequency (years)−1
to radiative forcing. In addition to the feedback K(s) in response to
the deviation between observed and desired temperature, we include a
feedforward of the best estimate of geoengineering forcing F̂ required to Fig. 4. Bode plot with PI gains kp = 4 and ki = 2π, chosen to give
maintain T = Tref in the presence of the disturbance Fd . reasonable bandwidth and phase margin if the level of solar reduction is
updated every N = 1 years based on the average temperature in the previous
N = 1 years. With updates every N = 2 years, the phase margin is poor,
of the climate system. Furthermore, the waterbed effect giving significant amplification of natural variability at frequencies near the
Z ∞ control bandwidth. The accurate discrete-time response is plotted including
zero-order-hold, averaging, and frequency fold-overs from aliasing [2].
log |S(iω)|dω = 0 (5)
0

is important in understanding the dynamic behavior resulting is illustrated by comparing the ratio of the closed-loop and
from solar geoengineering. The feedback acts on temperature open-loop amplitude spectra with the predicted sensitivity
errors due to both greenhouse gases and natural variability. function.
Frequencies where |S(iω)| > 1 correspond to amplification While an important first step, there are several details yet
of natural climate variability, which can be particularly to be filled in. First, the control gains were tuned using
significant if time delay is introduced. the frequency response of the same model that was used to
From Fig. 2, PI control K(s) = kp + ki /s is sufficient for evaluate performance. This is clearly not an option if geo-
this application (see the Bode plot in Figure 4). This yields engineering is ever implemented in the real world. A second
climate model could serve as a proxy for the unknown real-
s
Gfb (s) = (6) world dynamics, to verify that a controller designed on one
ki + s(λ + kp + β(s/κ)1/2 + Cs) model is robust to the inter-model uncertainty [30]. Second, a
and zero steady-state error; proportional gain alone is equiv- PI controller illustrates the potential and also some dynamic
alent to a change in intrinsic climate feedback parameter λ. characteristics associated with implementing feedback, but
more complicated controllers may be appropriate. Adaptation
C. Feedback in a GCM could be used to better learn the dynamics as time progresses
The results in Figure 5, based on simulations in [2], [31], or model-predictive control to enforce constraints on the
represent the first ever explicit feedback loop implemented amount and rate of climate change and on the rate of change
in a fully-coupled AOGCM of the Earth’s climate. We use of applied forcing. Finally, while many climate impacts can
HadCM3L, which has resolution of 3.75◦ in longitude by be related to global mean temperature, a more realistic goal
2.5◦ in latitude in both the atmosphere and ocean, with 19 would consider other variables such as precipitation or sea
vertical levels in the atmosphere and 20 in the ocean [29]. ice, as well as regional effects. Including multiple objectives
There are three take-away messages from Fig. 5. Most will require managing more than a single degree of freedom.
important, though unsurprising, is that properly designed The next section introduces this as an optimization problem;
feedback of the “observed” temperature can maintain the future work will integrate feedback into the multiple degree
global mean temperature at a desired target value, despite of freedom optimization problem to manage uncertainty.
uncertainty in both the model and the forcing. Second (also
unsurprising to anyone experienced in designing controllers) III. OPTIMIZATION
is the importance of proper representation of system dy- The second application of control theory to solar geoengi-
namics. A poorly designed controller can result in near- neering we discuss is optimization to (a) improve the com-
oscillatory behavior, illustrated here by increasing the time pensation between the climate effects of greenhouse gases
delay, resulting in poor phase margin (see Fig. 4). The and those due to solar geoengineering, (b) maximize the
third important result is that the dynamic behavior resulting benefit with minimum solar reduction, and (c) consider trade-
from implementing feedback in a complex climate model is offs between different metrics. Some of this appears in [1].
well predicted from the simple box-diffusion model. This The radiative forcing due to greenhouse gases is relatively

2403
 1.5 4

Amplitude spectrum ( C per √(yr) )

RCP4.5

−1
PI control, N=1 2
1 PI control, N=2
2020 target Amplification
1
(K)

o
Attenuation
ref

0.5
T−T

0.5

0
0.2

−0.5 0.1
2020 2030 2040 2050 2060 1/100 1/50 1/20 1/10 1/5 1/2
Time (years) Frequency (years)
−1

Fig. 5. Simulations of geoengineering feedback using the HadCM3L GCM showing time domain (left) and ratio of closed-loop to open-loop amplitude
spectrum (right). The reference case on the left is the predicted temperature response to a “Representative Concentration Pathway” RCP4.5 scenario [28],
with feedback initiated in 2040 with the goal of returning to 2020 temperatures. The feedback gains are as in Fig. 4 with updates either every year (N = 1)
or every two years (N = 2); the latter introduces time delay that exacerbates the amplification of natural variability. The spectra (right) are computed from
500 years of simulation; the predicted sensitivities from the box-diffusion model are shown in solid lines (green for N = 1 and black for N = 2).

