Climate and climate sensitivity to changing CO2 ... - Wiley Online Library

6 downloads 253 Views 2MB Size Report
Dec 13, 2014 - partial radiation perturbation method shows that amplified warming over idealized ... Counterintuitively, the transient response is not primarily related to .... schemes, the standard ECHAM6 convection scheme developed by Nordeng [1994] is ... In RCE, the boundary conditions are traditionally specified to be ...
PUBLICATIONS Journal of Advances in Modeling Earth Systems RESEARCH ARTICLE 10.1002/2014MS000369 Key Points:  A low surface heat capacity couples troposphere to diurnal maximum temperature  Land-ocean warming contrast is reproducible by restricting latent heat release  Bottom-up drying exhausts lapse rate feedback prior to water vapor feedback

Correspondence to: T. Becker, [email protected]

Citation: Becker, T., and B. Stevens (2014), Climate and climate sensitivity to changing CO2 on an idealized land planet, J. Adv. Model. Earth Syst., 6, 1205–1223, doi:10.1002/ 2014MS000369. Received 27 JUL 2014 Accepted 6 NOV 2014 Accepted article online 8 NOV 2014 Published online 13 DEC 2014

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

BECKER AND STEVENS

Climate and climate sensitivity to changing CO2 on an idealized land planet Tobias Becker1 and Bjorn Stevens1 1

Max Planck Institute for Meteorology, Hamburg, Germany

Abstract The comprehensive general circulation model ECHAM6 is used in a radiative-convective equilibrium configuration. It is coupled to a perfectly conducting slab. To understand the local impact of thermodynamic surface properties on the land-ocean warming contrast, the surface latent heat flux and surface heat capacity are reduced stepwise, aiming for a land-like climate. Both ocean-like and land-like RCE simulation reproduce the tropical atmosphere over ocean and land in a satisfactory manner and lead to reasonable land-ocean warming ratios. A small surface heat capacity induces a high diurnal surface temperature range which triggers precipitation during the day and decouples the free troposphere from the diurnal mean temperature. With increasing evaporation resistance, the net atmospheric cooling rate decreases because cloud base height rises, causing a reduction of precipitation. Climate sensitivity depends more on changes in evaporation resistance than on changes in surface heat capacity. A feedback analysis with the partial radiation perturbation method shows that amplified warming over idealized land can be associated with disproportional changes in the lapse rate versus the water vapor feedback. Cloud feedbacks, convective aggregation, and changes in global mean surface temperature confuse the picture.

1. Introduction An increase in land-ocean contrast is a robust response to global warming. It is evident in the transient response to enhanced greenhouse gas concentrations, and persists also in the equilibrium response [e.g., Manabe et al., 1991; Sutton et al., 2007]. Observations of the late twentieth century also show land warming more than oceans [Sutton et al., 2007]. Counterintuitively, the transient response is not primarily related to the differing heat capacities of land and ocean, which could be hypothesized to cause a more rapid warming over land, but rather to the same mechanisms responsible for the equilibrium response [Manabe et al., 1991; Joshi et al., 2008]. Joshi et al. [2008] explain the increasing land-ocean temperature contrast with warming using a simple thermodynamic argument. Because gravity waves are efficient at homogenizing the free-tropospheric temperature throughout the tropics, temperatures in the middle troposphere should be the same over land as they are over the ocean. Hence, the increasing land-ocean temperature contrast with warming can be understood as a response to increasing disparities between the effective lapse rates, which measure the difference between free-tropospheric and surface temperatures, over land versus over ocean with warming. Over the ocean, cloud bases are lower, and the lapse rate is moist-adiabatic over a deeper layer as compared to over land. Byrne and O’Gorman [2013] extend these ideas by additionally assuming that in the wet tropics (which they associate with the geographic region equatorward of 20 latitude), convection couples the land surface to the free troposphere, which implies that the surface moist static energy is the same throughout the tropics, something that is roughly observed (e.g., their Figure 1c). To the extent the temperature at the emission height of the free troposphere is mostly determined by the net absorbed solar radiation and the amount of long-lived greenhouse gases in the atmosphere, these findings imply that in compliance with Joshi et al. [2008], important aspects of the land-ocean warming contrast in the tropics can be understood by studying the land and ocean separately. Approaching the problem from the perspective of local equilibrium assumes that the net absorbed radiation is similar over land and ocean, and that the thermal structure of the middle and upper troposphere results in an effective emission height that is also similar over land and ocean. However, one expects a more pronounced diurnal cycle over land, which could decouple the mean surface temperature from that

C 2014. The Authors. V

1205

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

of the free troposphere. Likewise, the thermal structure of the atmosphere over land might begin to depart more significantly from that over the ocean as the land aridity increases. These changes may also call into play different feedbacks on radiative perturbations, which in turn might affect the sensitivity of the amplification factor / to perturbations. For instance, Fasullo [2010] argues that constraints on RH imply systematic changes in the cloud distribution and radiative feedbacks over land, which differ from those over the ocean. Rochetin et al. (2014), using a one-dimensional model based on the physics of the Laboratoire de Meteorologie Dynamique (LMD) general circulation model, also show that the behavior of clouds, particularly low clouds, may play an important role in deterring the equilibrium climate state over land. In this study, the authors, similarly to Rochetin et al. (2014), use the framework of radiative-convective equilibrium (RCE) to study how the equilibrium state of a land-like planet and its sensitivity to changing concentrations of atmospheric carbon dioxide depends on properties of the land surface. The idealization implied by RCE is a severe one. But previous work has shown it to be very useful for understanding aspects of climate and climate change for a water covered surface [Manabe and Strickler, 1964; Ramanathan and Coakley, 1978; Popke et al., 2013], and given the importance of local equilibrium to some of the arguments explaining differences in land-ocean temperatures, understanding the impact of surface properties on RCE appears worthwhile. A further motivation for studying RCE over land-like planets is that it provides a framework for developing a deeper understanding of how convection couples the surface to the atmosphere, a framework which is also well suited to study using cloud resolving models [e.g., Tompkins and Craig, 1998a; Muller and Held, 2012; Cronin et al., 2014]. As alluded to above, there is a long history of the use of RCE in a one-dimensional context using models with parameterized convection, and in two or three dimensions using models with explicit convection. The idea of using models with parameterized physics, such as state-of-the-art general circulation models (GCMs), to study RCE in three dimensions is relatively new. Held et al. [2007] used the full GCM physics package in a Cartesian geometry with fixed sea-surface temperatures (SSTs) to study the effects of different modeling assumptions on the representation of convection and cloud feedbacks. Popke et al. [2013] picked up this idea and performed climate change experiments by coupling the atmosphere to a mixed-layer ocean (MLO) instead of using fixed SSTs to show that over the tropical oceans the behavior of the RCE version of the ECHAM6 general circulation model differed very little as compared to the behavior of the full Earth System Model over the tropical oceans, further encouraging the use of the simplified framework. Compared to RCE studies with single-column models [Ramanathan and Coakley, 1978], the great advantage of a GCM is that atmospheric profiles of clouds and water vapor are free to interact with larger-scale circulations driven by convection. Besides, a GCM has the advantage that multiple equilibria appear to pose less of a problem, as the model is less likely to lock into a particular state. The study presented here uses the same model setup as Popke et al. [2013] and builds on their results, but concentrates on the land-ocean warming contrast in response to an increase of CO2 concentration. The article is structured as follows: in section 2, the model configuration and methodology of the study are described. In section 3, the transition from an RCE ocean climate to an idealized land climate with adapted thermodynamic surface properties is analyzed and controls on climate sensitivity are discussed. After concentrating on the general differences of climate and climate change over ocean and idealized land in RCE, the contribution factors are disentangled in section 4, focusing in one subsection on the impact of surface heat capacity and in another subsection on the reduction of latent heat flux by introducing an evaporation resistance that impedes evaporation from the ocean. The climate sensitivity analysis investigates the role of feedback factors like lapse rate, water vapor, and cloud feedback together with other key processes that are different over land and ocean. Conclusions are discussed in section 5.

