Examining vegetation feedbacks on global warming in the Community Earth System Model

Authors


Abstract

[1] Leaves close their stomates in response to increases of CO2. Such a rapid physiological response is included in the land component of comprehensive climate models. However, observational studies have shown that they can further close their stomates as a consequence of “down-regulation,” further reducing canopy conductance. However, they may also increase the area of their leaves, hence increasing their canopy conductance. Changes of canopy conductance change surface ET, a reduction leading to surface warming. A simulation considering these mechanisms of modifying canopy conductance is carried out for the assumption of a doubled atmospheric CO2concentration, using the Community Earth System model. It finds that down-regulation as formulated in previous studies could have as large a warming impact on land temperatures as the standard leaf physiological response. Increases in LAI, if they were to occur, appear to have but a small cooling effect. The reduction of latent cooling in the model is amplified by a reduction of low-level cloud cover, hence enhanced net absorption of solar radiation. Reduction of low level cloudiness appears to be necessary to maintain global radiation balance as reported in a previous study. Over mid to high latitudes, decreases in surface albedo associated with reduced snow cover also contribute to amplifying the warming. The physiological feedbacks of leaf stomates in the simulation increase warming by 0.6 ± 0.2°C over land and 0.3 ± 0.1°C globally, not inconsistent with previous studies. Enhanced interhemispheric temperature differences weaken the southward shift of the ITCZ associated with CO2radiative warming. Regions with relatively high LAI tend to have greater vegetation feedback; but increases in large-scale precipitation may weaken this local warming effect.

1. Introduction

[2] Early climate model studies of global warming addressed the hypothetical question: what would be the climate change for a doubling of atmospheric carbon dioxide (CO2) concentration from its preindustrial values? The steady state increases of temperature (i.e., an equilibrium response of global surface temperature) for this scenario generally has been between 1.5°C and 4.5°C depending on model sensitivity [e.g., Charney et al., 1979; Cubasch et al., 2001]. As it has become apparent that this particular scenario (∼560 ppmv) would be reached by the middle of the current century [Meehl et al., 2007, Figure 10.26], simulations intended to be useful to policy makers have adopted scenarios for atmospheric CO2 that grow to much larger concentrations, e.g., the “business as usual” (i.e., IS92a emission scenario [Leggett et al., 1992]). However, the doubled CO2 concentration scenario has remained popular as a simple test of the impacts of including new model components.

[3] Especially interesting is the question as to what will be the modification of global warming from vegetation feedbacks. Leaves take in carbon dioxide by diffusion through their stomates and then assimilation in their chloroplasts. The balance between CO2 inflow and water vapor outflow determines the leaf's stomatal conductance and the concentration of carbon dioxide internal to the leaf Ci, as described by the following equation. The carbon intake Ic is given by

display math

The diffusion term gc(Ce−Ci) (term a) is given by the product of conductance gc and the difference between leaf external Ce and internal carbon dioxide concentration. The assimilation term R(Ci)Icmax (term b) is given by the product of Icmax, the maximum possible assimilation for Ci → ∞, and a reduction term R(Ci) ≤ 1, depending on Ci. When not radiation limited, Icmax is given by Vcmax, the maximum possible assimilation by enzymes.

[4] If the Ic and Icmax were to remain constant as Ce increases, then Ci in term b would also remain constant making it independent of Ce. Term a would then have a larger Ce but the same Ci so gc would have to be less for Ic to remain constant. Such a decrease of gc for increasing Ce leads to less transpiration, hence a net warming. However, according to the modeled leaf physiology, an increase of Ic implies more Ci and hence a larger gc, weakening somewhat the implied warming. Further weakening would occur if plants grow more leaves in response to the increased CO2 hence increasing gc and if this weakening is enough, even a cooling. Sellers et al. [1996, Figure 1], introduced an additional consideration of down-regulation, i.e., a decrease ofIcmax that reduces Ic more than can be compensated by an increase of Ci so that gc is reduced (i.e., additional warming). In sum, the effects of the stomatal closure implied by more CO2can be weakened by increases in LAI and enhanced by down-regulation. Reduced or increased water stress and changes in cloud cover can further modify the temperature response to these leaf physiological effects. Thus, the sign of the overall effect depends on whether or not other such additional effects can collectively over-compensate for the stomatal closure response to CO2.

[5] How these collective adjustments are realized has been examined by Free-Air CO2 Enrichment experiments (FACE) [Nowak et al., 2004; Ainsworth and Long, 2005] and are generally consistent with the concepts of “optimization” [Dewar, 1996; Haxeltine and Prentice, 1996; Ainsworth and Rogers, 2007; Leakey et al., 2009].