2
intended to compensate. For each applied radiative forcing
% solar reduction

1.5 L1 pattern, we compute the monthly means over the second half
1 L0 of a 200-year simulation to minimize errors in estimating the
0.5 L2 long time-scale behavior.
0
The monthly-mean temperature and precipitation differ-
−90 −45 −30 −15 0 15 30 45 90 ences between the 2×CO2 and baseline cases for all m grid
Latitude
cells can be concatenated into a vector b ∈ R24m. Tem-
S0 peratures and precipitations are normalized by the standard
% solar reduction

1
S1 S2 S3 S4 deviation of their interannual natural variability [17,34]; this
0.5
measures change compared to “normal” variability. Similarly,
the response to the n patterns of solar reduction can be
0 described by A ∈ R24m×n . Define u ∈ Rn as the forcing
0 30 60 90 120 150 180 210 240 270 300 330 360 amplitude corresponding to each pattern. Assuming linearity,
Day (360 = 1 year)
the residual response is then
Fig. 6. Patterns of solar reduction used in optimization: spatial (top; L0, L1,
and L2 for uniform, linear and quadratic) and seasonal variations (bottom; z = b − Au (7)
S0 for uniform and S1–S4 for seasonal). The temporal patterns sum to a
uniform distribution, giving 12 independent degrees of freedom. Minimizing a quadratic function of climate changes rel-
ative to a pre-industrial baseline while constraining the av-
uniform in space and time (latitude and season), while solar erage insolation reduction yields a constrained least-squares
geoengineering is most effective in summer in the respective problem that can be solved using quadratic programming:
hemisphere [15]. Furthermore, the different mechanisms of
J ∗ = min uT (AT A)u + 2bT Au

with Cu ≤ 0 (8)
radiative forcing have different relative impacts on tempera- u
ture vs precipitation [32]. As a result, spatially and seasonally In addition to the constraint on average solar reduction, the
uniform solar reductions do not fully compensate climate constraint that we can only reduce sunlight, not increase it,
changes due to greenhouse gases in any model [16]. The can also be written as Cu ≤ 0, with each row of C enforcing
compensation can be improved by optimizing the distribution this at a particular latitude and time of year. (It is sufficient
of solar reduction. This might be achieved, for example, by to enforce this at only a few times and latitudes.) It is also
varying the latitude and time of year of stratospheric aerosol straightforward to include in the same framework variables
injection. Here we explore the potential benefit for the one might wish to maximize, such as Arctic sea ice extent.
solar reduction patterns in Fig. 6. In practice, neither these We also consider the forcing pattern that minimizes the
patterns nor the resulting climate response will be known, worst-case climate change in any grid-cell:
and feedback would be needed to manage this uncertainty.
We again use HadCM3L, and compute the climate re- J ∗ = min max |(b − Au)i | with Cu ≤ 0 (9)
u i
sponse for each spatial and temporal pattern in Fig. 6.
Assuming linear superposition to estimate the response to where (·)i is the ith element of the vector. Figure 8 is
any combination of these patterns (eq. (7)) is a good approx- obtained by optimizing as in (8) with constraints Au ≤ α −b
imation [33,34]. A pre-industrial CO2 (278 ppm) simulation and −Au ≤ α + b, where α is the allowable worst-case.
defines the “baseline” or desired climate, and a 2×CO2 case Allowing the distribution of solar reduction to vary in
used to define the perturbed climate that geoengineering is space and time improves the compensation of climate change

2404
 8
rms temperature and precipitation (normalized)
2xCO2 Minimize global rms only
7
4 Minimize worst−case only

Global−rms (normalized)
6 Seasonal Uniform solar
0.5% 0.5% 3 Spatial only reduction
5 only

1% 1% 2
4 3.5%

3 1 Non−uniform both
1.5% 1.5% spatially and seasonally
3%
2 2% 0
2% 6 8 10 12 14 16
2.5%
1 2.5% Worst−case (normalized)

NH September sea ice extent (x10 km )