2. Method and Model RCE simulations are performed in ECHAM6 with a coarse resolution, T31L47. Apart from the RCE modifications discussed below, this version of ECHAM6 is identical to the version described by Stevens et al. [2013], which serves as the atmospheric component of the fully comprehensive Earth System Model of the Max Planck Institute for Meteorology, hereafter the MPI-ESM [Giorgetta et al., 2013]. The MPI-ESM participated in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) [Taylor et al., 2012]. The RCE model

BECKER AND STEVENS

C 2014. The Authors. V

1206

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

setup is very similar to the setup of Popke et al. [2013]. The only differences are a slightly different model version, ECHAM-6.1.05 instead of ECHAM-6.0.13, a time step of 900 s instead of 450 s and a horizontal resolution of T31 instead of T63. The resolution-dependent tuning parameters [Mauritsen et al., 2012] are set to the T63 values in order to enhance comparability with the results of Popke et al. [2013]. Those are the fraction of convective cloud mass-flux that overshoots at the level of non-buoyancy (0.21), the conversion rate to rain in convective clouds (2 3 1024 s21) as well as the inhomogeneity factors of liquid and ice clouds (liquid: 0.77, ice: 0.8), which matter for radiative transfer. Contrary to Popke et al. [2013], who used different convection schemes, the standard ECHAM6 convection scheme developed by Nordeng [1994] is used in the experiments presented here. Output written to disk on six-hourly intervals is used as the basis for the analysis. 2.1. The RCE Version of ECHAM6 In RCE, the boundary conditions are traditionally specified to be homogeneous, so as to correspond to a one-dimensional setup, though respecting the influence of circulation on the mean state. In ECHAM, the spectral dynamical core maintains the homogeneity on the sphere, but because the physics are performed on a latitude-longitude grid, the grid size decreases toward the poles. However, in accordance with Popke et al. [2013], no influence on the simulated climate system was detected, although if the averaging period is very long, a small imprint of the physics grid can be expected. If the mean state of each vertical column is, on average, the same, the vertical thermal structure is, on average, determined solely by radiation and convection, where convection includes moist convection, turbulence, and boundary layer mixing [Ramanathan and Coakley, 1978]. To achieve homogeneous boundary conditions, the latitudinal insolation gradient is removed by prescribing the same diurnal cycle of insolation everywhere with a mean insolation of 340.3 Wm22 , and the atmosphere is coupled to a nonrotating, perfectly conducting slab with a specified heat capacity. This heat capacity is equivalent to the heat capacity of a water column with a depth h, and in accord with the setup of Popke et al. [2013], a default value of h 5 50 m is chosen. The surface slab, which is assumed to be perfectly conducting in the vertical, is perfectly insulating in the horizontal, so that horizontal heat transport through the slab is zero. Further details are discussed in Appendix A. 2.2. Experiments In the framework of RCE, idealized land climate is analyzed. Land differs from ocean in its heat capacity, albedo, roughness length, and water reservoir. The surface energy budget equation cp qh

@Ts 5SWnet 1LWnet 1SH1LH1G @t

(1)

illustrates that the surface temperature Ts depends on net radiative (shortwave - SWnet and longwave - LWnet ) and turbulent heat fluxes (sensible - SH and latent - LH), as well as on the surface heat capacity. In the RCE model, the ground heat flux G is set to zero and runoff is turned off. Neglected are the dynamical response of the lands to the history of precipitation and evaporation and the changes in surface albedo, which is set to a value of 0.07 to correspond to the mean ocean albedo. Neglected are also changes in roughness length because the idealized framework of RCE is particularly suitable for analyzing thermodynamic processes. Thereby, an emphasis on other properties of the land surface is implicitly maintained, such as the mean water availability, which is studied by varying the latent heat flux, introducing an evaporation resistance. In addition, this study investigates the influence of different surface heat capacities on atmospheric processes. Both effects are studied individually and together to identify possible nonlinear effects. An overview of the simulations is given in Table 1. In order to change the surface heat capacity, the parameter defining the depth of the water column h is reduced, from 50 to 0.02 m. The implementation of an additional resistor that reduces moisture fluxes at the surface is more complex. In ECHAM6, the turbulent flux of a variable v at the surface is obtained from the bulk transfer relation   ! x0 v0 s 52Cv jVl jðvl 2vs Þ; (2) ! where Cv is the transfer coefficient, Vl is the horizontal wind vector, and the subscripts l and s refer to values at the lowest model level and at the surface, respectively [Roeckner et al., 2003]. If v is replaced by the specific humidity q, the latent heat flux coefficient Cq is reduced by adding a second resistor in series:

BECKER AND STEVENS

C 2014. The Authors. V

1207

Journal of Advances in Modeling Earth Systems Table 1. Overview of Simulations With Differing Slab Depth h, Evaporation Resistance R, and CO2 Concentrationa Simulation h (m)

R

CO2

RCE experiments 50 0 13 50 0 43 1 0 13 1 0 43 0.1 0 13 0.1 0 43 0.02 0 13 0.02 0 43 50 3 13 50 3 43 50 15 13 50 15 43 50 63 13 50 63 43 0.1 3 13 0.1 3 43 0.1 15 13 0.1 15 43 0.1 63 13 0.1 63 23 Reference experimentsb 13 43

Simulation Length (Years)

Equilibrium Length (Years)

a PRP Module (Years)

70 60 60 60 30 30 60 60 40 40 40 40 40 40 30 30 30 30 30 30

60 50 50 50 20 20 50 50 30 30 30 20 30 20 20 20 20 20 20 20

10 10 10 10 20 20 10 10 10 10 10 10 10 10 20 20 20 20 20 20

50 50

30 30

a The years in which the PRP module is used are part of the years in equilibrium. b Simulation of Mauritsen et al. [2013].

Cq 5

10.1002/2014MS000369

1 R 1 Cq Cq

!21 :

(3)

The parameter R can be interpreted as an additional resistor which enhances evaporation resistance, making it comparable to evaporation from soil or plants. R is normalized with a typical value for the heat flux coefficient, Cq , and increased stepwise from R 5 0 to R 5 63. For R 5 0, Cq is equal to Cq , so this is the standard ECHAM setup for moisture fluxes from the ocean. For R 5 3, the turbulent moisture flux coefficient is approximately quartered. Further details concerning the particular implementation in ECHAM are provided in Appendix B. Climate sensitivity is investigated by abruptly quadrupling the CO2 concentration in a series of experiments. The equilibrium climate sensitivity (ECS) is then estimated as half of the surface temperature difference between the simulations with quadrupled and preindustrial CO2 [e.g., Gregory and Webb, 2008]: 1 DT23CO2 ’ ðT43CO2 2T13CO2 Þ: 2

(4)

For one experiment, with R 5 63 and h 5 0.1 m, the CO2 concentration is doubled to avoid instabilities that emerged in the simulation with quadrupled CO2. A factor that needs to be considered when analyzing climate sensitivity is atmospheric energy leakage, which is a known issue in all ECHAM6 versions until ECHAM 6.2 [Stevens et al., 2013]. However, the effects of changes in energy leakage have been calculated (for instance by comparing changes in energy leakage or differences between the surface and top of atmosphere (TOA) energy budget) and they are not found to play an important role or affect the interpretation of our results. The influence of the water vapor, lapse rate, and cloud feedback on climate sensitivity is diagnosed using the partial radiation perturbation (PRP) method [e.g., Wetherald and Manabe, 1988], which is described in detail by Klocke et al. [2013] and has been implemented in ECHAM6 by Mauritsen et al. [2013]. The PRP module is relatively fast, as it performs in parallel on all processors. The basic principle is to substitute the radiation relevant variables from the perturbed simulation separately into the control simulation, so the feedback factor kx 5

Dx Q Dx Dx DTs

(5)

can be calculated separately for any three-dimensional, time-resolved quantity x [Klocke et al., 2013]. Dx and DTs represent the difference between control and perturbed experiment, with DTs standing for the surface temperature change. Dx Q=Dx is from off-line radiation calculations, with Q representing the TOA net radiation flux. The PRP method is applied every 22 h twice: forward, by substituting a variable from the perturbed into the control climate and backward, by substituting a variable from the control into the perturbed climate. The radiative perturbation is estimated as the average of forward and backward perturbation [Colman and McAvaney, 1997]. The RCE experiments are compared with two experiments by Mauritsen et al. [2013], hereafter referred to as the reference experiments. Their ECHAM6 setup conforms to the atmosphere and land model used in the

BECKER AND STEVENS

C 2014. The Authors. V

1208

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

MPI-ESM-LR [Giorgetta et al., 2013], which participated in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) [Taylor et al., 2012], except that it is coupled to a 50 m deep slab ocean, with a prescribed oceanic heat transport instead of a dynamic ocean model. Mauritsen et al. [2013] set the preindustrial CO2 concentration to 284.7 ppm, slightly higher than in our RCE experiments (278 ppm). Only briefly discussed are some additional experiments, in which changes in the surface heat capacity were introduced, but with diurnally averaged insolation, so as to separate the effects of the diurnal cycle from the effects which result from the interaction with the atmosphere (through a fast response of the surface temperature to energy imbalances). All experiments include at least 20 simulated years in equilibrium state (Table 1).