[6] Early model sensitivity tests investigated the climate response to increased stomatal resistance (or decreased stomatal conductance) by uniformly doubling the stomatal resistance in land surface schemes and found a weak warming effect [e.g., Henderson-Sellers et al., 1995; Pollard and Thompson, 1995]. Sellers et al. [1996] and Betts et al. [1997] first examined plant physiological feedbacks to CO2 forcing in global climate models by modeling leaf conductance to respond to doubled atmospheric CO2 concentration and climate changes. These studies likewise found that their reduction of stomatal conductance for doubled CO2 led to less evapotranspiration and consequently a small warming. Sellers et al. [1996]found that their warming of 0.1°C was compensated by an equal cooling with inclusion of down-regulation even though it reduced further canopy conductance and evapotranspiration. Similarly,Betts et al. [1997] found that with their leaf area indexes (LAI) averaged over land increased by 7%, their 0.2°C warming switched to a 0.1°C cooling. More recently, Levis et al. [2000] revisited the question of the impact of dynamic vegetation with their IBIS land model and found a 0.1°C warming without structural (e.g., LAI) adjustments. Their LAI increased by 49% with their interactive vegetation and it induced large regional and seasonal climate changes but those were largely compensating and so did not change global temperature. Other more recent studies have only addressed the impact of standard physiology but have found larger effects than the above mentioned previous studies [e.g., Doutriaux-Boucher et al., 2009] found a 0.4°C [cf. also Boucher et al., 2009; Cao et al., 2010] found a 0.3°C warming (0.5°C over land) from physiology in a doubled CO2 climate. None of these previous studies addressed both the question of the effects of down regulation and changing leaf area and those that included one or the other did not indicate them to be of much importance for global temperature.

[7] Bounoua et al. [2010], however, included both factors and indicated both had substantial effect on a doubled CO2 climate, i.e., a global temperature decrease of 0.1°C associated with Vcmaxdown-regulation with fixed vegetation structure and a 0.3°C cooling (0.6°C cooling over land) with adjustment of LAI. Much larger regional effects, e.g., more than a 1.5°C cooling over the eastern U.S., were also found. The temperature response from the stomatal changes for standard physiology with increased CO2was included in their control and so not examined in their analysis. Their study based its formulation of down-regulation on that ofSellers et al. [1996]with both studies using the Simple Biosphere Model (SiB2) coupled with the Colorado State University climate model (hereafter SiB2-GCM) [Randall et al., 1996]. This model has a low climate sensitivity (1.94°C) and they suggested that the negative feedbacks they found could be stronger in a model with larger sensitivity. Based on early observational studies [Field et al., 1992; Tissue et al., 1993], Sellers et al. [1996] suggested that the maximum Vcmaxdown-regulation effect could reduce the carbon assimilation rate to the 1 × CO2 level. However, recent studies from FACE have suggested that photosynthetic capacity generally increases for C3 plants with elevated CO2 concentrations [Leakey et al., 2009] so that the assumption that the photosynthesis rate would remain constant under elevated CO2 concentrations is not necessarily realistic but rather provides an upper limit for Vcmaxdown-regulation.

[8] This paper reexamines the vegetation feedbacks described above for the climate change of a doubled atmospheric CO2 concentration using the Community Earth System model (CESM) (i.e., a climate sensitivity of about 3.2°C for CAM4 [Bitz et al., 2012]). Its land component is the Community Land Model version 4 (CLM4) [Oleson et al., 2010; Lawrence et al., 2012], is one of the most scientifically advanced such models. Can it confirm the negative feedbacks found by Bounoua et al. [2010]for down-regulation and increases of LAI?

[9] Both the CLM4 and SiB2 models use the Ball-Berry leaf conductance model [Collatz et al., 1991] that relates leaf stomatal conductance to atmospheric CO2 concentration, plant carbon assimilation with the latter modeled as indicated by equation (1) but with additional environmental factors, e.g., relative humidity at the leaf surface. Increases in atmospheric CO2 concentration or decreases in carbon assimilation reduce leaf conductance. The responses of vegetation foliage to increased CO2 and changed climate can by simulated either interactively by coupling dynamic vegetation models synchronously [e.g., Levis et al., 2000] or iteratively [e.g., Betts et al., 1997] to a climate model, or be a controlled approach that modifies prescribed foliage, e.g., Bounoua et al. [2010], who scaled up the Fraction of Photosynthetically Active Radiation (FPAR). This study follows the concept of the latter approach because of its simplicity and transparency but is unable to follow it in detail because of different model structures.

[10] The following section introduces the model and simulation design. Validation of the control simulations is presented in section 3, while section 4presents the results. Results are compared with previous SiB2-GCM simulations insection 5. Conclusions are summarized in section 6.

2. Methodology

[11] The CESM version 1.0.2 is an atmosphere-land-ocean coupled climate model that was developed at the National Center for Atmospheric Research. Besides CLM4, it also has an atmospheric component, i.e., the Community Atmosphere Model version 4 (CAM4) [Neale et al., 2010a; also The mean climate of the Community Atmosphere Model (CAM4) in forced SST and coupled experiments, submitted to Journal of Climate, 2012], a dynamic ocean model, i.e., Parallel Ocean Program version 2 (POP2) [Smith et al., 2010], and a sea ice model, i.e., the Los Alamos sea ice model version 4 (CICE4) [Hunke and Lipscomb, 2010]. For computational economy, we use its option for a slab ocean model (SOM) [Bitz et al., 2012] which predicts SST based on surface energy balance. This SOM takes only 15–20 years to reach an equilibrium state, much less time than that used by the dynamic ocean model. Such a slab model provides a good estimate of the equilibrium climate sensitivity [Danabasoglu and Gent, 2009]. We use the CAM4 instead of CAM5 [Neale et al., 2010b], which is also available in the CESM1.0.2, so the model framework is comparable to the Community Climate System Model version 4 (CCSM4) [Gent et al., 2011].