Pre−industrial baseline 8

2
0
0 2 4 6 8 10 12 7

6
Pre−industrial baseline
NH September sea ice extent (x106 km2) 6
Fig. 7. Trade-off between maximizing Arctic sea-ice extent and minimizing 5
global-rms temperature and precipitation changes (from [1]) if the solar
reduction is optimized in space and time. The trade-off is shown for 4
increasing uniform solar reduction (solid black line) and non-uniform 3 2xCO2 Uniform
reduction weighting only sea ice (upper dashed line), only temperature and Spatial & seasonal
precipitation (lower dashed line), or some combination (connecting lines, 2
Seasonal only
at constant average solar reduction). Non-uniform solar reduction improves 1 Spatial only
outcomes (as measured here), and can also result in climates that are not
achievable with only uniform reductions. 0
0 0.5 1 1.5 2 2.5 3
Average % solar reduction

Fig. 8. Optimization of different objectives, evaluating the benefit of

due to greenhouse gases in this model, and also allows spatially- or seasonally-varying solar reductions. The top panel considers the
trade-off between minimizing the global rms departure from pre-industrial
different trade-offs to be made depending on the weighting temperature and precipitation and the regional worst-case, addressing re-
of different objectives in the optimization. Results in [1] gional disparities. The lower panel illustrates the potential increase in sea
illustrate several trade-offs, including (a) the ability to reduce ice as a function of average solar reduction; for the case with all degrees
of freedom, this corresponds to the upper dashed lines in Fig. 7.
the worst-case climate change over any region while still
reducing the global rms climate change; this is relevant for
minimizing regional disparities, (b) the trade-off between IV. CHALLENGES AND OPEN QUESTIONS
minimizing global rms temperature changes and global rms Several open issues have been noted, including using
precipitation changes, and (c) the trade-off between mini- feedback (Sec. II) to manage uncertainty with multiple
mizing global rms temperature and precipitation changes and degrees of freedom (Sec. III). As a concrete example, an
maximizing Arctic sea ice. The last of these is reproduced in initial target for geoengineering might be the Arctic [35], to
Fig. 7 for illustration. The trade-off is shown as a function reduce risks associated with permafrost thaw, sea ice loss,
of the average solar reduction, a proxy for the negative or Greenland ice-sheet melt. Implementing such a scenario
impacts associated with implementing geoengineering (e.g., without adverse tropical precipitation effects [36] would
decreased ozone from stratospheric aerosols). For example, require a counterbalancing southern hemisphere solar reduc-
restoring pre-industrial sea ice in this model requires ∼2.5% tion. This would almost certainly require feedback to manage
reduction in sunlight if the reduction is spatially and season- uncertainty. Also noted is the need for more complex control
ally uniform, but the same sea ice extent can be obtained with algorithms incorporating learning, and constraints both on
only 0.5% solar reduction if solar changes are concentrated climate variables and to ensure implementation is feasible.
optimally in space and time. Controls engineers may also be useful in designing tests to
Here we further evaluate whether similar performance maximize signal-to-noise ratio for system identification [19].
improvement can be obtained with only spatially uniform Climate changes occur over decadal and longer time-
(but seasonally varying) or seasonally uniform (but spatially scales. While stratospheric aerosol concentrations could not
varying) solar reduction. Figure 8 demonstrates that while be rapidly modulated, marine cloud brightening could be
some improvement in worst-case climate changes are possi- modulated at time scales of order one week, opening up the
ble with only seasonal variation, the biggest benefit comes possibility of responding at the time-scale of hurricane evolu-
from allowing the solar reduction to vary as a function tion, for example [37]. (Hurricane control has been suggested
of latitude; introducing both provides only minor further earlier, but without a realistic mechanism for excitation [38]).
improvements. For different objectives, different degrees of Spatial and temporal control of solar reduction could also
freedom are important to include. The sea-ice improvement be used to alter the dynamics of El Niño [39]. Averaging
of Fig. 7 requires both spatial and seasonal variations in solar over space and time tends to reduce model uncertainty, and
reduction, although some benefit is again obtained with only thus attempts to control local dynamics at these shorter time-
spatial (concentrated in the Arctic). scales will be challenging.

