On the robustness of tropospheric adjustment in CMIP5 models

Authors


Abstract

[1] Effective radiative forcing associated with the tropospheric adjustment (TA) caused by atmospheric CO2 quadrupling (4×CO2) is quantified using fixed sea surface temperature experiments in CMIP5 models. Several features of TA in the tropics, partly related to weakening of the hydrological cycle, are found robust among the models: warming and drying of the free troposphere, near-surface moistening, strengthened stability in the lower troposphere, reduction in total cloud amount and shortwave cloud radiative effect (SWcld) over oceans. A positive SWcldchange prevailing in the tropical subsidence regime gives rise to large uncertainty in total effective radiative forcing, and is regulated quantitatively by changes in relative humidity (RH) that measure cloud decrease in the lower troposphere. A robust tropospheric warming primarily explains the RH decrease in the lower troposphere, but the change in specific humidity is largely model-dependent, which contributes to the inter-model spread of TA.

1. Introduction

[2] Equilibrium climate sensitivity, defined as an equilibrium response of global mean surface air temperature (SAT) to doubling of atmospheric CO2concentration, is a useful index for quantifying climate responses to external forcing simulated by different general circulation models (GCMs). The state-of-the-art GCMs participating in the Coupled Model Intercomparison Project version 5 (CMIP5) [Taylor et al., 2012] still have considerable diversity of equilibrium climate sensitivity [Andrews et al., 2012a], because of large spreads between models in climate feedback parameters and effective radiative forcing (RF). The latter is defined as radiative perturbation at the top of the atmosphere (TOA), without any changes to global mean SAT under the doubled CO2 condition. The effective RF includes the effects of rapid responses in stratosphere and troposphere to CO2 increase on the radiative balance at TOA, referred to as stratospheric adjustment and tropospheric adjustment (TA), respectively [Hansen et al., 2005; Knutti and Hegerl, 2008; Andrews et al., 2012b]. While the stratospheric adjustment is qualitatively robust among the models, cloud responses associated with TA is known to be the major contributor to uncertainty in effective RF [Gregory and Webb, 2008, hereinafter GW08; Webb et al., 2012].

[3] Changes in the atmospheric thermodynamic structure (temperature, humidity, and cloud) are key factors for TA and associated change in the shortwave component of cloud radiative effect (SWcld, defined as shortwave radiation difference at TOA between all sky and clear sky). Some modeling studies have shown decreasing relative humidity (RH) in the troposphere and associated reduction in total cloud fraction (Ctotal), producing positive change of SWcldSWcld) in TA [e.g., Dong et al., 2009; Colman and McAvaney, 2011, hereinafter CM11]. GW08 and Webb et al. [2012]compared effective RF simulated among CMIP3 and Cloud Feedback Model Intercomparison Project phase 1 (CFMIP1) models. They found that 1) TA-induced ΔSWcld varies greatly among models; 2) the number of models simulating positive ΔSWcldis greater than those simulating negative; 3) most models simulate increases in lower tropospheric stability (LTS); and 4) the inter-model spread of ΔSWcldin high-LTS regimes over the ocean has a great impact on uncertainty in net effective RF in CMIP3/CFMIP1 models.

[4] Kamae and Watanabe [2012, hereinafter KW12] explained the mechanisms of TA using one of the CMIP5 models, MIROC5 [Watanabe et al., 2010]. In response to instantaneous RF, the troposphere warms, which accompanies a strengthening of LTS and increase of surface RH over oceans. The latter suppresses evaporation (E), balanced by precipitation (P) decrease [Cao et al., 2012] and a shoaling of the marine boundary layer [Watanabe et al., 2012; Wyant et al., 2012]. These responses occur rapidly in a few days, with a resultant contrast between drying above the boundary layer and the near-surface moistening. The former dominates the latter, leading to negative and positive changes inCtotalCtotal) and SWcld, respectively.