[12] Five simulations were carried out as listed in Table 1. The simulations are labeled following Bounoua et al. [2010] although somewhat different assumptions were necessary for the RPVB case because of different model structure. The control simulation (CTL) fixed CO2concentrations at their present-day (year 2000) level of 367 ppmv. The forcing data for the SOM was derived from a fully coupled simulation driven by the radiative forcing of the 1990s. The average solar irradiance was decreased by 12 W m−2 to correct a warm bias in the model global temperature. The R simulation doubled the CO2 concentration in the CAM4 atmosphere model while keeping fixed its concentration for the physiological scheme in the land model. The R test was designed to isolate the impact of CO2 radiative forcing alone as also done by Sellers et al. [1996] and Betts et al. [1997]. In the RP simulation, the CO2 concentration was also doubled to 734 ppmv in the CLM4 land model. Therefore, RP simulated both radiative forcing from the CO2 and feedbacks associated with its reduction of leaf stomatal conductance.

Table 1. List of Simulationsa
NameCO2 ConcentrationVcmaxLAI
  • a

    A dash indicates no modification is applied. R stands for radiative effect alone, P for physiological effects directly related to enhanced CO2 concentration, V for Vcmaxdown-regulation, while B for biophysical effects related to increased LAI.

CTL1 × CO2--
R2 × CO2 for the CAM; 1 × CO2 for the CLM--
RP2 × CO2--
RPV2 × CO2Down-regulated-
RPVB2 × CO2Down-regulatedIncreased

[13] An effect of down-regulation was added in the RPV simulation formulated similar to that used bySellers et al. [1996] and Bounoua et al. [2010], i.e., the Vcmax was reduced by the ratio of gross primary production (GPP) in the control simulation to that of the doubled CO2 simulation (GPP_CTL/GPP_RP). The reduction was applied to Vcmaxat the plant functional type (PFT) sub-grid level. As shown inequation (1), Vcmax is to the first order proportional to GPP. Applying this ratio is intended to reduce the carbon assimilation rate to that of the 1 × CO2 level. However, plant photosynthesis is also limited by radiation and environmental variables, so GPP does not exactly decrease to that of CTL.

[14] The RPVB simulation down-regulatedVcmax while increased LAI by the ratio of GPP_RP to GPP_CTL. Bounoua et al. [2010] suggested that in a doubled CO2 climate, more leaf area would grow in response to enhanced CO2 concentration, CO2-induced climate changes, and increased water availability due toVcmaxdown-regulation, but they scaled FPAR rather than LAI, not possible here because of different model structure. Since FPAR is linearly related to GPP, increasing FPAR by this ratio compensated the carbon loss due toVcmaxdown-regulation and the overall GPP in their RPVB simulation was very close to that of the RP simulation. Here we directly increased LAI. In CLM4, gross photosynthesis is calculated as following:

display math

where Asun, Asha are photosynthesis of sunlit and shaded leaves, Lsun and Lsha are sunlit and shaded leaf area indices, fsun = (1 − eKL)/KL and fsha = 1 − fsun are sunlit and shaded fraction of canopy, L is LAI, K is optical depth of direct beam per unit leaf and stem area [Oleson et al., 2010]. As shown in equation (2), LAI is nonlinearly related to GPP, i.e., when LAI is increased fsha increases and fsun decreases. Thus scaling up LAI would not increase GPP to that of RP but somewhat less (more discussion in the result section). A few more tests were also conducted to examine the sensitivity of overall vegetation feedbacks to the magnitude and location of LAI increases (discussed in section 4.2). A constraint was also applied to the LAI, i.e., LAI only increased over regions where the soil wetness, i.e., the soil water stress factor BTRAN in the CLM4, was equal (when lack of water stress, i.e., BTRAN ≥ 0.99) or greater in the RP simulation than in the CTL. This constraint mimicked limitation of vegetation growth by soil water stress [Bounoua et al., 2010]. If the ratio of soil wetness in the RP to that of the CTL (BTRAN_RP/BTRAN_CTL) was greater than the ratio of GPP_RP to GPP_CTL, LAI was further increased by the ratio of BTRAN_RP to BTRAN_CTL to simulate an increase of LAI due to enhanced water use efficiency. In CLM4, changes of LAI modified snow-free albedo and roughness length and other LAI-dependent variables and thus changed vegetation feedbacks.

[15] The different simulations developed here illustrate the possible contributions by plants to future climate change. The assumed changes in their physiology and structure are not rigorously informed from observations but are plausible scenarios consistent with known mechanisms and are intended to assess the consequences of the assumptions made.

[16] Medium resolution grids, about 1.9° × 2.5° for the land and atmosphere models and 1° × 1° for the slab ocean model, were used. (Sellers et al. [1996] and Bounoua et al. [2010] used 7.2° by 9° horizontally.) Precipitation simulations are usually improved with a higher model resolution that resolves topography better. All the simulations started from the same initial conditions and were carried out for 35 years. In the RP simulation, the model reached equilibrium in about 20 years. The model climatologies were calculated by averaging the simulation output over its last 10 years, i.e., years 26–35. All results shown in tables and figures are annual averages.

