Climate model studies in which CO2-induced global warming is offset by engineered decreases of incoming solar radiation are generally robust in their prediction of reduced amounts of global precipitation. While this precipitation response has been explained on the basis of changes in net radiation controlling evaporative processes at the surface, there has been relatively little consideration of the relative role of biogeochemical carbon-cycle interactions. To address this issue, we employ an Earth System Model that includes oceanic and terrestrial carbon components to isolate the impact of biogeochemical carbon coupling on the precipitation response in geoengineering experiments for two types of solar radiation management. We show that carbon coupling is responsible for a large fraction of the global precipitation reduction in such geoengineering experiments and that the primary effect comes from reduced transpiration through the leaves of plants and trees in the terrestrial component of the carbon cycle due to elevated CO2. Our results suggest that biogeochemical interactions are as important as changes in net radiation and that climate models that do not account for such carbon coupling may significantly underestimate precipitation reductions in a geoengineered world.
 Climate model studies in which CO2-induced global warming is offset by decreased incoming solar radiation show reduced amounts of global precipitation [e.g., Govindasamy et al., 2003; Bala et al., 2008; Royal Society, 2009; Schmidt et al., 2012]. While this precipitation response has been explained as the consequence of net radiation changes impacting evaporative processes at the surface [Bala et al., 2008], there has been little consideration of the relative role of biogeochemical carbon-cycle interactions. To address this issue, we have designed a novel set of experiments using the second generation Canadian Earth System Model (CanESM2), which includes interactions between the terrestrial and oceanic carbon cycle components and the physical climate system [Arora et al., 2011], to isolate the impact of such carbon coupling on the precipitation response in equilibrium Solar Radiation Management (SRM) geoengineering experiments. Below, we briefly describe our experiments, leaving the specific model and experimental details to section 2.
 In our experiments SRM, is used to offset the global average surface warming induced by an abrupt doubling or quadrupling of the concentration of atmospheric CO2 relative to a pre-industrial “control” climate. We consider two distinct SRM approaches in this study: (1) “Sunshade” SRM in which space-based reflectors are used to uniformly reduce incident solar radiation [Govindasamy et al., 2003]. (2) “Aerosol” SRM in which sulfate particles injected into the stratosphere produce a global dimming effect by reflecting solar radiation [McCusker et al., 2012]. The application of two distinct SRM approaches in this study allows an investigation of the potential impact of SRM choice, as well as an investigation of the robustness of the impact of biogeochemical interactions on the precipitation response in geoengineering experiments.
 To isolate the impact of biogeochemical interactions, in addition to these “full” simulations, we perform “decoupled” experiments in which the increase of atmospheric CO2 concentration is not communicated to the terrestrial and oceanic carbon cycle components of the model. In this decoupled mode of operation, the response of the model is representative of the generation of climate models that do not include biogeochemical feedbacks associated with the carbon cycle. By differencing the response of our full and carbon-decoupled SRM geoengineering simulations, we obtain an estimate of the impact of biogeochemical interactions, or carbon coupling, on global average precipitation. All of the differences identified in our study are significant above the 99% confidence level.
2 Model and Experiments
 CanESM2 is a comprehensive ESM which includes interactions between the terrestrial and oceanic carbon cycle components and the physical climate system. The atmospheric component of CanESM2 is a spectral model employing T63 triangular truncation with physical tendencies calculated on a 128 × 64 (~2.81°) horizontal linear grid and based on the Canadian Centre for Climate Modelling and Analysis fourth generation atmospheric general circulation model. The physical ocean component of CanESM2 is based on the National Center for Atmospheric Research community ocean model (NCOM1.3) and has 40 levels with approximately 10 m resolution in the upper ocean, and the horizontal resolution is approximately 1.41° (longitude) × 0.94° (latitude). The Canadian Model of Ocean Carbon, the ocean carbon cycle component of CanESM2, incorporates an inorganic chemistry module (solubility pump) and a Nutrients-Phytoplankton-Zooplankton-Detritus ecosystem model (organic and carbonate pumps) for simulating the ocean-atmosphere exchange of CO2 (Zahariev et al. ). The terrestrial carbon cycle processes in CanESM2 are modeled using the Canadian Terrestrial Ecosystem Model (CTEM) which simulates carbon in three live vegetation pools (leaves, stem, and root) and two dead pools (litter and soil organic carbon) for nine plant functional types [Arora and Boer, 2010]. The photosynthesis sub-module of CTEM is based on a biochemical model and uses a single-leaf photosynthesis approach. The coupling between photosynthesis and stomatal conductance in CTEM is based on vapor pressure deficit [Leuning, 1995] and this parameterization governs the response of stomatal conductance to changes in atmospheric CO2 concentration.