[5] The aforementioned mechanisms have yet to be confirmed using CMIP5 multi-model ensemble (MME). Although some modeling studies showed positive ΔSWcld associated with the tropospheric drying and negative ΔCtotal [Dong et al., 2009; CM11; KW12], model comparison works report that there exist models showing the opposite signs of ΔSWcld and/or ΔCtotal [GW08; Wyant et al., 2012; Webb et al., 2012]. It is therefore worth examining the robustness and sources of spread in TA across CMIP5 models for understanding mechanisms that determine effective RF. One way to estimate the effective RF and TA is fixed sea surface temperature (SST) method [Hansen et al., 2005] in which atmospheric GCMs (AGCMs) are run with fixed SST and sea ice but with different CO2 concentrations. It should be noted that the land surface warms in response to CO2 increase in this method and it has some impacts on dynamical circulation, hydrological cycle, and spatial pattern of cloud radiative effect [GW08; Dong et al., 2009; Bala et al., 2010; Lambert et al., 2011; Wyant et al., 2012; Andrews et al., 2012b] although they are not essential factors for effective RF and associated TA (KW12). In this study, we evaluate effective RF and TA in CMIP5 models using results of fixed SST experiments.

2. Models and Experiments

[6] We used the multi-model dataset archived under the CMIP5 umbrella [Taylor et al., 2012]. For calculating effective RF, we used fixed-SST control and 4×CO2 experiments (sstClim and sstClim4×CO2 in the CMIP5 experiment table) [Taylor et al., 2012]. These were conducted with AGCMs, driven by climatological SST and sea-ice concentration derived from pre-industrial control simulations in each model. Atmospheric CO2 concentrations in sstClim and sstClim4×CO2 were set to 280 and 1120 ppmv, respectively. We calculated differences in climatology between the two runs, averaged over 30 years. Many CMIP5 models have also conducted control and 4×CO2experiments by prescribing SST and sea-ice with observations for 1979–2008 (amip and amip4 × CO2). The differences in climatology between amip and amip4×CO2 are generally consistent with those between sstClim and sstClim4×CO2 (figures not shown). We used outputs from 12 models for sstClim 4×CO2experiments and one model (CNRM-CM5) for amip 4×CO2 experiments (Table S1 in the auxiliary material).

3. Effective Radiative Forcing and Tropospheric Adjustment

3.1. Global-Mean Radiation Budgets

[7] First, we compare global-mean effective RF among different models and ensembles.Figure 1shows multi-model effective RF, simulated in amip and sstClim 4×CO2 experiments. Results from the CMIP3/CFMIP1 ensemble, derived from GW08, are also plotted. The effective RFs and its components in CMIP5/CFMIP2 ensembles are scaled to be 2 × CO2 equivalent. All components of effective RF are generally consistent among three ensembles, but some differences are apparent. Changes in net effective RF (ΔNet) and shortwave component of clear-sky flux (ΔSWclear) calculated by CMIP3/CFMIP1 models tend to be smaller than those of CMIP5/CFMIP2 ensembles. This difference mostly arises from different methods of calculating effective RF, and is consistent with Andrews et al. [2012b] and KW12. In the CMIP3/CFMIP1 ensemble, effective RF is estimated using a regression method [Gregory et al., 2004] applied to 2 × CO2 experiments with AGCMs coupled to a slab ocean, which tends to produce smaller ΔSWclearover the Arctic Ocean relative to the fixed-SST method (GW08).

Figure 1.

Global-mean 2 × CO2 net effective RF (ΔNet) and its components (W m−2) in CMIP3 and CMIP5 ensembles. Effective RFs in CMIP3 models (red) calculated by the regression method (using 2 × CO2experiment with atmosphere and slab-ocean coupled GCMs) ofGregory et al. [2004] are derived from GW08. CMIP5 data are calculated by fixed-SST method using AGCMs (black: amip 4×CO2 experiments; blue: sstClim 4×CO2 experiments; see Table S1). Effective RFs are scaled to be 2 × CO2 equivalent.