3. Evaluation of Modeled Control Climate

[17] Table 2 compares the 2m air temperature and precipitation over land simulated by the control run with those observed as given by the CRU TS3.1 data [Mitchell and Jones, 2005]. The results from Sellers et al. [1996] and Bounoua et al. [2010] are also listed for reference. Definitions and names of regions follow Sellers et al. [1996]. “All land” in this table refers only to land located to the north of 60°S, i.e., Antarctica is omitted, in order to compare the results of the control run with the CRU observations, while elsewhere in the paper it means over all land grids.

Table 2. Annual Mean Surface Air Temperature (°C) and Precipitation (mm day−1)a
 CTLCRUSellers96 (B2010)
Surface Air Temperature
Tropics (14.4°S–14.4°N)25.1225.6728.1
Mid-latitudes (28.8°–50.4°N)11.2710.6217.4
North latitudes (50.4°–72.0°N)−4.02−3.254.8
All land13.8113.8919.6 (19.55)
 
Precipitation
Tropics4.794.794.36
Mid-latitudes1.771.462.70
North latitudes1.691.412.35
All land2.402.292.90 (2.75)

[18] The present-day climatology is adequately reproduced by our control simulation with only small biases, i.e., a 0.08°C cold bias of land surface temperature and a 0.11 mm day−1 excess precipitation bias (Table 2). Evapotranspiration (ET) connects the terrestrial ecosystem and the atmosphere. Our control run simulated an annual mean global ET of 1.515 mm day−1 or about 80 × 103 km3 yr−1, higher than the observationally based estimate 65 × 103 km3 yr−1 for 1982–2008 [Jung et al., 2010], an overestimation also indicated by Lawrence et al. [2011]. This overestimation is over low to mid latitudes but at high latitudes ET is underestimated in comparison with Zhang et al. [2009, 2010]. Of the total ET, about 43% is contributed from the canopy transpiration, 41% from ground evaporation and about 16% from canopy evaporation. The partitioning is close to a multimodel mean of 48% from transpiration, 36% from ground evaporation and 16% from canopy evaporation [Dirmeyer et al., 2006].

[19] GPP in CLM4 is biased high due to some structure errors [Bonan et al., 2011]. In this study, when CLM4 was coupled with CAM4, its positive bias for GPP is smaller than in off-line simulations (146 versus 165 Pg C yr−1 while estimation from the FLUXNET diagnostic model ensemble mean for 1998–2005 is 123 ± 8 Pg C yr−1 [Beer et al., 2010]). A more reliable prediction of GPP would need to correct these errors. Since here we mainly focus on vegetation climate feedbacks associated with changes of vegetation physiology and LAI, these biases in the base physiology are likely to be of little significance.

4. Results

4.1. Global and Regional Climate Feedbacks

[20] Figure 1 shows the differences of exposed LAI (ELAI; LAI that is not covered by snow) between the RPVB and RPV simulations. Following Bounoua et al. [2010], LAI was increased, provided soil wetness was also increasing (black thick contour). ELAI does not change or changes little (due to changes of snow cover) over regions where soil is dryer in the doubled CO2 climate. Changes in soil water stress are closely related to (but not limited by) precipitation variations. In the RP simulation, annual mean precipitation decreases over Mexico, northeastern Amazon, southern Africa, and southwestern Europe. These variations are similar to those given by the Intergovernmental Panel on Climate Change (IPCC) AR4 [Christensen et al., 2007] multimodel projection for 2080–2099 under A1B emission scenario for these regions (their Figures.11.2, 11.5, 11.12, and 11.15). Over regions where the soil water stress is not limiting leaf growth (i.e., BTRAN_RP ≥ 0.99), e.g., Peru and Bolivia, western tropical Africa, southeastern South Africa, and southern Indonesia, ELAI increases. Tropical ELAI increases most over eastern Africa (∼75%), southwestern Amazon (∼50%), and southeastern Brazil (∼70%). Relatively large increases also occur over northern latitudes (i.e., northwestern China, northern Siberia, and northern Canada). The global land average ELAI increases by about 30% (0.37 m2/m2). This magnitude of change is comparable to the range of LAI changes simulated by a dynamic vegetation model for the mid of the 21st century to the end of the 21st century (21%–36% [Jiang et al., 2011]) but smaller than the 49% increase given by another dynamic vegetation model under a doubled CO2 climate [Levis et al., 2000].

Figure 1.

Differences of exposed leaf area index (ELAI; m2/m2) between the RPVB and RPV simulations. Thick black contour indicates the zero line of soil wetness differences between the RP and CTL simulations. Shading intervals are 0.2 m2/m2, except the first two which are 0.001–0.1 and 0.1–0.2 m2/m2.

[21] The pattern of LAI changes shares some similarities with those of Bounoua et al. [2010]. For instance, both simulations found large increases of LAI over the Amazon, tropical Africa, Southeast Asia, southwestern Canada, northeastern U.S., northern Europe, and central Russia. The magnitude of LAI increase is less than the 58% increase reported by Bounoua et al. [2010], in part because they adjusted FPAR rather than LAI directly as well as due to different climate (e.g., precipitation) and physiological (e.g., GPP) responses in the RP simulation. Later analysis (section 4.2) shows that the sign of overall vegetation feedback is not very sensitive to the LAI changes determined here.

[22] The increases in LAI do not bring GPP back to its value in the RP simulation; rather it is less by 8.6 Pg C yr−1. When the soil wetness constraint is removed, the difference is reduced to −2.6 Pg C yr−1. This difference is due to changes of other environmental variables between RP and RPVB simulations and the fact that LAI is not linearly related to GPP.