3. RCE Climate Over Land and Ocean 3.1. Phenomenology In RCE, spatially persistent large-scale circulations like the Hadley circulation or the Walker circulation on earth are inhibited due to homogeneous boundary conditions. However, in RCE, ECHAM produces largescale precipitation clusters, ranging in size from a few degrees to continental size. Large-scale circulations develop in association with these clusters but are more dynamic as they evolve with the life-cycle of the clusters themselves. As a result, on time scales of days, the distribution of vertical motion in RCE is similar to the monthly averaged circulations found in more realistic configurations, and the covariance between the large-scale vertical motion, cloud radiative effects, and atmospheric thermodynamic properties in RCE is similar to what is found in the fully coupled model [Popke et al., 2013]. Popke et al. [2013] have already shown that RCE reproduces the main features of the mean tropical climate over ocean, with differences between the RCE climate and the mean climate of the same model running in a more realistic configuration being all smaller than differences that arise from changes to the convection scheme. The use of a lower resolution (T31), but otherwise similarly configured version of the RCE model in the present study does not change this basic finding. Figure 1 demonstrates that the RCE climate reproduces the atmospheric temperature profile of the reference experiments and the double peaked relative humidity profile with a dry mid troposphere in between. Relative humidity maximizes in the lower troposphere, where shallow cumulus detrains, and at the tropopause, where cirrus clouds form. Cloud and condensate profiles show that in RCE shallow cumulus is concentrated more in the boundary layer, but otherwise the climate of the reference model over the tropical ocean is well captured, consistent with what was found by Popke et al. [2013]. To specify land properties, values of R 5 15 and h 5 0.1 m were chosen to match the diurnal temperature range and Bowen ratio over tropical land with what is produced by the RCE experiments. Setting R 5 15 is equivalent to a reduction of the latent heat flux coefficient by a factor of 16. Figure 1 shows that the RCE climate with surface properties specified in this fashion also captures many of the differences between tropical land and ocean. The lower troposphere is drier and the bimodal relative humidity structure is less

Figure 1. Vertical equilibrium profiles of atmospheric properties in different RCE model configurations (mixed-layer ocean planet with h 5 50 m and idealized land planet with h 5 0.1 m and R 5 15) compared to the ECHAM6 reference experiment of Mauritsen et al. [2013], considering only the values over the tropical ocean in one case and over the tropical land in the other case. Tropics are defined from 30 N to 30 S.

BECKER AND STEVENS

C 2014. The Authors. V

1209

Journal of Advances in Modeling Earth Systems Table 2. Equilibrium Global Mean Surface Temperature and Climate Sensitivity for All Experiments With Differing Slab Depth h and Evaporation Resistance Ra h (m)

R

RCE experiments b 50 0 1 0 0.1 0 0.02 0 50 3 50 15 50 63 0.1 3 c 0.1 15 0.1 63 Reference experimentse Ocean Land

T13CO2 (K)

T43CO2 (K)

DT23CO2 (K)

300.1 299.8 301.3 298.6 306.7 309.2 306.7 308.1 313.0 305.0

304.5 304.5 304.5 303.4 309.8 316.6 311.3 310.6 322.2 307.2d

2.2 2.4 1.6 2.4 1.6 3.7 2.3 1.2 4.6 2.2

298.7 297.4

303.9 306.3

2.6 4.4

a T13CO2 and T43CO2 denote surface temperatures with preindustrial and quadrupled CO2 concentrations, respectively, and DT23CO2 is the climate sensitivity. b RCE, ocean. c RCE, idealized land. d Accounts to a doubling of CO2 concentration. e Simulation of Mauritsen et al. [2013], 30 N to 30 S average.

10.1002/2014MS000369

evident than over ocean. Due to enhanced triggering of convection by the diurnal cycle, the shallow cumulus detrainment layer deepens, leading to more mixing and increased cloud fraction in the 700 hPa level and a moister middle troposphere as compared to oceanic profiles. The RCE model is warmer over land as compared to ocean, consistent with many models and observations, although curiously this feature is not evident in the base climate of the reference experiments. Precipitation reduces in the RCE model to 1.5 mm/d over idealized land, which is similar to conditions in semiarid land climates, e.g., in the Sahel.

However, some of the land characteristics are exaggerated in the RCE simulation. The lower troposphere (below 700 hPa) is somewhat drier and warmer over land, and there is a substantial reduction in shallow (liquid water) clouds. The lack of low-level clouds is consistent with a much drier lower troposphere. The lack of low-level clouds, and the somewhat lower surface albedo, implies a greater amount of absorbed solar radiation in the RCE climate, consistent with the warmer mean state. Enhanced low-level drying may also be an indication of a too high value of evaporation resistance. 3.2. Land-Ocean Warming Contrast in RCE In the reference experiment, the surface temperature increase in response to a doubled CO2 concentration ranges from 2.6 K over tropical ocean to 4.4 K over tropical land. In the RCE setup, the climate sensitivity is 2.2 K over ocean and 4.6 K over idealized land (Table 2 and Figure 2). This corresponds to a land-ocean warming ratio of /51:7 in the reference experiment and /52:1 in the RCE experiment. For comparison, / ranges from 1.4 to 1.8 in the CMIP5 models [Joshi et al., 2013]. Though the strong landocean contrast in RCE might be a coincidence, this study shows that it is possible to reproduce a strong land-ocean warming contrast just with different local thermodynamic surface properties, and if at all, circulations that arise from land-ocean contrasts act to reduce these differences. Nonetheless, some additional effects from the abnormally high temperatures in the idealized land setup cannot be excluded, as will be discussed in section 4.4. The ECS of 2.2 K in the RCE experiment over ocean agrees well with the ECS of 2.1 K which Popke et al. [2013] found with a higher resolution (T63), but otherwise the same ECHAM6 RCE model setup. This further supports the present lower resolution model setup. The higher climate sensitivity over the tropical ocean in the reference experiments (using the standard version of ECHAM6 with a mixed-layer ocean) can be explained with additional warming over land and poleward amplification of the warming, both of which felt by the tropical ocean.

4. Detailed Analysis of Surface Properties The previous results motivate a deeper investigation of the factors responsible for the differences between the RCE climate over land versus over ocean. In this section, both the surface heat capacity (as measured by h) and evaporation resistance, R, are altered stepwise in a series of experiments. Feedback factors are analyzed to help identify physical mechanisms which could explain how changes in land surface properties affect climate sensitivity. In addition, to better understand the influence of different surface properties on climate, these series of experiments help identify how clouds and the diurnal cycle are coupled to the mean climate over land.

BECKER AND STEVENS

C 2014. The Authors. V

1210

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

4.1. Influence of Surface Heat Capacity on Climate in RCE Reducing h from 50 to 2 cm decreases the surface heat capacity by a factor of 2500 and enables a much quicker adaptation of the surface to an energy imbalance. For reference, the heat capacity of the atmosphere is equivalent to the heat capacity of a water column with a depth of about 2.5 m [Trenberth and Stepaniak, 2004], so that for surface heat capacities smaller than this, the time scale of the surface response is shorter than that of the atmosphere. Neglected in this comparison are the dependence of the atmospheric response on mixing time scales [Tompkins and Craig, 1998b] and effects of Figure 2. Equilibrium mean surface temperature Ts with preindustrial and quadmoisture [Cronin and Emanuel, 2013], rupled CO2 concentrations in different RCE model configurations (mixed-layer ocean planet with h 5 50 m and idealized land planet with h 5 0.1 m and which would induce a slower atmosR 5 15) compared to the ECHAM6 reference experiment of Mauritsen et al. pheric response. Surface layers with [2013], considering only the values over the tropical ocean and tropical land, heat capacities much larger than that of respectively. Tropics are defined from 30 N to 30 S. the atmosphere adapt on time scales much longer than expected time scales over which atmospheric properties (and hence the surface energy budget) change, so that surface temperature variability is expected to be small. Atmospheric processes are expected to be influenced in two ways by changes in surface heat capacity. First, an interaction with the atmosphere should become more evident as soon as the surface adapts on time scales similar to atmospheric energy imbalances. Second, the temperature of a surface layer with low heat capacity will become increasingly susceptible to the diurnal cycle of insolation, leading to an increase of diurnal temperature range (DTR). To distinguish between those two effects, experiments with a constant (diurnally averaged) insolation of 340.3 W m22 have been performed in addition to the experiments with a diurnal cycle. If the surface heat capacity is reduced with constant insolation, the distribution of surface temperature Ts broadens for slab depths larger than 1 m but remains similar for smaller slab depths. This indicates that most of the atmospheric processes which influence the energy balance change on time scales on which a slab with a heat capacity equivalent to that of a 1 m deep water column can already adapt. However, the increase of interaction and the broadening of the distribution of Ts neither affect the atmospheric profiles of temperature, relative humidity, clouds, cloud condensate, the global mean surface temperature, which ranges from 299.2 to 299.6 K, nor climate sensitivity, which ranges from 2.1 to 2.4 K. Hence, the increase of interaction between surface and atmosphere itself is only of minor importance for the different climate states over land and ocean. For this reason, the experiment series with constant insolation is not discussed further. The picture is quite different if insolation is defined according to a tropical diurnal cycle. Now atmospheric properties can be altered both by the enhanced interaction with the surface and by a coupling to the DTR. Table 2 shows that equilibrium mean surface temperatures are similar in the experiments with h set to 50 and 1 m, but increase by more than 1 K in the experiment with h 5 0.1 m and decrease by more than 2 K in the experiment with h 5 0.02 m. Figure 3 hints at the reasons for a strong temperature decrease. For h 5 0.02 m, the distribution of daily mean surface temperatures develops a pronounced (20.7) negative skewness, indicative of a physical mechanism that can trigger very low surface temperatures. The increase of negative skewness is accompanied by a tendency toward a broader distribution of surface temperatures, with the variance increasing successively in all experiments with smaller values of h except for the case when h 5 0.1 m. To understand these tendencies, the influence of the DTR on atmospheric processes needs to be analyzed. Figure 4 depicts the diurnal cycle of surface temperature and precipitation. The diurnal surface temperature range increases from 0.04 K in the h 5 50 m experiment and 1.7 K in the h 5 1 m experiment to 11 K in the