[8] In two CMIP5 ensembles (amip and sstClim 4×CO2 runs), effective RFs agree with each other, but minor differences are present owing to different models used in the two ensembles (Table S1). In the individual models, effective RFs in the two runs are very similar (figures not shown). Table 1shows effective RF in 13 models of the CMIP5 ensemble. The inter-model spread of ΔNet follows that of ΔSWcld, which has the largest spread among the components. Change in clear-sky longwave radiation (ΔLWclear) associated with land surface warming and thermodynamic structures in troposphere and stratosphere also has a large spread. In the CMIP5 ensemble, all but one model show positive ΔSWcld. As a part of TA, P and E decrease by 0.10 to 0.18 mm day−1, consistent with previous studies [e.g., Allen and Ingram, 2002; Lambert and Webb, 2008; Bala et al., 2010]. All models have perfect compensations between P and E changes. Spatial distributions of ΔNet, ΔLWcld and ΔSWcld, and land- and ocean-mean values averaged in low latitudes (30°S–30°N) in the individual models are presented inFigures S1–S3 and Table S2. Both ΔLWcld (negative) and ΔSWcld(positive) over the ocean are dominant in global-mean values and most of their signs are consistent, indicating that global-mean cloud radiative effects associated with TA is mainly attributable to changes over the ocean. Changes in vertical pressure velocity at 500 hPa (Δω500), an index of dynamical motion associated with land-sea thermal contrast in TA [Lambert et al., 2011; Wyant et al., 2012], are negative (upward) and positive (downward) over land and ocean in low latitude, respectively. The positive ΔLWcld and negative ΔSWcld over land (Figures S2 and S3) associated with the changes in dynamical motion and increasing cloud amount over land (Table S2) weaken the global-mean radiative effects of tropospheric cloud adjustment (KW12). P and E over ocean are suppressed but P increases over land (Figures S4 and S5). The latter may be associated with CO2 physiological forcing [Andrews et al., 2012b, and references therein]. These results indicate that TA and associated cloud radiative effects revealed by KW12 are generally consistent among CMIP5 models.

Table 1. Global-Mean Effective Radiative Forcing (RF) and Adjustment of Precipitation (ΔP) and Evaporation (ΔE) due to 4×CO2a
 ΔNet [W m−2]ΔLWclear [W m−2]ΔSWclear [W m−2]ΔLWcld [W m−2]ΔSWcld [W m−2]ΔP [mm day−1]ΔE [mm day−1]
  • a

    Experiments for all models except CNRM-CM5 are climatological SST experiments (sstClim4×CO2 minus sstClim). Values of CNRM-CM5 are calculated by AMIP experiment (amip4×CO2 minus amip). Model uncertainties are ±1 standard deviations in inter-annual variations. The uncertainties for the ensemble mean are ±1 standard deviations in 13 models.