 The pre-industrial control simulation uses a specified CO2 concentration of 284.7 ppm. The concentrations in the doubled and quadrupled CO2 simulations are 569.3 and 1138.6 ppm, respectively. In the sunshade SRM case, the global mean warming caused by CO2 is compensated for by reductions of about 1.90% and 4.00% (2xCO2 and 4xCO2) in the solar constant (1365 W/m2) for carbon-coupled simulations and 1.85% and 3.85% (2xCO2 and 4xCO2) for carbon-decoupled simulations. For the aerosols SRM case, stratospheric aerosols, assumed to be 75% H2SO4 and 25% H2O by weight [Toon and Pollack, 1973], are specified with concentrations of 7.18 × 10−6 and 1.56 × 10−5 g/m3 (2xCO2 and 4xCO2) for the carbon-coupled simulations, and 6.87 × 10−6 and 1.51 × 10−5 g/m3 (2xCO2 and 4xCO2) for the carbon-decoupled simulations. The aerosols are specified globally in a vertically uniform 10 km thick layer just above the tropopause with an effective particle radii of 0.35 µm which results in global burdens of 9.0 and 19.6 TgS for the 2xCO2 and 4xCO2 carbon-coupled simulations, and 8.6 and 19.0 TgS for the 2xCO2 and 4xCO2 carbon-decoupled simulations. Additional experiments with particle effective radius of 0.2 and 0.5 µm show basic insensitivity of our main results over this range.
 The reduction in solar constant and addition of stratospheric aerosols, in the sunshade and aerosols SRM cases, yield the global average and time average surface temperature to within ±0.05°C of its pre-industrial value of 13.7°C, offsetting the approximate 3.7°C and 7.4°C warming that is obtained in this model when CO2 is doubled or quadrupled, respectively. The SRM geoengineering simulations are run for several hundred years allowing equilibrated conditions and steady statistical properties to be obtained.
 The percentage change in precipitation over ocean and land is presented in Figure 1 for all simulations in this study. For sunshade SRM, in the absence of carbon coupling, precipitation changes of roughly −1.9% and −3.9% occur over land, while changes of roughly −1.6% and −3.1% occur over ocean for CO2 doubling and quadrupling, respectively. These reductions are consistent with previous estimates based on models without carbon coupling [Lunt et al., 2008; Pongratz et al., 2012]. In the presence of carbon coupling, sunshade SRM shows additional precipitation reductions of roughly 69% and 94% over land, and 18% and 37% over ocean for CO2 doubling and quadrupling, respectively. We note that carbon coupling has a substantial impact on precipitation change over land relative to ocean for both CO2 doubling and quadrupling. To account for such precipitation changes, previous arguments based on radiative effects alone have pointed to the fact that tropospheric warming induced by longwave interactions with CO2 is distributed throughout the full depth of the troposphere, while the reduction in solar heating, designed to cancel this warming, is concentrated at the surface. This asymmetry in the vertical distribution of longwave heating and shortwave cooling means that radiative budgets evaluated at the surface (Figure 2) are necessarily dominated by shortwave cooling, implying that there is less energy available for evaporative processes and a concomitant reduction in global precipitation in the geoengineered simulation [Bala et al., 2008].