BECKER AND STEVENS

C 2014. The Authors. V

1211

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

h 5 0.1 m experiment and 15 K in the h 5 0.02 m experiment. Hence, the DTR becomes sufficiently strong to expect an impact on atmospheric processes for h  0:1 m, an expectation which is supported by a shift of maximum precipitation from nighttime to daytime. The nocturnal precipitation maximum over ocean is consistent with measurements [Bowman et al., 2005]. Kraus [1963] argues that the nocturnal maximum is due to the stabilizing effect of solar heating of cloud tops during the day and the destabilizing effect of nighttime radiative cooling at cloud tops, though recent studies suggest that water vapor is at least as important as clouds [Takahashi, 2012]. However, the amplitude of daily precipitation variations is much stronger if it is triggered by the DTR. While precipitation peaks Figure 3. Histogram of daily averaged surface temperature perturbations Ts in the h 5 0.1 m experiment in the afterat all grid points relative to the global mean value for different surface heat noon, it peaks in the h 5 0.02 m experiment capacities (h 5 50, 1, 0.1, and 0.02 m), considering only the time period in equilibrium. Each grid point is weighted by its area. The bin size is 0.1 K. already in the morning, with a value that is 4 times as large as at night. Hence, with h 5 0.02, precipitation peaks too early compared to the real tropics over land. Precipitation peaking too early during the day is a common problem in GCMs [e.g., Bechtold et al., 2004; Lawrence and Slingo, 2005]. In the present series of experiments, this change can be related to an excessive triggering of convection in the morning, from a too rapid warming of the land surface, resulting in a strong reduction of precipitable cloud water later in the afternoon. Hence, precipitation decreases through the day even if the atmosphere becomes increasingly unstable. 0

In the experiments with h  0:1 m, the shallow cumulus detrainment layer deepens and moisture is mixed through a deeper layer because convection is triggered by moisture convergence from increased evaporation that accompanies a rapid increase of Ts during the day. Hence, boundary layer clouds dissipate while

Figure 4. Diurnal cycle of (a) surface temperature and (b) precipitation for different surface heat capacities. The values are averaged over 6 h and are plotted at the center of the averaging interval.

BECKER AND STEVENS

C 2014. The Authors. V

1212

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Figure 5. Vertical equilibrium profiles of atmospheric properties for different surface heat capacities.

cloud fraction increases at 700 hPa (Figure 5), as this is a level where shallow convection frequently termi€bis and Stevens, 2012]. Since the properties of these clouds are reminiscent of stratocunates in ECHAM6 [Mo mulus cumulogenitus clouds, this name will be used henceforth. The dissipation of boundary layer clouds induces a surface temperature increase because low clouds have a cooling effect and explains the decrease of variance in the h 5 0.1 m experiment because Ts is altered less by the boundary layer clouds. The vertical atmospheric profiles in Figure 5 emphasize the coupling of the diurnal maximum surface temperature, Ts;dmax , to the atmosphere. Although the vertical profiles are similar for h  1 m, a transition occurs for lower surface heat capacities, resulting from the strong diurnal cycle of Ts . In those experiments, the atmosphere decouples from the daily mean surface temperature because moist convection is triggered near the time when Ts maximizes. Therefore, the moist-adiabatic lapse rate depends on Ts;dmax and decreases disproportionately to the change in the mean surface temperature. The temperature difference between the 200 and 1000 hPa levels decreases from 81.6 K in the h 5 1 m experiment to 78.5 K in the h 5 0.1 m experiment. The decoupling from the mean surface temperature is also evident in the h 5 0.1 m and h 5 0.02 m experiment, in which the vertical temperature profiles as well as Ts;dmax are similar, although the diurnal mean surface temperatures differ by more than 2 K. With decreasing surface heat capacity, Ts;dmax remains approximately constant while Ts;dmin decreases strongly. That is because latent and sensible heat fluxes restrain the warming during the day, but during the night the amount of cooling is only limited by the length of the night (and whether or not nocturnal cloud layers develop). As discussed by Cronin et al., [2014], this rectification effect is likely an important reason for the strong decrease of equilibrium mean surface temperature, when decreasing h from 0.1 to 0.02 m. Another reason for the strong decrease of Ts is related to the 700 hPa clouds, which are more pronounced in the subsidence regime over a 0.02 m thick slab than over a 0.1 m thick slab (Figure 6). The 700 hPa cloud formation is favored because strong longwave cooling at the cloud tops is very efficient when the mid troposphere is very dry. The cooling supports the development of an upper level inversion which inhibits mixing with the drier free-atmosphere above and helps sustain the clouds. These same clouds reflect shortwave radiation and cool the surface effectively but are sufficiently high that their effect on the surface longwave budget is more marginal, which amplifies the 700 hPa temperature inversion. A lack of boundary layer clouds thus enhances the coupling of the stratocumulus cumulogenitus layer to the surface. Figure 6 shows that in the h 5 0.02 m experiment, this positive feedback mechanism dominates over the cloud dissipation effect from mixing with dry and warm subsiding air, leading to a further cooling of the already cold slab (Figure 7). This explains the negative skewness of the surface temperature distribution in the h 5 0.02 m experiment (Figure 3). In the h 5 0.1 m experiment, the effect of dry and warm subsiding air becomes more prominent, which is why the 700 hPa clouds have less impact, both on the mean surface temperature and on its skewness. It is difficult to argue that the balances found by the model in this respect are indicative of what nature would do, nonetheless they show the important role lower tropospheric clouds play in regulating the diurnal temperature range over land. The simulations suggest that a heat capacity corresponding to h  0:1 m is necessary to match observed diurnal temperature ranges over land, and hence is land-like. The diurnal 2 m temperature range is about 6 K in the h 5 0.1 m experiment and 10 K in the h 5 0.02 m experiment. According to Jackson and Forster [2010], the mean diurnal temperature range over the global land surface is 11 K, with smaller values in

BECKER AND STEVENS

C 2014. The Authors. V

1213

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Figure 6. Vertical cross sections of mean cloud cover and wind field for different surface heat capacities (h 5 1, 0.1, and 0.02 m). The cross sections are centered at the location with the minimum monthly mean surface temperature, Ts;min , and show the vertical profiles along that longitude (example in Figure 7). The cycle is closed with the longitude on the other side of the planet. The vertical velocity is scaled with a factor of 100. The data are averaged over all time steps in equilibrium. Representativeness is increased by averaging the data points with the same longitudinal distance to the minimum surface temperature.

tropical areas, especially in wet regions. Hence, a slab depth of either 0.1 or 0.02 m is consistent with observed diurnal temperature ranges. As a consequence, a slab depth of 0.1 m will be considered as landlike in the following. Although it could be argued that a smaller, h 5 0.02 m land heat capacity is also plausible, those simulations are less stable because they develop strong winds in the highest model levels. In addition, they have a more distorted cycle of precipitation relative to the reference experiments.