[23] Figures 2a–2d show the differences of annual mean 2m land air temperature for the R minus CTL, RP minus R, RPV minus R, and RPVB minus R simulations, respectively. Results are summarized by means and standard deviations over global land (i.e., as derived from individual annual averages). With only changing CO2 radiative forcing, land surface temperature increases by 3.33 ± 0.16°C with relatively large warming over mid to high latitudes (Figure 2a), while decreases in stomatal conductance associated with enhanced CO2concentration and changed climate further increase land temperature by 0.27 ± 0.15°C. Warming occurred over the tropics and northern hemisphere high-latitude forest areas at the 95% confidence level (Figure 2b). Temperature over land is increased by 0.59 ± 0.18°C from the standard physiology and down-regulation and with the assumed increase in LAI, it is somewhat less such that the average change over land including all terms is 0.48 ± 0.19°C (about 14 ± 7% of that from CO2 radiative forcing).

Figure 2.

Surface 2m air temperature anomalies (°C) for the (a) R minus CTL, (b) RP minus R, (c) RPV minus R, and (d) RPVB minus R simulations. Differences significant at the 95% confidence level (t-test) are masked with dots. “Mean” indicates means of global land surface temperature anomalies and ranges of temperature variations as given by ±one standard deviation for annual samples.

[24] This warming from vegetation changes involves several processes. Canopy transpiration is a dominant component of ET over vegetated areas. Changes in plant physiological properties perturb canopy transpiration and ET, and thus modulate the land-atmosphere interactions through reshaping surface heat budget and hydrological cycles. To better understand the climate responseFigures 3a–3c show zonal means of ET and its components for the RP minus R, RPV minus RP, and RPVB minus the RP simulations, respectively.

Figure 3.

Zonal mean evapotranspiration (ET) and its components (canopy transpiration, canopy evaporation, and ground evaporation; mm day−1) for the (a) RP-R, (b) RPV-RP, and (c) RPVB-RP simulations.

[25] Reduction in leaf stomatal conductance from increased CO2 decreases canopy transpiration (green line), dominating the changes in ET (black line), with greatest reduction over regions with highest LAI, e.g., forest in the tropics and midlatitude (Figure 3a), a result consistent with previous simulations [Pollard and Thompson, 1995; Henderson-Sellers et al., 1995; Martin et al., 1999]. Ground evaporation (orange line) increases in association with increased humidity differences between the canopy air and ground surface, while variations in the canopy evaporation (blue line) are mainly associated with changes in precipitation.

[26] Canopy transpiration in the RPV simulation (Figure 3b) with down-regulated gross photosynthesis has less ET than the RP simulation. Its maximum negative anomalies are over the tropics where magnitudes are about half as large as the direct stomatal response to increased CO2. In the RPVB simulation (Figure 3c), these reductions in transpiration and ET are weakened as a consequence of the increased LAI.

[27] Surface heat balances are also informative. Figures 4a–4c show the differences of zonal mean land surface heating terms, net solar (positive downward; red), net longwave (positive downward; blue), sensible heat (positive upward; black), and latent heat (positive upward; green) fluxes, and surface 2m air temperatures (gray) for the RP minus R, RPV minus RP, and RPVB minus RP simulations, respectively. As shown in Figure 4a, consistent with ET reduction associated with stomatal closure (Figure 3a), latent heat fluxes are reduced, with their greatest decrease over the tropics, while net solar fluxes are enhanced over the tropics and mid to high latitudes. When Vcmaxis down-regulated (Figure 4b), latent cooling is further reduced and solar heating is slightly increased. Decreases in latent heat fluxes are weakened when LAI is increased, while net solar fluxes are enhanced over mid to high latitudes (Figure 4c). Over the Northern Hemisphere subtropics, enhanced downward longwave fluxes also contribute to increasing the surface temperatures. Figure 4 demonstrates that surface temperature increases associated with vegetation feedbacks are mainly a response to reduced latent fluxes and enhanced net solar fluxes. The former dominate over the tropics while the latter are more important over the high latitudes.

Figure 4.

Zonal mean land surface heat fluxes (W m−2) read on left and 2m air temperature (°C) read on right for the (a) RP-R, (b) RPV-RP, and (c) RPVB-RP simulations. Net solar (positive downward), net longwave (positive downward), sensible heat (positive upward), and latent heat (positive upward) fluxes and air temperatures are denoted by red, blue, black, green, and gray lines, respectively.

[28] Four regions (see Figure 2c) with relatively large responses of surface temperature to plant physiological changes are chosen to further understand regional temperature variations. These are the Amazon (10°S-5°N, 55°W-75°W), Congo Basin (10°S-5°N, 15°E-35°E), eastern U.S. (30°N-50°N, 75°W-90°W), and Siberia (50°N-70°N, 80°E-140°E).Figure 5 shows annual mean 2m air temperature (T), net solar fluxes (SWnet), net longwave radiative fluxes (LWnet), latent heat fluxes (LH), and sensible heat fluxes (SH) averaged in the four regions for the RP minus R (gray), the RPV minus RP (black), and the RPVB minus RP simulations (blue).

Figure 5.