2405
 More broadly, introducing a feedback loop around a cli- [13] A. Robock, “20 reasons why geoengineering may be a bad idea,”
mate model as in Sec. II is relevant beyond geoengineering Bulletin Atom. Sci., vol. 64, pp. 14–18, 2008.
[14] S. Tilmes, R. Müller, and R. Salawitch, “The sensitivity of polar ozone
for understanding human-climate interactions (e.g., incorpo- depletion to proposed geoengineering schemes,” Science, vol. 320, no.
rating gradual learning to change predicted optimal pathways 1201–1205, 2008.
for reducing emissions), understanding natural climate feed- [15] B. Govindasamy and K. Caldeira, “Geoengineering Earth’s radiation
balance to mitigate CO2 -induced climate change,” Geophys. Res. Lett.,
backs (e.g., permafrost thaw leading to CO2 and methane vol. 27, pp. 2141–2144, 2000.
release leading to further warming), or better understanding [16] B. Kravitz et al., “Climate model response from the Geoengineering
of model behavior (e.g., automated model tuning). Model Intercomparison Project (GeoMIP),” J. Geophys. Res., vol. 118,
2013.
[17] K. L. Ricke, M. Granger Morgan, and M. R. Allen, “Regional climate
V. SUMMARY COMMENTS response to solar-radiation management,” Nature Geoscience, vol. 3,
For a given level of climate change, there is clearly less pp. 537–541, 2010.
[18] A. Robock, M. Bunzl, B. Kravitz, and G. L. Stenchikov, “A test for
climate risk if this is reached through reduced greenhouse geoengineering?” Science, vol. 327, pp. 530–531, 2010.
gas emissions rather than geoengineering. However, because [19] D. G. MacMynowski, D. W. Keith, K. Caldeira, and H.-J. Shin, “Can
of technical and societal inertia, atmospheric CO2 concen- we test geoengineering?” Energy Environ. Sci., vol. 4, pp. 5044–5052,
2011.
trations are likely to continue to increase, and for a given [20] M. Kuebbeler, U. Lohmann, and J. Feichter, “Effects of stratospheric
atmospheric concentration, there may be less risk if some sulfate aerosol geo-engineering on cirrus clouds,” Geophys. Res. Lett.,
amount of geoengineering is used rather than none. While vol. 39, no. 23, 2012.
[21] D. G. MacMynowski, H.-J. Shin, and K. Caldeira, “The frequency re-
there are many concerns associated with geoengineering, it sponse of temperature and precipitation in a climate model,” Geophys.
is essential to research the problem to know what can and Res. Lett., vol. 38, 2011, l16711.
cannot be achieved, and to be prepared to implement it [22] K. Caldeira and N. Myhrvold, “Projections of the pace of warming
following an abrupt increase in atmospheric carbon dioxide concen-
intelligently rather than in a crisis. Control and dynamics are tration,” Env. Res. Lett., vol. 8, no. 3, 2013.
essential element of geoengineering research, indeed, this is [23] S. A. Lebedoff, “Analytic solution of the box diffusion model for a
the largest control problem ever considered! global ocean,” J. Geophys. Res., vol. 93, no. D11, pp. 14 243–14 255,
1988.
ACKNOWLEDGMENTS [24] M. Morantine and R. G. Watts, “Upwelling diffusion climate models:
Analytical solutions for radiative and upwelling forcing,” J. Geophys.
BK is supported by the Fund for Innovative Climate and Res., vol. 95, no. D6, pp. 7563–7571, 1990.
[25] I. M. Held, M. Winton, K. Takahashi, T. Delworth, F. Zeng, and G. V.
Energy Research (FICER). The Pacific Northwest National Vallis, “Probing the fast and slow components of global warming by
Laboratory is operated for the U.S. Department of Energy returning abruptly to preindustrial forcing,” J. Climate, vol. 23, pp.
by Battelle Memorial Institute under contract DE-AC05- 2418–2427, 2010.
75RL01830. DGM thanks Carnegie Institution of Science [26] J. Hansen, A. Lacis, D. Rind, G. Russell, P. Stone, I. Fung, R. Ruedy,
and J. Lerner, “Climate sensitivity: Analysis of feedback mechanisms,”
and Ken Caldeira in particular for hospitality and computing in Climate Processes and Climate Sensitivity, ser. Geophysical Mono-