4.2. The Effects of Surface Heat Capacity on Climate Sensitivity in RCE RCE experiments with quadrupled CO2 concentration show that ECS does not change much for h  1 m, where the DTR does not influence atmospheric processes yet (Table 2 and Figure 8a). For lower surface heat capacities (h  0:1 m), there is a general trend for the ECS to increase, to 2.4 K in the experiment with h 5 0.02 m, albeit somewhat obscured by the one-off decrease of ECS to 1.6 K in the experiment with h 5 0.1 m. The variation of ECS with decreasing heat capacity is mostly related to changes in the unperturbed (1 3 CO2) experiments because the scatter of equilibrium mean surface temperature is much smaller in the perturbed (4 3 CO2) experiments. An important reason for this difference is that there is much less effect from boundary layer cloud dissipation in the latter experiments, because there are hardly any boundary layer clouds for any surface heat capacity. Another reason is that the impact of 700 hPa clouds decreases because an atmosphere with quadrupled CO2 concentration is more opaque, resulting in less effective radiative cooling, both at the top of the stratocumulus cumulogenitus layer and at the surface. The feedback factors show that the low climate sensitivity in the h 5 0.1 m experiment is related to a decrease of shortwave cloud feedbacks (Figure 8b). This is due to the dissipation of boundary layer clouds in the unperturbed experiment for the case of h50:1 m. For experiments with larger values of h, the shortwave cloud feedback is large, but similar to one another, as clouds only dissipate when the atmospheric CO2 concentration is increased. Likewise, for experiments with smaller values of h, the shortwave feedback is small, but similar across experiments, as there is a lack of low clouds irrespective of the concentration of CO2. The sum of water vapor and lapse rate feedback increases only slightly from 20.2 to 0.6 W m22 K21 with decreasing h, as a decrease in the positive water vapor feedback is largely balanced by a similar decrease in magnitude in the negative lapse rate feedback (Figure 8b). This slight increase in the combined water vapor and lapse rate feedback with decreasing h likely reflects the coupling of the free troposphere with the warmest surface temperatures with the emergence of the diurnal cycle. This would lead to a small but steady increase of ECS with increasing DTR if clouds did not confuse the picture. Separating these effects from those associated with cloud feedbacks is important because although they are crucial for

BECKER AND STEVENS

C 2014. The Authors. V

1214

Journal of Advances in Modeling Earth Systems

understanding ECHAM, cloud feedbacks are likely to depend more strongly on the particular formulation of the model.

#

X #

X X 60° # 120° 180°

10.1002/2014MS000369

# X

# X

Relative to global mean annual mean values of water vapor (1.94–1.39 W m22 K21) and lapse rate feedback (20.97 to 20.22 W m22 K21) in CMIP5 models [Vial et al., 2013], higher absolute values are reasonable in the RCE framework because water vapor concentrations are enhanced, similar to conditions in the tropics. With increasing DTR, nearly all the convection is initiated during the day. Hence, the atmosphere decouples from the diurnal mean surface temperature. This induces a reduction of the absolute value of water vapor and lapse rate feedback because in a warmer climate, surface evaporation is enhanced particularly during the day, weakening the daily maximum temperature increase relative to the daily mean temperature increase.

Overall, the ECS increases slightly from 2.2 to 2.4 K if the surface heat capacity is reduced from 50 to 0.02 m. Cloud feedbacks confuse the picture and mask the trend of increasing ECS with decreasing surface heat capacity. Without the one-off decrease of shortwave cloud feedback, the trend would certainly be more pronounced. Although the underlying trend would probably be larger without the one-off decrease, it is too weak to explain the landocean warming contrast observed on earth. Nonetheless, the DTR becomes land-like if h  0:1 m, which is important because atmospheric processes like convection decouple from the diurnal mean surface Figure 7. Random snapshots of the surface temperature field relative to the global mean surface temperature for a surface heat capacity equivalent to a 1 and 0.02 m water column, averaged over 1 month. In addition, the surface wind field is shown. The longitude with the minimum surface temperature is marked.

Figure 8. (a) Equilibrium mean surface temperatures with preindustrial and quadrupled CO2 concentrations and (b) the contribution of the net feedback factors with decreasing surface heat capacity. The feedback factors are calculated with the PRP module. wv 1 lr is the sum of water vapor and lapse rate feedback factor.

BECKER AND STEVENS

C 2014. The Authors. V

1215

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

temperature. Additional experiments with intermediate slab depths of 5 and 0.05 m (not shown) are consistent with these findings. 4.3. Influence of Evaporation Resistance on Climate in RCE A decrease of the latent heat flux coefficient through the introduction of an evaporation resistance R at the surface (details in section 2.2 and Appendix B) modifies both the surface energy budget and the atmospheric water supply. Two series of experiments have been conducted, one with a large surface heat capacity and one with a land-like surface heat capacity equivalent to a 0.1 m water column (Table 1). This subsection concentrates on the latter, while the next subsection considers both series of experiments. With increasing Figure 9. Mean equilibrium surface fluxes in the experiments with difevaporation resistance, the latent heat flux ferent evaporation resistances and a surface heat capacity equivalent to a 0.1 m water column. SW and LW denominate the net shortwave and decreases (Figure 9), but not as much as the longwave radiation at the surface, respectively, and SH and LH stand for latent heat flux coefficient, and rather linearly sensible and latent turbulent heat fluxes, respectively. Upward fluxes than exponentially. On a global scale, precipiare defined to be positive. tation at the surface must balance evaporation from the surface. Thus, precipitation also decreases approximately linearly with increasing R, from 4.4 to 0.3 mm d21. In the absence of moisture transport from the ocean, these precipitation rates are partly smaller than mean precipitation over land (2.7 mm d21 in the reference experiment), but are similar to precipitation in semiarid or arid climate zones. In RCE, a definition of aridity via the water balance is not reasonable because surface evaporation and precipitation are by definition on average equal since moisture cannot be advected from other areas. Here aridity is understood rather as a general lack of humidity, measured at the surface by the Bowen ratio. The decrease of latent heat flux is partially compensated by an increase of sensible heat flux, resulting in an increase of Bowen ratio. In MPI-ESM simulations, the Bowen ratio is about 1:2 in very humid areas of the tropics over land, with no upper limit in arid regions like in Sahara desert. According to this, the Bowen ratio in our RCE experiments can be considered land-like for R  3 (Bowen ratio is 1:10, 1:2, 2:1, and 8:1 for R 5 0, 3, 15, and 63, respectively). Because of the increasing Bowen ratio, the ascending air is drier and has to cool more before it saturates at the lifting condensation level (LCL). Figure 10 confirms that the mean cloud base rises, and relative humidity decreases in the layer below 600 hPa. The changes of clouds and water vapor induce an increase of the effective emission height of the longwave radiation downwelling at the surface. Thus, the atmosphere cools less, which explains the decrease of convective heating and therefore of precipitation from an atmospheric point of view. This result suggests that the presence of land reduces global mean precipitation through its influence on the radiation budget. Because the sensible heat flux does not increase as much as latent heat flux decreases, the net upward longwave radiation at the surface must increase (Figure 9). In addition, in response to a decrease of planetary albedo, the net shortwave radiation increases slightly, amplifying this trend. The net upward longwave radiation increases either due to rising surface temperatures or due to a reduction in the downwelling component. The latter strongly depends on the amount of water vapor and cloud water in the atmosphere. Table 2 and Figure 14 show that the global mean surface temperature increases with R until a certain point, as for R  15; Ts begins to decrease. A decrease in Ts with R  15 is robust both for ocean-like and land-like surface heat capacities and is what one expects given that in the limit, an atmosphere without water will have a much smaller greenhouse effect than the present atmosphere. From this perspective, the increase in Ts with R for small values of R is counterintuitive. The two view points can be reconciled by noting that as the evaporation resistance increases, the atmosphere becomes more arid, reducing its greenhouse effect,

BECKER AND STEVENS

C 2014. The Authors. V

1216

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Figure 10. Vertical equilibrium profiles of atmospheric properties in the experiments with different evaporation resistances and a surface heat capacity equivalent to a 0.1 m water column. The dashed line in the temperature profile represents the dry adiabatic lapse rate, starting at the surface temperature of the experiment with an evaporation resistance R 5 63.

causing a surface cooling. But for a given emission temperature a drier atmosphere will have a warmer surface by virtue of a larger temperature lapse rate. Because increasing the evaporation resistance at the surface dries the atmosphere from the bottom-up, the lapse rate effect initially dominates, with the weakening of the greenhouse effect only becoming evident for somewhat larger increases in R. Not only do the vertical profiles of the mean thermodynamic structure change with increasing evaporation resistance, but also the horizontal distribution of convective clusters changes. Figure 11 shows that precipitation far away from the location with maximum precipitation approaches zero for R  3, indicating a transition to only one precipitating cloud cluster. The two random snapshots in Figure 12 illustrate this transition. Because the upward mass flux in this single convective cluster has to balance the subsidence due to radiative cooling everywhere else, the overturning circulation at the location of maximum precipitation increases for R  3 (Figure 13). In addition, this composite suggests that the surface wind does not converge at the center of the cluster anymore, but at its boundaries. The reason is that for a very arid atmosphere, an evaporatively driven cold pool develops beneath the precipitation cluster (Figure 13). This strong virga effect implies that a substantial portion of the precipitation does not reach the surface and that the main source of atmospheric humidity is evaporating rain. This convective structure is reminiscent of a low precipitation super-cell [Bluestein and Parks, 1983], albeit on a planetary scale, of the type that are frequently found in the arid regions of North America, especially the High Plains. It would be interesting to explore to what extent cloud resolving simulations of RCE over a more arid surface behave similarly.