CCSM48.84 ± 0.228.15 ± 0.150.60 ± 0.14−1.30 ± 0.091.39 ± 0.22−0.18 ± 0.01−0.18 ± 0.01
CNRM-CM57.86 ± 0.176.86 ± 0.111.04 ± 0.07−1.75 ± 0.101.72 ± 0.18−0.16 ± 0.01−0.16 ± 0.01
CSIRO-Mk3.6.06.20 ± 0.217.54 ± 0.100.10 ± 0.04−2.00 ± 0.130.54 ± 0.18−0.10 ± 0.01−0.10 ± 0.01
CanESM27.35 ± 0.247.00 ± 0.120.39 ± 0.11−1.20 ± 0.081.15 ± 0.18−0.14 ± 0.01−0.14 ± 0.01
HadGEM2-A7.00 ± 0.267.22 ± 0.140.11 ± 0.25−1.37 ± 0.091.25 ± 0.16−0.18 ± 0.01−0.18 ± 0.01
INM-CM46.24 ± 0.197.10 ± 0.110.29 ± 0.12−0.75 ± 0.10−0.40 ± 0.20−0.11 ± 0.01−0.11 ± 0.01
IPSL-CM5A-LR6.49 ± 0.286.05 ± 0.160.99 ± 0.07−2.61 ± 0.112.05 ± 0.20−0.16 ± 0.01−0.16 ± 0.01
MIROC57.94 ± 0.227.94 ± 0.100.41 ± 0.06−1.80 ± 0.121.39 ± 0.20−0.15 ± 0.01−0.15 ± 0.01
MPI-ESM-LR8.63 ± 0.298.00 ± 0.110.42 ± 0.06−1.56 ± 0.071.77 ± 0.26−0.13 ± 0.01−0.13 ± 0.01
MPI-ESM-MR8.59 ± 0.257.94 ± 0.100.43 ± 0.07−1.40 ± 0.101.63 ± 0.26−0.13 ± 0.01−0.13 ± 0.01
MPI-ESM-P8.60 ± 0.187.92 ± 0.100.45 ± 0.08−1.50 ± 0.081.72 ± 0.21−0.13 ± 0.01−0.14 ± 0.01
MRI-CGCM37.19 ± 0.207.17 ± 0.150.88 ± 0.10−1.20 ± 0.100.35 ± 0.14−0.12 ± 0.01−0.12 ± 0.01
NorESM1-M6.97 ± 0.286.82 ± 0.160.37 ± 0.10−1.30 ± 0.091.39 ± 0.22−0.15 ± 0.01−0.15 ± 0.01
Ensemble7.53 ± 0.897.36 ± 0.380.50 ± 0.09−1.52 ± 0.211.23 ± 0.47−0.14 ± 0.00−0.14 ± 0.00

3.2. Change in Thermodynamic Structure and its Dependence on Dynamical Regime

[9] In any set of MMEs shown in Figure 1, the global-mean cloud radiative effects associated with TA are greatly controlled by changes over low-latitude (30°S–30°N) oceans (hereafter LLO;Table S2 and Figures S2 and S3) [Webb et al., 2012]. We analyzed dependencies of cloud radiative effects over LLO associated with TA on dynamical circulation regimes [Wyant et al., 2012; KW12] in the CMIP5 models. Figure 2shows multi-model ΔCtotal, ΔSWcld, and ΔLWcld over LLO sorted by ω500 in the control simulation, following the method of Bony et al. [2004]. Although inter-model spread as defined by the standard deviation (σ) around the MME average is large (shading in Figure 2), most of the models show the same signs of ΔCtotal, ΔSWcld, and ΔLWcld in all circulation regimes. MME averages are negative for ΔCtotal (−0.9 to −1.5%; Figure 2a) and ΔLWcld (−1.5 to −3.9 W m−2; Figure 2c), and positive for ΔSWcld (1.5 to 3.0 W m−2; Figure 2b) in all regimes, but the spreads of ΔCtotal and ΔSWcld are larger in the subsidence regime of ω500 > 0 (σ = 0.8–1.2% and 1.0–1.9 W m−2) than in the convective regime. Note that ΔSWcld and ΔLWcldinclude cloud masking effects, because it is defined here by the difference between all-sky and clear-sky radiative fluxes at TOA [Soden et al., 2008; Andrews et al., 2012b; Wyant et al., 2012]. A large part of the negative ΔLWcld may be attributed to cloud masking effect prevailing in the convective regimes over LLO [Wyant et al., 2012].

Figure 2.

Regime composite of (a) ΔCtotal (%), (b) ΔSWcld, and (c) ΔLWcld (W m−2) over low-latitude ocean (LLO; 30°S–30°N) with respect toω500 (hPa day−1) in CMIP5 fixed-SST 4 × CO2 experiments. Thick black curve and shading indicate MME average and one standard deviation, respectively.