resources used in this work. graph. Am. Geophys. Union, 1984, vol. 29, pp. 130–163.
[27] G. H. Roe and M. B. Baker, “Why is climate sensitivity so unpre-
R EFERENCES dictable?” Science, vol. 318, pp. 629–632, 2007.
[28] Vuuren et al., “The Representative Concentration Pathways: An
[1] D. G. MacMartin, D. W. Keith, B. Kravitz, and K. Caldeira, “Man- Overview,” Climatic Change, vol. 109, pp. 5–31, 2011.
agement of trade-offs in geoengineering through optimal choice of [29] C. Jones, “A fast ocean GCM without flux adjustments,” J. Atm.
non-uniform radiative forcing,” Nature Climate Change, vol. 3, pp. Oceanic Tech., vol. 20, pp. 1857–1868, 2003.
365–368, 2013. [30] B. Kravitz, D. G. MacMartin, D. T. Leedal, P. J. Rasch, and A. J.
[2] D. G. MacMartin, B. Kravitz, D. W. Keith, and A. J. Jarvis, “Dynamics Jarvis, “Feedback and management of uncertainty in meeting climate
of the coupled human-climate system resulting from closed-loop goals with solar geoengineering,” submitted, 2014.
control of solar geoengineering,” Clim. Dyn., 2013. [31] A. Jarvis and D. Leedal, “The Geoengineering Model Intercomparison
[3] D. Keith, “Geoengineering the climate: History and prospect,” Annual Project (GeoMIP): A control perspective,” Atm. Sci. Lett., vol. 13, pp.
Rev. Energy Environ., vol. 25, pp. 245–284, 2000. 157–163, 2012.
[4] P. J. Crutzen, “Albedo enhancement by stratospheric sulfur injections: [32] G. Bala, P. B. Duffy, and K. E. Taylor, “Impact of geoengineering
A contribution to resolve a policy dilemma?” Climatic Change, vol. 77, schemes on the global hydrological cycle,” PNAS, vol. 105, no. 22,
pp. 211–219, 2006. pp. 7664–7669, 2008.
[5] The Royal Society, Geoengineering the climate: Science, governance [33] G. A. Ban-Weiss and K. Caldeira, “Geoengineering as an optimization
and uncertainty. Royal Society, London, UK, 2009. problem,” Environ. Res. Lett., vol. 5, 2010.
[6] M. I. Budyko, “Climatic changes,” American Geophysical Society, [34] J. Moreno-Cruz, K. Ricke, and D. W. Keith, “A simple model to
Washington D.C., Tech. Rep., 1977, 244 pp. account for regional inequalities in the effectiveness of solar radiation
[7] B. J. Soden, R. T. Wetherald, G. L. Stenchikov, and A. Robock, management,” Climatic Change, vol. 110, no. 3-4, pp. 649–668, 2011.
“Global cooling following the eruption of Mt. Pinatubo: a test of [35] M. C. MacCracken, H.-J. Shin, K. Caldeira, and G. A. Ban-Weiss,
climate feedback by water vapor,” Science, vol. 296, pp. 727–730, “Climate response to imposed solar radiation reductions in high
2002. latitudes,” Earth Syst. Dynam., vol. 4, pp. 301–315, 2013.
[8] J. Latham et al., “Marine cloud brightening,” Phil. Trans. Royal Soc. [36] J. M. Haywood, A. Jones, N. Bellouin, and D. Stephenson, “Asym-
A, vol. 370, pp. 4217–4262, 2012. metric forcing from stratospheric aerosols impacts Sahelian rainfall,”
[9] R. Angel, “Feasibility of cooling the Earth with a cloud of small Nature Climate Change, 2013.
spacecraft near the inner Lagrange point (L1),” PNAS, pp. 17 184– [37] J. Latham, B. Parkes, A. Gadian, and S. Salter, “Weakening of
17 189, 2006. hurricanes via marine cloud brightening (MCB),” Atmospheric Science
[10] P. J. Irvine, A. Ridgwell, and D. J. Lunt, “Climatic effects of Letters, vol. 13, no. 4, pp. 231–237, 2012.
surface albedo geoengineering,” Journal of Geophysical Research: [38] J. M. Henderson, R. N. Hoffman, S. M. Leidner, T. Nehrkorn, and
Atmospheres, vol. 116, no. D24, 2011. C. Grassotti, “A 4D-Var study on the potential of weather control and
[11] J. McClellan, D. W. Keith, and J. Apt, “Cost analysis of stratospheric exigent weather forecasting,” Quarterly J. Royal Met. Soc., vol. 131,
albedo modification delivery systems,” Env. Res. Lett., vol. 7, 2012. pp. 3037–3051, 2005.
[12] D. W. Keith and H. Dowlatabadi, “A serious look at geoengineering,” [39] D. G. MacMynowski, “Can we control El Niño?” Env. Res. Lett.,
Eos, Trans. AGU, vol. 73, no. 27, 1992. vol. 4, 2009.

2406