Differences of regional averaged surface air temperatures (T; °C) and surface heat fluxes (W m−2): net solar fluxes (SWnet; positive downward), net longwave radiative fluxes (LWnet; positive downward), latent heat fluxes (LH; positive upward), and sensible heat fluxes (SH; positive upward), for the RP minus R (gray) and RPV minus RP (black), and RPVB minus RP (light blue) simulations.

[29] Vegetation physiological and LAI changes amplify the impact of CO2 radiative forcing in all regions (Figure 5), i.e., over the Amazon by about 0.8°C (a 35% increase), over the Congo basin by 0.2°C (7%), over the eastern U.S. by 1.1°C (35%), and over Siberia by 1.4°C (34%). Reduced latent heat flux is one of the primary contributors to this warming, and is affected most by the direct stomatal response to the elevated CO2(RP-R).

[30] Changes in net solar fluxes also contribute to the enhanced warming. These depend on changes in cloud and precipitation which are more region-dependent. In the RP simulation, enhanced net solar flux is the largest contributor to the amplified warming over the Amazon, eastern U.S., and Siberia. It becomes a secondary contributor to the warming over the Amazon and eastern U.S. in response toVcmaxdown-regulation and contributes little over Siberia. Over the Congo basin, net solar flux switches from cooling (negative) to a weak warming effect (small positive values). With increased LAI, it again becomes an important contributor to the warming in each region.

[31] To understand the changes associated with net solar fluxes, variations in cloud, surface albedo, and precipitation in each region are examined. Figure 6shows the percentage changes in total cloud, cloud at low, middle, and high levels, surface albedo, and precipitation for the RP minus R, RPV minus RP, and RPVB minus RP simulations. Low-level cloud cover decreases in all four regions in the RP simulation in association with reduced ET (Figure 5) that decreases the wetness in the boundary layer. Doutriaux-Boucher et al. [2009] and Andrews et al. [2011] also found a rapid reduction (within 5 years) of low level cloudiness in association with stomatal closure due to enhanced CO2 concentration. Our results show that this low cloud reduction persists to steady state. In the RPV and RPVB simulations, as a consequence of the ET reduction, low clouds are further reduced except over Siberia, where cloud cover increases at all levels associated with its precipitation increase.

Figure 6.

Changes in total cloud (cldtot), low-level cloud (cldlow), mid-level cloud (cldmed), high-level cloud (cldhgh), precipitation (Precip), and surface albedo (in percentage) for each region.

[32] Changes in surface albedo in the RP and RPV simulations are quite small over the Amazon and Congo basin and are mainly related to the dependence of ground albedo on soil moisture as LAI is not changed in these scenarios. Over the eastern U.S. and Siberia, surface albedo is reduced as snow cover decreases. In the RPVB simulation, surface albedo over the Amazon and Congo basin slightly increases, likely due to enhanced albedo in the near-infrared band with more LAI [Mei and Wang, 2010]. Over the eastern U.S., albedo deceases associated with reduced snow cover and increased LAI, while over Siberia albedo is slightly enhanced from increased snow cover.

[33] Figure 6suggests that the enhanced net solar fluxes over the Amazon and Congo basin are mainly due to reduced low-level cloud cover, while over the eastern U.S. and Siberia both reduced low-level cloud cover and lower surface albedo contribute to enhancing surface solar fluxes. However, an increase of precipitation associated with enhanced cloud cover may weaken the connection between low cloud cover and ET perturbation. For instance, over the Congo basin in the RP simulation and over Siberia in the RPV simulation net solar fluxes decrease or change little due to enhanced cloud cover.

[34] Figures 7a–7d show the June–July–August (JJA) precipitation anomalies for the R minus CTL, RP minus R, RPV minus R, and RPVB minus R simulations. The CO2 radiative forcing alone increases global precipitation by 0.16 ± 0.02 mm day−1. The ITCZs over the Pacific, Atlantic and Indian Oceans are shifted southward (Figure 7a) perhaps associated with reduced interhemispheric temperature differences (NH minus SH) as these ITCZs tend to be displaced to the warmer hemisphere [Broccoli et al., 2006]. A recent study found 4 out of 9 slab ocean model simulations from the WCRP CMIP3 data set also simulated a southward shift of ITCZ in a doubled CO2 climate while the other 5 simulated a northward shift [Frierson and Hwang, 2012], indicating that model estimate of tropical precipitation change still contain large uncertainties, as also revealed by the IPCC AR4 ensemble simulations [Meehl et al., 2007, Figure 10.9].Vegetation feedbacks slightly reduce summer rainfall globally but increase it over land. The southward shift of the ITCZs in JJA is partially offset (Figures 7b–7d) as vegetation feedbacks enhance CO2 radiative warming more over the Northern Hemisphere than over the Southern Hemisphere (Figures 2b–2d). Such a change of the ITCZs is also found in the annual mean field but is weaker.

Figure 7.

Precipitation anomalies (mm day−1) for the (a) R-CTL, (b) RP-R, (c) RPV-R, and (d) RPVB-R simulations averaged over June–July–August (JJA). Differences significant at the 95% confidence level (t-test) are masked with dots. “Mean” indicates global means of precipitation anomalies and ranges of variations as given by ±one standard deviation.