[10] It is plausible that changes of the thermodynamic structure in the troposphere have essential roles for ΔCtotal and associated ΔSWcld over LLO in TA (CM11; KW12). Figures 3a–3dshow vertical structures of MME-mean ΔT and ΔRH, sorted by ω500. The pattern and magnitude of ΔT are robust, showing that the troposphere warm, with a maximum in the lower troposphere (700–850 hPa). This is consistent with the height of maximum instantaneous radiative heating caused by CO2 increase [Collins et al., 2006; KW12]. The lower tropospheric warming (0.4 K on average around 700–850 hPa; Figure 3b) has little inter-model spread, and acts to decrease lower-tropospheric RH across the subsidence regime (−1.5% on average;Figures 3c and 3d). In contrast, ΔRH over LLO is robustly positive about 0.5% near the surface (925–1000 hPa levels; Figures 3c and 3d) because of an increase in water vapor (detailed below). The increase in near-surface RH and associated declines inE and P (Table 1 and Figures S4 and S5) are consistent with previous studies [Cao et al., 2012; KW12]. The warming-induced tropospheric drying is stronger than the near-surface wetting, resulting in negative ΔCtotal in most models (Figure 2a and Table S2; KW12).

Figure 3.

(a) Regime composite of MME in ΔT (K), (c) ΔRH (%), (f) ΔRH(ΔT) (%), and (h) ΔRH(Δq) (%) over LLO. Hatching denotes regions where 11 out of 13 models agree on sign of the value plotted. (b, d, g, i) MME average (thick curve) and standard deviation (shading). (e) Scatter diagram of ΔSWcld over LLO against ΔRH averaged between 700 and 925 hPa levels in subsidence regime (ω500 > 0 hPa day−1; red rectangle in Figure 3c). Colors of filled circles correspond to those in Figure 2.

[11] The inter-model spread of ΔRH around 700–925 hPa, averaged over the subsidence regime (ΔRHsub,low; red rectangle in Figure 3c), is significant in MME (Figure 3d). Figure 3e shows the relationship between ΔRHsub,low and ΔSWcld over LLO among models. ΔRHsub,low and ΔSWcldover LLO have large inter-model spreads (ranging from −1.9 to −0.4% and −0.2 to 3.0 W m−2, respectively), which are related to each other. That is, models with larger negative ΔRHsub,low tend to have larger positive ΔSWcld (and larger negative ΔCtotal; not shown) over LLO; the correlation coefficient R between the two (ΔRH and ΔSWcld) averaged over the subsidence regime reaches −0.69. Even for all the circulation regimes, ΔRHsub,lowcan faithfully measure inter-model variance of ΔCtotal and ΔSWcld over LLO (R between ΔRHsub,low and ΔSWcld is −0.64; cf. Figure 3e), because inter-model spreads of ΔCtotal and ΔSWcld over the subsidence regime are the major contributors to those over LLO (Figure 2).

[12] Since RH is determined by T and specific humidity q, ΔRH can be divided into two components: ΔT-driven and Δq-driven changes. We write ΔRH ≈ ΔRH(ΔT) + ΔRH(Δq), where the first and second terms denote perturbations of RH with q and T set to values in the control run, indicating contributions of ΔT and Δq to ΔRH, respectively. We confirmed that the residual term is negligibly-small relative to them (figures not shown). ΔRH(ΔT) (Figure 3f) indicates robust drying with little inter-model spread, because of robust positive ΔT (Figure 3a) peaking in the lower and upper troposphere. In contrast, ΔRH(Δq) shows robust increases near the surface and upper troposphere, but is not robust in the middle and lower troposphere in terms of sign and magnitude (Figures 3h and 3i). Figure 4 shows ΔRHsub,low and its components; ΔRHsub,low(ΔT) and ΔRHsub,low(Δq) in individual CMIP5 models. It is evident that the robustness of the negative ΔRHsub,low is accomplished by the robust warming, ΔRHsub,low(ΔT) (orange bars). Diversity in the magnitude of ΔRHsub,low is largely due to ΔRHsub,low(Δq) (blue bars), consistent with Figure 3i. This indicates that models with greater increases (decreases) in lower tropospheric q tend to have smaller (larger) negative ΔRHsub,low. Further evidence is provided by high correlation (R = −0.68) between ΔRHsub,low(Δq) and ΔSWcld, averaged over LLO. These results strongly suggest that various responses in lower-troposphericq across the subsidence regime are key ingredients for narrowing uncertainty in effective RF owing to quadrupling CO2 concentration.