Figure 11. Precipitation sampled along the longitude of maximum precipitation (example in Figure 12) for a surface heat capacity equivalent to a 0.1 m water column and evaporation resistances R 5 0, 3, 15, and 63, closing the cycle with the longitude on the other side of the planet and moving the grid point with maximum precipitation to the center. The data are averaged over all time steps in equilibrium. Representativeness is increased by averaging the data points with the same longitudinal distance to the maximum value.

BECKER AND STEVENS

C 2014. The Authors. V

Although a change in the structure of convective systems is plausible as the evaporation resistance of the surface increases, the reasons why these systems should aggregate on the planetary scale are less clear. An important consequence of convective aggregation is that the cloud-free environment tends to be much drier [e.g., Bretherton et al., 2005], leading to increased longwave cooling. In aggregated states, Tobin et al. [2012] detected a drying in the subsidence regime in several satellite data sets and reanalysis. Bretherton et al. [2005] and Muller and Held [2012] relate convective aggregation to an upgradient horizontal transport of moist static energy (MSE). In the present experiments, cloud radiative effects appear to be more important than upgradient MSE transport in

1217

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

the boundary layer for the fusion of the precipitating clusters into one single cluster. High clouds warm the atmos# phere and low clouds cool the atmosphere. The warming of the atmosphere by high clouds can be thought of as a # X reduction in the atmospheric cooling, as less radiation is emitted through the top of the atmosphere in the presence of high clouds, whose tops are much X P max colder than the effective emission X 60° height of the clear atmosphere. These # 120° heating effects will become more pro180° # nounced as the atmosphere dries, effecX tively decreasing the effective emission height, or increasing the effective emission temperature. For this reason, and perhaps because low clouds also decrease with increasing evaporation resistance, the net cloud radiative effect # of the atmosphere, CREatm , increases X from 16 to 22 W m2 due to changes in the longwave radiative budget accompanying an increase in the evaporation resistance. For comparison, the NASA Surface Radiation Budget data set shows that the global mean CREatm is Figure 12. Random snapshots of the distribution of precipitation relative to the global mean precipitation for evaporation resistances R 5 0 and R 5 15 with a approximately zero, though CREatm is surface heat capacity equivalent to a 0.1 m water column, averaged over 1 positive in the tropics [Allan, 2011]. In month. In addition, the surface wind field is shown. The longitude with the maxithe experiments with a surface heat mum precipitation is marked. capacity equivalent to a 50 m water column, the CREatm increase with evaporation resistance is even more pronounced, increasing from 7 to 22 W m22. In these experiments, convective aggregation also occurs for R  3. Consequently, the cloud radiative effect acts as a source of moist static energy in convectively active atmospheric regions relative to clear sky areas, triggering convective aggregation. Muller and Held [2012] found hysteresis in their experiments, which means that the equilibrium climate state is restricted by the initial conditions. However, there is no hint of hysteresis in ECHAM6-RCE. At least in an experiment with R 5 0, initiated in an aggregated state, the single cluster rapidly breaks up. In addition, Emanuel et al. [2014] hypothesize that convective aggregation mostly depends on Ts . Further experiments with an adjustable surface albedo could clarify that matter. 4.4. The Effects of Evaporation Resistance on Climate Sensitivity in RCE In a completely dry atmosphere, water vapor, lapse rate, and cloud feedback will, by definition, vanish. In such an extreme climate, one expects a warming of about 1 K [Stevens and Bony, 2013]. However, it is not obvious how ECS changes if water accessibility is reduced stepwise. It turns out that, for reasons not unlike those used to explain why surface temperatures maximize for R 5 15, climate sensitivity maximizes for R 5 15, before decreasing with R in very arid conditions (Table 2 and Figure 14). The ECS increase is independent of surface heat capacity and has a similar magnitude as the land-ocean warming contrast in the reference experiments. The ECS increase is most evident when comparing the R 5 3 and the R 5 15 experiments but is also evident from the sense of the combined water vapor and lapse rate feedbacks (Figure 15). From the perspective of the combined water vapor and lapse rate feedback, an increasing ECS with R is also independent of surface heat capacity, albeit more pronounced in the setup with h 5 0.1 m. Whether or not the ECS changes as expected for any particular experiment is however subject to cloud changes, the physical realism of which is difficult to judge. The main reason for an increasing climate sensitivity with increasing R is that the bottom-up drying exhausts the lapse rate feedback more readily than the water vapor feedback. A disproportional change in

BECKER AND STEVENS

C 2014. The Authors. V

1218

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Figure 13. Vertical cross sections of mean potential temperature and wind field for a surface heat capacity equivalent to a 0.1 m water column and evaporation resistances R 5 0, 3, 15, and 63, sampled analogously to Figure 11 along the longitude of maximum precipitation (example in Figure 12). The vertical velocity is scaled with a factor of 100.

the lapse rate and water vapor feedbacks arises because, with increasing evaporation resistance, primarily the area below the LCL dries. Consequently, the lapse rate feedback weakens because the lower troposphere is less often saturated and the global mean lapse rate is closer to the dry adiabat. Unlike the moist adiabat, the dry adiabat does not depend on temperature. Joshi et al. [2008] identify this mechanism as the main contributor to amplified CO2 induced warming over land. In the upper troposphere, the moist adiabat is close to the dry adiabat and almost independent of Ts . Hence, the lapse rate feedback acts primarily in the lower troposphere, while the water vapor feedback mainly depends on the relative change of specific humidity in the upper troposphere, both because the temperature difference to the surface increases with height and because the lower troposphere is nearly opaque at wavelengths of strong water vapor absorption [Soden and Held, 2006]. Hence, with increasing evaporation resistance, the lapse rate feedback weakens before the water vapor feedback decreases, resulting in a high Figure 14. Equilibrium mean surface temperatures with preindustrial ECS for R 5 15. With h 5 0.1 m, the water vapor and quadrupled CO2 concentrations for different evaporation resistanfeedback even increases for R  15. This is ces (R 5 0, 3, 15, and 63) and surface heat capacities (h 5 50 and 0.1 m). related to an increase of Ts . Meraner et al. [2013]

BECKER AND STEVENS

C 2014. The Authors. V

1219

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Figure 15. Change of the net feedback factors with increasing evaporation resistance with a surface heat capacity equivalent to a (a) 50 m and (b) 0.1 m water column, calculated with the PRP module. The dashed lines guide the eyes to what we assert is the robust response of the model.

state that the enhanced water vapor feedback in a warmer climate can be attributed to an increase of tropopause height. Indeed, the tropopause height increases in the perturbed (43CO2) experiment with R 5 15 and h 5 0.1 m from 100 to 70 hPa, also inducing a strong longwave cloud feedback. However, tropopause height does not increase in the experiments with h 5 50 m. Hence, the high surface temperatures are not solely responsible for the high climate sensitivity. The tendency of increasing ECS related to the disproportional response of the water vapor and lapse rate feedbacks is confused by convective aggregation on the one hand and changes in the vertical cloud distribution on the other hand, leading to a smaller ECS for R 5 3 than for R 5 0. Convective aggregation can influence ECS because the atmospheric lapse rate is determined primarily in the convective area, while the water vapor feedback is also influenced by changes in the subsidence regime. Consequently, the drying in the subsidence regime [Tobin et al., 2012] leads to a reduction of the sum of water vapor and lapse rate feedback. In addition, in the setup with h 5 50 m, boundary layer clouds dissipate in the unperturbed experiment, causing a decrease of positive cloud feedback. To summarize, the relationship between the water vapor and lapse rate feedback depends on the aridity of the base climate. Although convective heating in the lower troposphere is most important for the lapse rate feedback, the water vapor feedback also depends on changes in the subsidence regime and in the upper troposphere, particularly on tropopause height. Because, with increasing evaporation resistance, relative humidity strongly decreases in the lower troposphere while it does not change much above the base of convective clouds, the land-like RCE climate state can have a very high climate sensitivity. Although land-like climate features are somewhat exaggerated, similarities are robust (section 3), which is why the experiment with h 5 0.1 m and R 5 15 can be interpreted as an idealized land planet, and which is why the land-ocean contrast increases with warming in the present climate. If evaporation resistance is further increased, eventually the atmosphere both dries and cools, leading to a strong decrease of cloud, lapse rate, and water vapor feedback, which results in an overall decrease of ECS in very arid conditions, and hence less land amplification. Experiments with intermediate evaporation resistances of R 5 1, R 5 7, and R 5 31 confirm the presented tendencies and results.