4.2. Sensitivity Tests

[35] Soil water stress was used here in the RPVB simulation and by Bounoua et al. [2010] as a constraint on increases of LAI. However, water availability is not the only limitation for LAI to increase. For instance, the growth of the Amazon rain forest is also constrained by solar radiation [e.g., Myneni et al., 2007] and by limited phosphorus [e.g., Davidson et al., 2004]. Also, when using the ratio of GPP_RP/GPP_CTL to infer an increase of the LAI, the influence of soil water stress has already been considered since both the CLM4 and SiB2 models use soil water stress for the calculation of Vcmax, and thus for GPP. Another assumption in the simulations is that increases in LAI are proportional to decreases in Vcmax. This assumption is quite idealized. These so inferred increases in LAI enhance the canopy transpiration, tending to counteract the warming associated with stomatal closure related to enhanced CO2 concentration and Vcmaxdown-regulation. How sensitive are the total vegetation feedbacks on climate to the amount of LAI increase?

[36] Four more simulations (a–d) were carried out to examine this issue. These were the same as the RPVB simulation except that in (a) the soil water stress constraint was removed (RPVBa) and in (b) the ratio to increase LAI was doubled (2 × GPP_RP/GPP_CTL) to test an extreme condition for LAI variations (RPVBb). As previously mentioned, increases in LAI are less than those by Bounoua et al. [2010]; thus two additional simulations that more closely mimic their LAI increase in terms of magnitude and methodology were tested, respectively, i.e., (c) RPVBc, which omitted the soil wetness constraint and uniformly increased LAI by 50%, and (d) RPVBd, in which the LAI increase was determined by an increase of FPAR using the simplified relationship FPAR = 1-exp(−0.5 × LAI), while the ratio GPP_RP/GPP_CTL was applied to FPAR.

[37] We found that removal of the soil water constraint did not change the direction of global temperature variation in the simulation (warming by 0.42°C for RPVBa-R compared to 0.48°C with the soil water stress constraint) suggesting that the warming is quite a robust conclusion. Only with an extreme increase of LAI (∼134%) in RPVBb was the land surface temperature anomaly (RPVBb-RP) negative and only by a small amount (−0.04°C), much less than the −0.57°C cooling reported byBounoua et al. [2010]. The ELAI increased by 50% and 76% for the RPVBc and RPVBd tests, respectively, and surface temperature was warmer than in the RP simulation by 0.18°C (0.17°C) over land and 0.14°C (0.14°C) globally for the RPVBc (RPVBd) test. Evidently, the overall vegetation feedback in the CESM was not strongly sensitive to the increases of LAI.

5. Discussion

[38] Table 3 summarizes the results for the CESM with those of Bounoua et al. [2010] for the simulations in common listed for references. The temperature increase determined here for standard physiology (0.27°C over land and 0.17°C globally) was a somewhat larger than that found in earlier studies, e.g., 0.1–0.2°C globally [Betts et al., 1997; Sellers et al., 1996; Levis et al., 2000], but is smaller than the more recent estimates of 0.3–0.4°C globally by Doutriaux-Boucher et al. [2009] and Cao et al. [2010] These global differences may be within sampling error. Figure 2suggests the global temperature estimates determined here are uncertain by at least 0.1°C and regional temperatures considerably more because of statistical sampling. Our ten-year averaging is comparable to that employed by previous studies but is somewhat marginal for judging temperature changes of order a few tenths of a degree unless they are physically consistent, e.g., with ET changes as we believe is the case for our results, at least those for land and global average. TheVcmaxdown-regulation increases land global average temperatures by an additional 0.32 ± 0.23°C in comparison with largest changes from those radiatively driven over the Amazon, Congo basin, southern South America, northern and eastern U.S., and northern hemisphere high latitudes (Figure 2c). Increases in LAI have two opposite feedbacks to climate. Increases in canopy transpiration enhance the overall ET and thus increase surface latent cooling, but they also may reduce the surface albedo over regions covered by snow or with sparse vegetation and so increase surface heating. Their overall effect depends on the balance between these two effects. Increases in LAI reduced land surface temperature by 0.1°C, a relatively small amount compared to that of most of the other studies. Land precipitation changed little in the CESM simulation from the Vcmaxdown-regulation but was enhanced by the LAI increase. The CESM simulated a slight increase of surface runoff associated withVcmaxdown-regulation while its increase in LAI reduced surface runoff (Table 3).

Table 3. Summary of Results From the CESM Simulations and Bounoua et al. [2010] (in Parentheses)
 CTLRP-CTLRPV-RPRPVB-RPV
Global Land and Ocean
Global surface 2m air temperature (°C)14.37 (18.53)3.11 (1.94)0.18 (−0.10)−0.04 (−0.16)
Global precipitation (mm day−1)2.85 (2.88)0.16 (0.14)−0.01 (−0.03)0.01 (0.01)
 
Land
GPP (Pg C yr−1)145.6 (124.8)49.4 (44.6)−31.0 (−31.2)22.4 (31)
LAI (m2 m−2)1.22--30% (58%)
2m air Temperature (°C)9.77 (19.55)3.61 (2.80)0.32 (−0.13)−0.11 (−0.44)
Precipitation (mm day−1)2.40 (2.75)0.17 (0.18)−0.00 (−0.09)0.01 (0.03)
Surface Runoff (mm day−1)0.37 (1.15)0.02 (0.11)0.01 (−0.04)−0.01 (−0.03)

[39] Table 4 shows global and regional changes of ET and surface 2m air temperatures associated with Vcmaxdown-regulation simulated (RPV minus RP) by the CESM and for reference those simulated byBounoua et al. [2010] and Sellers et al. [1996], both of which used the SiB2 model coupled to Colorado State University model [Randall et al., 1996] with the same resolution and climate sensitivity. The RPV simulations show a decreased ET over land in the CESM was about the same as in the SiB2-GCM, −2.7% versus −3% (Table 4). However, the ET of the CESM simulation decreased more in the tropics than over mid to high latitudes, while such meridional differences were weaker in the SiB2-GCM.