Figure 4.

Decomposition of ΔRH (%, gray) averaged between 700 and 925 hPa levels in subsidence regime (ω500 > 0 hPa day−1) over LLO, into ΔRH(ΔT) (%, orange) and ΔRH(Δq) (%, blue) components. The models are sorted from left to right by decreasing ΔRH.

4. Summary and Discussion

[13] Major parts of effective RF and TA simulated by fixed-SST 4×CO2 experiments were qualitatively consistent across CMIP5 models. The robust processes are confirmed as: 1) tropospheric warming and strengthened vertical stability in the lower troposphere over oceans; 2) positive and negative ΔRH in near surface and lower troposphere; 3) negative ΔEfrom sea surface which is consistent with the positive near-surface ΔRH; and 4) drying of the lower troposphere and associated negative ΔCtotal and positive ΔSWcld that are consistent with positive ΔT and negative ΔE.Inter-model spreads in ΔCtotal and ΔSWcld, large over the subsidence regime, are significantly correlated with spread in ΔRH at 700–925 hPa levels over the subsidence regime. This ΔRH spread is explained by varying Δq in terms of sign and magnitude, which therefore appears as the major source of uncertainty in effective RF.

[14] CM11 discussed that warming by instantaneous RF is essential but not sufficient for drying of the middle and lower troposphere in TA. The result of this study reveals that the intensities of the lower tropospheric drying and associated ΔSWcld are controlled by Δq in GCMs. However, physical mechanisms determining Δqin the lower troposphere were not clarified here. Processes related to the marine boundary layer over cool subtropical oceans, e.g., surface evaporation, turbulent mixing, shallow convection, and large-scale subsidence, would behave differently to the CO2 increase in different models. In addition, there are models that show responses in ΔCtotal (positive) and ΔSWcld (negative) opposite to those in MME averages (Figure 2). TA and associated effective RF may also be regulated by the reproducibilities of q and Cvertical profiles in the model's control simulations. One model generating opposite responses (INM-CM4) has considerable biases in cloud ice and water relative to A-Train satellite data [Jiang et al., 2012]. Cloud responses simulated by such models should be evaluated carefully. Over the strong subsidence regime (ω500 > 50 hPa day−1), another model (MRI-CGCM3) reveals positive ΔCtotal, opposite to the MME average (Figure 2a). Wyant et al. [2012]reported that TA-induced change of low cloud fraction over cool subtropical oceans in their cloud resolving model strongly depended on the horizontal resolution. Further studies on responses to CO2 increase of boundary layer depth, strength of inversion capping the boundary layer, plus vertical profiles of q and C and their reproducibilities within individual models in a subsidence regime may provide insight into uncertainty of cloud adjustment.

[15] In the present study, we mainly focused on adjustments of RH, Ctotal and SWcld, because of limited data availability of the CMIP5 archive as of this writing. It is worth evaluating the relative contribution of different types and properties of clouds (amount, altitude, and optical thickness) [Zelinka et al., 2012] to effective RF, and their relationship to thermodynamic structure among multi-models. The robustness and source of spread of TA should also be investigated using perturbed parameter ensembles of single GCMs [e.g.,Webb et al., 2012; Shiogama et al., 2012], which can measure a different type of uncertainty than CMIP5 MME. Such efforts may advance our understanding of RF and feedback to CO2 increase.

Acknowledgments

[16] We acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1) for producing and making available their model output. For CMIP5, the US Department of Energy's Program for Climate Model Diagnosis and Intercomparison provided coordinating support, and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This work was supported by the Program for Risk Information on Climate Change (PRICC) and Grants-in-Aid 23310014 and 23340137 from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.

[17] The Editor thanks Lorenzo Tomassini and an anonymous reviewer for their assistance in evaluating this paper.