5. Summary and Conclusions Factors influencing differences in the land versus ocean surface temperature response to increasing concentrations of atmospheric CO2 are explored by modifying the comprehensive general circulation model, ECHAM6, to simulate RCE for land-like surface properties. The simulations are performed at T31 resolution

BECKER AND STEVENS

C 2014. The Authors. V

1220

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

with a nonrotating planet, without spatial gradients in insolation and with a perfectly conducting slab with a certain heat capacity as lower boundary condition. Some additional minor changes ensure spatially homogeneous boundary conditions. The model setup matches the ECHAM6-RCE composition used by Popke et al. [2013] to determine the influence of the convection scheme. To better understand controls on climate sensitivity over land and ocean, the most important thermodynamic surface properties responsible for distinguishing land from water surfaces, the heat capacity and the evaporative flux, are reduced stepwise, aiming for a land-like climate. Therefore, three series of experiments are performed, with decreasing surface heat capacity, with increasing evaporation resistance and combining both. Climate sensitivity is investigated by comparing simulations with preindustrial and quadrupled CO2 concentrations. With an evaporation resistance R 5 15 and a surface heat capacity equivalent to that of a 0.1 m water column, RCE captures the main differences in the mean climate over tropical land as compared to over tropical oceans. For example, DTR and Bowen ratio increase to 6 K and 2:1, respectively. Due to the strong diurnal temperature cycle at the surface, convection and precipitation are primarily triggered during the day. Hence, the atmosphere decouples from the mean surface temperature and couples to the diurnal maximum surface temperature instead. Furthermore, the lower troposphere dries with increasing evaporation resistance and a reduction of surface heat capacity leads to a deepening of the shallow cumulus detrainment layer. Both developments cause a reduction of cloud and condensate concentrations in the boundary layer. The land-ocean warming contrast is reproducible in RCE by restricting surface latent heat fluxes. A reduction of surface heat capacity and the emergence of a diurnal cycle are of at most secondary importance. Though cloud feedbacks change with decreasing surface heat capacity, they are likely to be model-dependent wild cards which we hesitate to interpret too deeply. The trends of water vapor and lapse rate feedback are likely to be more robust, and their sum increases only marginally with a change to the surface heat capacity. Conversely, the bottom-up drying in response to the increase of evaporation resistance exhausts the lapse rate feedback more readily than the water vapor feedback, causing high climate sensitivities over idealized land (R 5 15 and h 5 0.1 m), before both feedback factors approach zero in very arid conditions. Water vapor and lapse rate feedback behave disproportionately because relative humidity decreases primarily in the lower troposphere, reducing the lapse rate feedback. Applied to the real tropics, these results suggest that the land-ocean warming contrast is most amplified in semiarid regions, where the air is rarely saturated with water vapor, but where the amount of water vapor in the upper troposphere still suffices for a strong water vapor feedback. These tendencies are confused not only by uncertain cloud feedbacks but also by convective aggregation. Convective aggregation causes a drying in the subsidence regime, which is linked to a reduction of water vapor feedback and to a decrease of climate sensitivity. In contrast to studies in which the evaporation resistance is not allowed to vary, self-aggregation in the present work appears to be related to an increase of the net cloud radiative effect of the atmosphere in more arid climates, which then acts as a source of moist static energy in the convective regime. The question whether convective aggregation is a physical phenomenon or only a model issue is still an object of scientific discussion.

Appendix A: Further Details on RCE in ECHAM6 In the RCE version of ECHAM6, the latitudinal insolation gradient is removed by defining the incoming solar radiation at the TOA as FðtÞ5I0 l0 ðtÞ;

(A1)

where I0 is the solar constant and l0 is the cosine of the solar zenith angle, which depends on time only. Propagating tidal effects are avoided by specifying a diurnal cycle that is coherent across the planet as a whole, so that each point experiences a diurnal cycle in phase with every other point. Because the present focus is on the tropical atmosphere, the diurnal cycle corresponds to that of the equator at equinox, where the sun passes directly overhead at noon. To account for the fact that the tropical atmosphere exports heat to the extra tropics, the solar constant is decreased by a factor of p4 from 1361 to 1069 Wm22 so that the global mean insolation is 340.3 Wm22 . The solar irradiance in ECHAM6 is resolved in 14 spectral bands. Hence, the irradiance in every band is rescaled with the factor p4. Horizontally uniform trace gas and aerosol BECKER AND STEVENS

C 2014. The Authors. V

1221

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

concentrations are achieved by setting methane and nitrous oxide volume mixing ratios to the preindustrial values, 650 and 270 ppbv; chlorofluorocarbon and aerosol concentrations are set to zero. The CO2 concentration is set to its preindustrial value, 278 ppm. The ozone concentration varies with height but is horizontally uniform. A preindustrial equatorial ozone profile is defined according to Popke et al. [2013, Figure 1]. In ECHAM6, the planetary boundary layer height is defined as the minimum of the convective and dynamic boundary layer height. As the dynamic boundary layer height depends on the Coriolis parameter, the planetary boundary layer height is equated with the convective boundary layer height. This is consistent with the boundary layer behavior in the tropics. Another necessary adaptation is an increase of atmospheric mass, to compensate for the lack of mountains and to keep the mean sea level pressure (SLP) at 1013.25 hPa. The RCE model version was not tuned, so the same tuning constants are used as in the default ECHAM6 model setup [Mauritsen et al., 2012]. No further modifications of the standard ECHAM6 setup than discussed here or in section 2.1 are necessary. Thus, the complete model physics is sustained.

Appendix B: Implementation of Evaporation Resistance in ECHAM6 In ECHAM6, there is no transport coefficient Cq specifically for latent heat, there is only a transport coefficient Ch for heat in general. The parameter zcfhw in ECHAM6 is proportional to Ch : zcfhw5

! cvdifts  Dt  g  pl  jVl j  Ch ; Rd  Tl

(B1)

where cvdifts is the time step weighting, Dt is the forecast time step, g is the gravitational acceleration, pl and Tl are pressure and temperature at the lowest model level, respectively, and Rd is the specific gas constant of dry air. Ch depends on the bulk Richardson number, the height of the lowest model level and the roughness length for momentum and heat. Among others, zcfhw is used to calculate the moisture flux zqhflw at the surface: zqhflw5

1  zcfhw  ðql 2qs Þ: cvdifts  Dt  g

(B2)

In this case and in all other cases where zcfhw affects the surface moisture flux, zcfhw is adapted according to equation (3). As zcfhw has typical values of about 200, zcfhw5200 is assumed.

Acknowledgments The authors thank J€ urgen Bader for his methodical and scientific support, Dagmar Popke for the scientific foundation this study builds upon, Aiko Voigt for his contributions to the formulation of the model setup, Thorsten Mauritsen for the implementation of the PRP method in ECHAM6, and Cathy Hohenegger for comments on an earlier version of this manuscript. The research was made possible through the support of the Max Planck Society for the Advancement of Science. Computing resources were provided by the German Climate Computing Center (DKRZ), Hamburg. Original processing scripts, model code, and data used in the plots presented in this paper can be accessed through the MPI information officer, contact information for which is maintained on the institute website.