 From Figure 1, for aerosol SRM in the absence of carbon coupling, precipitation changes of roughly −1.7% and −3.7% occur over land, while changes of roughly −2.8% and −5.9% occur over ocean for CO2 doubling and quadrupling, respectively. In the presence of carbon coupling, aerosol geoengineering shows additional precipitation reductions of roughly 45% and 84% over land, and 26% and 18% over ocean for CO2 doubling and quadrupling, respectively. Comparing the precipitation reductions in aerosol and sunshade SRM, and the impact of carbon coupling, we conclude that (1) precipitation reductions are greater in aerosol SRM than in sunshade SRM, and this result is robust against carbon coupling. (2) Carbon coupling alone produces a substantial reduction in total precipitation, particularly over land, and this result is robust against SRM approach. Notably, these results suggest that the impact of carbon coupling on the global hydrological cycle can be as large as the well-documented radiative impact, and as large as the impact of employing aerosol rather than sunshade geoengineering.
 To understand why greater precipitation reductions are realized for aerosol SRM relative to sunshade SRM, we may appeal to the same radiative arguments used to understand global precipitation reductions in sunshade SRM. For aerosol SRM, there is an additional radiative effect associated with the absorption, and emission, of outgoing longwave radiation by the aerosols in the lower stratosphere, similar to volcanic aerosols [Robock, 2000]. The longwave radiation re-emitted by aerosols back into the troposphere interacts with CO2 to enhance the warming there. Consequently, to offset the same amount of CO2 forcing, aerosol SRM must induce a larger reduction of incoming solar radiation compared to sunshade. Once again, due to asymmetry in the vertical distribution of longwave heating and shortwave cooling in SRM geoengineering, this greater reduction of incoming radiation in aerosol SRM translates into a greater reduction in the net radiation at the surface. The result is that there is even less energy available at the surface for evaporation, and so precipitation reduction is larger in aerosol compared to sunshade geoengineering. This is quantified in Figure 2 by surface radiation budgets for the simulations performed in this study.
 Relative to the carbon uncoupled runs, different SRM changes (i.e., higher reductions in solar forcing) were required in the carbon-coupled simulations to maintain the same global average surface temperature. This raises the possibility that radiative arguments might also explain the precipitation reductions obtained in the carbon-coupled verses carbon-decoupled simulations in our study. This issue was investigated by an additional set of carbon-decoupled runs that employed the same amount of SRM that was employed in the carbon-coupled runs to offset CO2 quadrupling. These additional simulations are indicated by the black circular symbols in Figures 1–3. While these SRM changes result in a net reduction of radiation at the surface, which accounts for the change obtained in the carbon-coupled simulations (R′ black circles verses adjacent red bars in Figure 2), they account for only a small portion of the precipitation changes realized in the carbon-coupled simulations, particularly over land (black circles verses adjacent red bars in Figure 1). Consequently, radiative arguments do not explain the precipitation reductions induced by carbon coupling in our study.
 To understand why carbon coupling in the SRM geoengineering experiments induces a reduction in global precipitation, we must consider the impact of biogeochemical feedbacks on evaporative processes at the surface. Our first observation is that carbon coupling impacts evaporation primarily over land rather than ocean (Figure 3). Over land, changes in evaporative processes, relative to the pre-industrial control climate, in these experiments may be expressed as:
where E′trans and E′inter, respectively, represent differences in transpiration through, and the evaporation of intercepted precipitation from, the leaves of plants and trees—both of which are directly affected by carbon coupling. It is anticipated then that E′trans and E′inter are central in determining the evaporative response to carbon coupling in our experiments. When carbon coupling is turned on, an increase in CO2 concentration has two direct effects on vegetation, which have opposing influences on evaporation. The first is that vegetation leaf area increases, leading to increased transpiration through E′trans and evaporation of intercepted precipitation through E′inter. The second effect is that stomatal resistance increases, leading to reduced transpiration; this affects E′trans only. Both the increase in leaf area and the increase in stomatal resistance are well known physiological responses of terrestrial vegetation to elevated atmospheric CO2 concentrations [Kergoat et al., 2002; Andrews et al., 2011]. E′soil and E′snow represent differences in evaporation from soil and snow, respectively, and are not directly affected by biogeochemical feedbacks.
