Abstract

This study represents an initial effort in the context of the coupled atmosphere–surface climate feedback-response analysis method (CFRAM) to partition the temporal evolution of the global surface temperature from 1981 to 2005 into components associated with individual radiative and nonradiative (dynamical) processes in the NCAR CCSM4’s decadal hindcasts. When compared with the observation (ERA-Interim), CCSM4 is able to predict an overall warming trend as well as the transient cooling occurring during the period 1989–94. However, while the model captures fairly well the positive contributions of the CO2 and surface albedo change to the temperature evolution, it has an overly strong water vapor effect that dictates the temperature evolution in the hindcast. This is in contrast with ERA-Interim, where changes in surface dynamics (mainly ocean circulation and heat content change) dominate the actual temperature evolution. Atmospheric dynamics in both ERA-Interim and the model work against the surface temperature tendency through turbulent and convective heat transport, leading to an overall negative contribution to the evolution of the surface temperature. Impacts of solar forcing and ozone change on the surface temperature change are relatively weak during this period. The magnitude of cloud effect is considerably smaller compared to that in ERA-Interim and the spatial distribution of the cloud effect is also significantly different between the two, especially over the equatorial Pacific. The value and limitations of this process-based temperature decomposition are discussed.

1. Introduction

Decadal prediction targets near-term (10–30 yr) climate evolutions and is critical for the assessment of climate change risks and the design of mitigation and adaptation strategies. Different from century-scale climate projections, a model’s decadal hindcasts can be rigorously verified against observations (Doblas-Reyes et al. 2013). Among many challenges, representations of various climate feedbacks in a coupled model remain the largest source of uncertainty in decadal prediction (Bellucci et al. 2015a) and model initialization is thought to be at the heart of the decadal prediction problem (Meehl et al. 2009).

Phase 5 of the Coupled Model Intercomparison Project (CMIP5) includes decadal hindcast experiments to assess the predictability of climate across decadal time scales and to evaluate the skill of decadal prediction with the state-of-the-science climate models (Taylor et al. 2012). In these experiments, different methods of ocean initialization (e.g., full-field and anomaly) are explored (Bellucci et al. 2015b; Hazeleger et al. 2013a; Meehl et al. 2009; Meehl et al. 2014b; Smith et al. 2013; Thoma et al. 2015) and the relative importance of initialization versus external forcing in different periods is examined (Branstator and Teng 2012; Corti et al. 2015). It has been reported that model runs with initialized ocean states tend to outperform the uninitialized historical climate runs (Branstator and Teng 2012; Fyfe et al. 2011; García-Serrano and Doblas-Reyes 2012; Meehl and Teng 2014a; Yeager et al. 2015). The models demonstrate reasonable skill at predicting major decadal climate transitions (e.g., the mid-1970s shifts and the early-2000s hiatus) (Meehl and Teng 2012, 2014a,b; Meehl et al. 2014a) and tend to derive the skill of prediction from major modes of decadal variability including the Atlantic multidecadal oscillation (AMO) and the Pacific decadal oscillation (PDO) (Chikamoto et al. 2012; García-Serrano and Doblas-Reyes 2012; García-Serrano et al. 2015; Hazeleger et al. 2013b; Huang et al. 2015; Kim et al. 2012; Msadek et al. 2014; Pohlmann et al. 2013; Swingedouw et al. 2013). Although mainly determined by the presence of long-term trends (Bellucci et al. 2013; Choi et al. 2016), the skill of prediction is also found to be higher over the North Atlantic and Indian Oceans than over the North Pacific (Bellucci et al. 2013; Bombardi et al. 2015; Doblas-Reyes et al. 2013; Guemas et al. 2012; Guemas et al. 2013; Hermanson et al. 2014; Meehl et al. 2009; Meehl et al. 2014b). With these findings, Bellucci et al. (2015a) suggest that more attention should be given to the roles of sea ice, land surface, stratosphere, and aerosols in decadal prediction and the initializations of these fields; however, there remain major challenges.