BECKER AND STEVENS

References Allan, R. P. (2011), Combining satellite data and models to estimate cloud radiative effect at the surface and in the atmosphere, Meteorol. Appl., 18(3), 324–333, doi:10.1002/met.285. Bechtold, P., J.-P. Chaboureau, A. Beljaars, A. K. Betts, M. K€ ohler, M. Miller, and J.-L. Redelsperger (2004), The simulation of the diurnal cycle of convective precipitation over land in a global model, Q. J. R. Meteorol. Soc., 130(604), 3119–3137, doi:10.1256/qj.03.103. Bluestein, H. B., and C. R. Parks (1983), A synoptic and photographic climatology of low-precipitation severe thunderstorms in the Southern Plains, Mon. Weather Rev., 111(10), 2034–2046. Bowman, K. P., J. C. Collier, G. R. North, Q. Y. Wu, E. H. Ha, and J. Hardin (2005), Diurnal cycle of tropical precipitation in Tropical Rainfall Measuring Mission (TRMM) satellite and ocean buoy rain gauge data, J. Geophys. Res., 110, D21104, doi:10.1029/2005JD005763. Bretherton, C. S., P. N. Blossey, and M. F. Khairoutdinov (2005), An energy-balance analysis of deep convective self-aggregation above uniform SST, J. Atmos. Sci., 62(12), 4273–4292, doi:10.1175/JAS3614.1. Byrne, M. P., and P. A. O’Gorman (2013), Link between land-ocean warming contrast and surface relative humidities in simulations with coupled climate models, Geophys. Res. Lett., 40, 5223–5227, doi:10.1002/grl.50971. Colman, R. A., and B. J. McAvaney (1997), A study of general circulation model climate feedbacks determined from perturbed sea surface temperature experiments, J. Geophys. Res., 102(D16), 19,383–19,402, doi:10.1029/97JD00206. Cronin, T. W., K. A. Emanuel, and P. Molnar (2014), Island precipitation enhancement and the diurnal cycle in radiative-convective equilibrium, Q. J. R. Meteorol. Soc., Boston, USA, doi:10.1002/qj.2443, in press. Cronin, T. W., and K. A. Emanuel (2013), The climate time scale in the approach to radiative-convective equilibrium, J. Adv. Model. Earth Syst., 5, 843–849, doi:10.1002/jame.20049. Emanuel, K., A. A. Wing, and E. M. Vincent (2014), Radiative-convective instability, J. Adv. Model. Earth Syst., 6, 75–90, doi:10.1002/ 2013MS000270. Fasullo, J. T. (2010), Robust land-ocean contrasts in energy and water cycle feedbacks, J. Clim., 23(17), 4677–4693, doi:10.1175/ 2010JCLI3451.1. Giorgetta, M. A., et al. (2013), Climate and carbon cycle changes from 1850 to 2100 in MPI-ESM simulations for the Coupled Model Intercomparison Project phase 5, J. Adv. Model. Earth Syst., 5, 572–597, doi:10.1002/jame.20038. Gregory, J., and M. Webb (2008), Tropospheric adjustment induces a cloud component in CO2 forcing, J. Clim., 21(1), 58–71, doi:10.1175/ 2007JCLI1834.1. Held, I. M., M. Zhao, and B. Wyman (2007), Dynamic radiative-convective equilibria using GCM column physics, J. Atmos. Sci., 64(1), 228– 238, doi:10.1175/JAS3825.11.

C 2014. The Authors. V

1222

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000369

Jackson, L. S., and P. M. Forster (2010), An empirical study of geographic and seasonal variations in diurnal temperature range, J. Clim., 23(12), 3205–3221, doi:10.1175/2010JCLI3215.1. Joshi, M. M., J. M. Gregory, M. J. Webb, D. M. H. Sexton, and T. C. Johns (2008), Mechanisms for the land/sea warming contrast exhibited by simulations of climate change, Clim. Dyn., 30(5), 455–465, doi:10.1007/s00382-007-0306-1. Joshi, M. M., F. H. Lambert, and M. J. Webb (2013), An explanation for the difference between twentieth and twenty-first century land-sea warming ratio in climate models, Clim. Dyn., 41(7–8), 1853–1869, doi:10.1007/s00382-013-1664-5. Klocke, D., J. Quaas, and B. Stevens (2013), Assessment of different metrics for physical climate feedbacks, Clim. Dyn., 41(5–6), 1173–1185, doi:10.1007/s00382-013-1757-1. Kraus, E. B. (1963), The diurnal precipitation change over the sea, J. Atmos. Sci., 20(6), 551–556. Lawrence, D. M., and J. M. Slingo (2005), Weak land-atmosphere coupling strength in HadAM3: The role of soil moisture variability, J. Hydrometeorol., 6(5), 670–680, doi:10.1175/JHM445.1. Manabe, S., and R. F. Strickler (1964), Thermal equilibrium of the atmosphere with a convective adjustment, J. Atmos. Sci., 21(4), 361–385. Manabe, S., R. J. Stouffer, M. J. Spelman, and K. Bryan (1991), Transient responses of a coupled ocean atmosphere model to gradual changes of atmospheric CO2. Part I. Annual mean response, J. Clim., 4(8), 785–818. Mauritsen, T., et al. (2012), Tuning the climate of a global model, J. Adv. Model. Earth Syst., 4, M00A01, doi:10.1029/2012MS000154. Mauritsen, T., R. G. Graversen, D. Klocke, P. L. Langen, B. Stevens, and L. Tomassini (2013), Climate feedback efficiency and synergy, Clim. Dyn., 41(9–10), 2539–2554, doi:10.1007/s00382-013-1808-7. Meraner, K., T. Mauritsen, and A. Voigt (2013), Robust increase in equilibrium climate sensitivity under global warming, Geophys. Res. Lett., 40, 5944–5948, doi:10.1002/2013GL058118. M€ obis, B., and B. Stevens (2012), Factors controlling the position of the intertropical convergence zone on an aquaplanet, J. Adv. Model. Earth Syst., 4, M00A04, doi:10.1029/2012MS000199. Muller, C. J., and I. M. Held (2012), Detailed investigation of the self-aggregation of convection in cloud-resolving simulations, J. Atmos. Sci., 69(8), 2551–2565, doi:10.1175/JAS-D-11-0257.1. Nordeng, T. E. (1994), Extended versions of the convective parametrization scheme at ECMWF and their impact on the mean and transient activity of the model in the tropics, ECMWF Tech. Memo. 206, Eur. Cent. for Medium-Range Weather Forecasts, Reading, U. K. Popke, D., B. Stevens, and A. Voigt (2013), Climate and climate change in a radiative convective equilibrium version of ECHAM6, J. Adv. Model. Earth Syst., 5, 1–14, doi:10.1029/2012MS000191. Ramanathan, V., and J. A. Coakley (1978), Climate modeling through radiative-convective models, Rev. Geophys. Space Phys., 16(4), 465– 489. Rochetin, N., B. R. Lintner, K. L. Findell, A. H. Sobel, and P. Gentine (2014), Radiative convective equilibrium over a land surface, J. Climate., doi:10.1175/JCLI-D-13-00654.1, in press. Roeckner, E., et al. (2003), The atmospheric general circulation model ECHAM5—Part I, Tech. Rep. 349, Max Planck Inst. for Meteorol., Hamburg, Germany. Soden, B. J., and I. M. Held (2006), An assessment of climate feedbacks in coupled ocean-atmosphere models, J. Clim., 19(14), 3354–3360, doi:10.1175/JCLI3799.1. Stevens, B., and S. Bony (2013), Water in the atmosphere, Phys. Today, 66(6), 29–34. Stevens, B., et al. (2013), Atmospheric component of the MPI-M Earth System Model: ECHAM6, J. Adv. Model. Earth Syst., 5, 146–172, doi: 10.1002/jame.20015. Sutton, R. T., B. Dong, and J. M. Gregory (2007), Land/sea warming ratio in response to climate change: IPCC AR4 model results and comparison with observations, Geophys. Res. Lett., 34, L02701, doi:10.1029/2006GL028164. Takahashi, K. (2012), Thermotidal and land-heating forcing of the diurnal cycle of oceanic surface winds in the eastern tropical Pacific, Geophys. Res. Lett., 39, L04805, doi:10.1029/2011GL050692. Taylor, K. E., R. J. Stouffer, and G. A. Meehl (2012), An overview of CMIP5 and the experiment design, Bull. Am. Meteorol. Soc., 93(4), 485– 498, doi:10.1175/BAMS-D-11-00094.1. Tobin, I., S. Bony, and R. Roca (2012), Observational evidence for relationships between the degree of aggregation of deep convection, water vapor, surface fluxes, and radiation, J. Clim., 25, 6885–6904. Tompkins, A. M., and G. C. Craig (1998a), Radiative-convective equilibrium in a three-dimensional cloud-ensemble model, Q. J. R. Meteorol. Soc., 124(550), 2073–2097, doi:10.1002/qj.49712455013. Tompkins, A. M., and G. C. Craig (1998b), Time-scales of adjustment to radiative-convective equilibrium in the tropical atmosphere, Q. J. R. Meteorol. Soc., 124(552), 2693–2713, doi:10.1256/smsqj.55207. Trenberth, K. E., and D. P. Stepaniak (2004), The flow of energy through the earth’s climate system, Q. J. R. Meteorol. Soc., 130(603), 2677– 2701, doi:10.1256/qj.04.83. Vial, J., J.-L. Dufresne, and S. Bony (2013), On the interpretation of inter-model spread in CMIP5 climate sensitivity estimates, Clim. Dyn., 41(11–12), 3339–3362, doi:10.1007/s00382-013-1725-9. Wetherald, R. T., and S. Manabe (1988), Cloud feedback processes in a general circulation model, J. Atmos. Sci., 45, 1397–1416.

BECKER AND STEVENS

C 2014. The Authors. V

1223