 The decomposition of changes in evaporative processes over land, E′land (equation (1)), is also displayed in Figure 3, where contributions to E′snow and E′inter have not been shown due to their small magnitude. E′inter is small in these experiments because the effect of increased leaf area appears to be compensated for by decreased precipitation. Figure 3 indicates that the change in evaporation over land due to carbon coupling is associated with large changes in E′trans and E′soil. Since carbon coupling has no direct impact on soil evaporation, the change in E′soil is primarily a response to precipitation change over land. From Figure 3, we see that the reduction in evaporation from the important E′trans term is roughly seven to eight times larger when carbon coupling is present for CO2 doubling and quadrupling. This suggests that the increase in stomatal resistance has a far greater influence on evaporation than the effect of increases in leaf area [Matthews and Caldeira, 2007] and it is the primary reason for the reduction in global precipitation when carbon coupling is included in our experiments. Plots of the spatial distributions of the critical evaporative transpiration term, E′trans, for carbon-coupled simulations (not shown) indicate that reduction in transpiration, as expected, occurs over regions with vegetation but with higher values in tropical regions [Andrews et al., 2011]. The spatial patterns of reduced evaporative transpiration are also found to be remarkably robust against SRM approach.
 The precipitation response in SRM geoengineering experiments has so far, with few exceptions [e.g., Matthews and Caldeira, 2007], been explained on the basis of net radiation changes controlling evaporative processes at the surface. Such radiative arguments also seem to explain our finding that aerosol SRM induces a larger reduction in global precipitation than sunshade SRM. In analyzing the impact of biogeochemical interactions on the precipitation response in geoengineering experiments, we have found that such biogeochemical processes are as important as radiative processes in explaining precipitation changes.
 While there exist differences in terrestrial carbon cycle components of ESMs [Friedlingstein et al., 2006], our conclusion is expected to be robust given that the vegetation feedback processes involved in the present study represent basic, well-known, physiological responses to changing CO2 concentration [e.g., Sellers et al., 1996; Betts et al., 2007; Kergoat et al., 2002; Andrews et al., 2011; Cao et al., 2012]. This is verified in Figure 4 which shows the surface latent heat flux over land in eight distinct ESM simulations where the radiation code “sees” fixed pre-industrial CO2 but the carbon cycle sees 1% per year CO2 rise. This experiment from Phase Five of the Coupled Model Intercomparison Project (CMIP5) provides a direct estimate of the impact of biogeochemical interactions on evaporative processes at the surface. Across the eight models, the latent heat flux generally shows a decrease with increasing CO2, with CanESM2 (solid blue curve) closely following the model mean (solid black curve).
 We conclude by stressing that there remains a significant uncertainty across models in the reduction in transpiration, and hence the latent heat flux, due to elevated CO2 (as seen in the spread of results shown in Figure 4). This uncertainty is caused by both the uncertainty in stomatal closure and the strength of the CO2 fertilization effect (which increases the leaf area) in response to elevated CO2. Arora et al.  show that the carbon-concentration feedback over land, which is dependent on the strength of the CO2 fertilization effect, varies significantly across the land carbon-cycle components of the CMIP5 models. This said, the important role played by terrestrial biogeochemical processes in affecting the hydrological cycle suggests that the climate impacts of geoengineering are more reliably assessed with ESMs which explicitly consider the interactions between the carbon cycle and the physical climate system.
 We thank George Boer (CCCma), Greg Flato (CCCma), and Neil Swart (University of Victoria) for their insightful comments on early drafts of the paper. We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison, and the WCRP's Working Group on Coupled Modelling for their roles in making available the WCRP CMIP multi-model datasets. Support of this dataset is provided by the Office of Science, U.S. Department of Energy.