Despite the substantial efforts devoted to evaluating the CMIP5 decadal hindcasts, we are still missing in the literature a systematic examination of both the regional-scale and global surface temperature evolutions in these experiments. Specific questions to ask include the following: 1) Do the CMIP5 models’ decadal hindcasts capture the interannual variations and decadal trends in the surface temperature? 2) If a model predicts such temperature evolutions well, does such a performance originate from accurate representations of various radiative and dynamical processes determining surface temperature fluctuations or does it actually benefit from a cancellation of errors among misrepresented processes? 3) If the model prediction differs substantially from the observation, what problems at the process level are the main culprits? As an initial effort toward addressing these questions, here we use the coupled atmosphere–surface climate feedback-response analysis method (CFRAM; Cai and Lu 2009; Lu and Cai 2009) to quantify and compare with reanalysis-based observations the relative contributions of multiple physical and dynamical processes to surface temperature evolutions in the decadal hindcasts conducted with the NCAR Community Climate System Model version 4 (CCSM4). CFRAM has proven to be an efficient offline diagnostic tool and has provided new insights into the formation of surface temperature anomalies associated with internal modes of variability (e.g., ENSO and the northern annular mode) and the nature of regional surface temperature biases in climate models (Deng et al. 2012, 2013; Hu et al. 2016; Park et al. 2012). Following this introduction, section 2 provides a brief description of the CCSM4 decadal hindcast experiment and observational data used and also outlines the CFRAM method. The main results are reported and discussed in section 3. Section 4 gives the concluding remarks.

2. Data and methods

The decadal hindcasts conducted with the NCAR CCSM4 are the main evaluation target in this analysis (Gent et al. 2011). The CCSM4’s skill of predicting various ocean and climate transitions has been examined in earlier studies (Karspeck et al. 2015; Meehl and Teng 2012; Meehl et al. 2016; Yeager et al. 2012). The CCSM4 model consists of four components (atmosphere, ocean, land, and sea ice). The atmospheric component [i.e., the Community Atmosphere Model version 4 (CAM4)] has a finite-volume dynamical core with a nominal 1° horizontal resolution (0.9° × 1.25°) and 26 vertical hybrid levels. The ocean model is the Parallel Ocean Program (POP) with a nominal 1° horizontal resolution and 60 vertical levels. The land and sea ice model share the same horizontal grid with the atmosphere and ocean model, respectively. The 30-yr decadal hindcast we used is initialized in 1981 and includes 10 ensemble members providing predictions up to 2010 (https://www.earthsystemgrid.org). The ensemble mean is used in our analysis. More details of the design of these CMIP5 “near-term” prediction experiments can be found in Taylor et al. (2012). The hindcasts for the period 1981 to 2005 were analyzed here since it made use of the historical, observation-based external forcing data (e.g., greenhouse gas concentrations, solar irradiance, ozone, and volcanic aerosols). The initial conditions of the ocean and sea ice come from a forced ocean–sea ice simulation designed to reproduce the evolution of the ocean and sea ice states from the start of 1948 through the end of 2007 (here the initial conditions are the results of year 1981; Yeager et al. 2012).

The observation used for prediction verification is the European Centre for Medium-range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim; ECMWF 2009; Dee et al. 2011). ERA-Interim is one of the latest global atmospheric reanalysis covering the period 1979 to the present and has a 0.75° horizontal resolution with 37 pressure levels in the vertical ranging from 1000 to 1 hPa. It is important to note that reanalysis data blend real observations with model simulations and variables in the ERA-Interim with varying degrees of uncertainty depending on how much observational constraint has been imposed. For example, cloud fields in the reanalysis are strongly influenced by microphysics and cumulus parameterization schemes used in the model for reanalysis production and are therefore highly model dependent.