Table 4. Annual Mean Anomalies of ET (in Percentage) and Surface Air Temperatures (SAT; °C) for the RPV Minus RP Simulationsa
 ET (RPV-RP)SAT (RPV-RP)
Tropics (14.4°S-14.4°N)−3.7 (−2.5)0.2 (0.4)
Mid-latitudes (28.8°–50.4°N)−2.4 (−2.1)0.5 (−0.3)
North latitudes (50.4°–72.0°N)−0.7 (−2.3)0.8 (−0.6)
All land−2.7 (−3.0)0.3 (−0.1, −0.1)
Global−0.2 (−0.8)0.2 (−0.1, −0.1)

[40] Over the tropics, this study found a 0.2°C warming associated with the down-regulation ofVcmax and over mid to high latitudes a warming by about 0.5–0.8°C consistent with its reduced ET at all latitudes (Figure 3). The SiB2-GCM also simulated decreases of ET at all latitudes but its simulated temperatures in mid to high latitudes decreased in the range −0.3 to −0.6°C, dominating the global mean [Sellers et al., 1996]. By contrast, the warming of CESM over mid to high latitude was greater than that over the tropics as amplified by albedo feedback, and consistent with previous simulations [Levis et al., 2000; Notaro et al., 2007; O'ishi et al., 2009] that also demonstrated warming associated with stomatal closure greatest at high latitudes. The SiB2-GCM strong high latitude cooling from down-regulation might be judged a sampling issue except that it appears to have been reproduced by theBounoua et al. [2010] study.

[41] Although the reduction of ET is clearly likely to warm land, the mechanism for additional oceanic warming in the current and previous simulations is less obvious. Generally shifts in global average temperature are associated with changes in global top of the atmosphere radiative balances. Doutriaux-Boucher et al. [2009] ascribed the global temperature response to a reduction in low clouds as also seen in our Figure 6.

6. Conclusions

[42] This paper has demonstrated again that stomatal closure makes an important contribution to the global warming resulting from increases of CO2. It has also demonstrated for the first time that the down-regulation of CO2 physiology could add as much warming over land as that implied by standard leaf physiology. Possible changes in leaf area can weaken this warming, but their effect, as computed here, is much smaller than that reported by Bounoua et al. [2010] and so possibly is strongly dependent on climate model details.

[43] The largest reductions in ET are over tropical and midlatitudes forest, offset somewhat by increases in LAI. The surface heat balance is reshaped with reduced latent cooling and enhanced net solar heating. The former is dominant over the tropics while the latter is the major contributor to the warming at mid to high latitudes. Reduced ET decreases low-level cloud while reduced snow cover decreases surface albedo, both of which contribute to enhancing net solar fluxes.

[44] Overall, our simulations find that vegetation feedbacks increase surface temperature in CESM by 0.6° ± 0.2°C (0.3 ± 0.1°C) over land (globally), or about 14% (11%) of the modeled CO2 radiative warming. This warming could be weakened by 0.1°C from LAI adjustment, or even more if large increases in LAI were to happen. Regional effects are greater. For instance, over the Amazon, eastern U.S. and Siberia, the CO2 radiative warming is enhanced by 0.8°C, 1.1°C, and 1.4°C, respectively.

[45] Vegetation feedbacks also significantly modify interhemispheric temperature differences and thus affect global circulation and precipitation. The southward shift of the ITCZ associated with CO2radiative warming is weakened as vegetation feedbacks increase surface temperature more over the Northern Hemisphere than over the Southern Hemisphere. Changes in large-scale precipitation in turn may weaken (e.g., in the Congo basin) or amplify (e.g., in the Amazon) vegetation feedbacks by modulating cloud cover at mid and high levels, which could counteract or reinforce the low-level cloud cover variations associated with ET perturbation and thus weaken or enhance the secondary warming effect associated with variations in surface net solar fluxes.

[46] Comparison with previous simulations found that the warming effect associated with stomatal closure from increased CO2 concentration and the cooling effect from LAI increase are consistent and within the range of previous studies, while the warming associated with Vcmaxdown-regulation is much stronger than indicated by previous SiB2-GCM simulations [Sellers et al., 1996; Bounoua et al., 2010]. Since this comparison indicates a strong dependence of vegetation feedbacks on processes in the atmospheric component of a climate model and a significant enhancement of global warming, further intercomparison of the vegetation feedbacks examined here with other land models using agreed upon test scenarios (a VEGMIP) appears warranted.

Acknowledgments

[47] This work is supported by the grant DE-SC0002246 from the U.S. Department of Energy. M. Shaikh's help on model testing is gratefully appreciated. The Texas Advanced Computing Center (TACC) at University of Texas at Austin provided high performance computing resources for the work. Valuable comments from editors and three anonymous reviewers improved the paper.