The main diagnostic tool, CFRAM, is based upon the total energy balance in an atmosphere–surface column that consists of M atmospheric layers and one surface layer (Cai and Lu 2009; Lu and Cai 2009). Writing the total energy balance equation separately for two climate states (climate states A and B) and then taking the difference (Δ) between them (i.e., state B − state A), we obtain

 
formula

where E is the vertical profile of the total energy for the atmosphere–surface column including dry static energy, latent heat, and kinetic energy; R is the vertical profile of the divergence of longwave radiative flux; and S is the vertical profile of the convergence of shortwave radiative flux. For the atmospheric layers in the column, represents the vertical profile of the convergence of total energy due to turbulent, convective, and large-scale atmospheric motions. For the surface layer of the column, over ocean surfaces, corresponds to the net convergence of energy due to oceanic motions of all scales while over land/snow/ice surfaces represents the energy gain in the column due to soil heat diffusion, runoff, and snow/ice melting and freezing. The term represents the rate of energy and heat storage change, which is negligible in the atmosphere if annual mean states are considered but can be substantial in the ocean. All terms in Eq. (1) have units of watts per meter square.

Linearizing the radiative energy perturbations by neglecting the interactions of effects among different radiative-active species, we may express and as the sum of partial radiative energy flux convergence/divergence due to individual processes:

 
formula

In Eq. (2), the superscripts SR, CLD, WV, O3, AL, and CO2 stand for solar irradiance, cloud, water vapor, ozone, surface albedo, and CO2, respectively. Elements of are the temperature differences in each layer between the two climate states being evaluated, and is the Planck feedback matrix quantifying the change of the vertical profile of longwave radiative flux divergence due to changes in the atmospheric and surface temperatures. Substituting Eq. (2) into Eq. (1), multiplying both sides by , and rearranging the terms, we obtain

 
formula

where at the surface layer, representing the sum of the changes in the surface turbulent latent and sensible heat flux, and in the atmospheric layers, which is estimated (in the actual calculations) as a residual from the radiative energy perturbations, that is, . Similarly, and it is zero in all the atmospheric layers. Equation (3) allows us to express the vertical profile of the temperature difference at a given location between two climate states as the sum of the vertical profile of partial temperature differences due to (from left to right) changes in solar irradiance, CO2, cloud, water vapor, ozone, surface albedo, atmospheric dynamics (i.e., energy transport by atmospheric motions of all scales), and surface dynamics (including mainly energy transport by oceanic motions of all scales and ocean heat storage change).

With the Fu–Liou radiative transfer model (Fu and Liou 1992, 1993), individual radiative energy perturbations in Eq. (3), for example , are obtained by conducting “off-line” radiative transfer calculations separately for climate state A (the “base state”) and a perturbed state where all the atmospheric/surface properties are kept the same as in climate state A except that cloud properties are replaced with those in climate state B. The difference in the net radiative heating between the perturbed state and the base state gives us , that is, changes in radiative heating due to changes in clouds alone. The Planck feedback matrix is estimated by conducting multiple radiative transfer calculations documenting the changes in the vertical profile of the longwave radiative heating due to 1-K warming (with respect to the base state temperature) at an atmospheric or surface layer. Finally, given the way the CFRAM calculation is conducted, the interpretation of both the atmospheric dynamics and surface dynamics terms needs some extra caution. Since both terms are estimated as residuals from the equation of atmospheric or surface energy balance, they inevitably include errors associated with the offline radiative transfer calculation and linearization done in the CFRAM. Even though dynamical energy transports and changes in heat storage (especially over ocean) tend to dominate these two terms over a global scale, it must be recognized that processes not included in the current CFRAM analysis also contribute to these two terms and such contributions could be significant at local and regional scales. For example, atmospheric dynamics includes also effects of aerosols and trace gases that are not explicitly considered in the CFRAM (e.g., CH4, N2O).

The input variables required by the offline radiative transfer calculations include solar irradiance at the top of the atmosphere (TOA), air/surface temperatures, specific humidity, cloud amount, cloud liquid/ice water content, CO2 mixing ratio, ozone mixing ratio, surface albedo, and surface latent/sensible heat flux. All of these fields are obtained separately from the CCSM4 decadal hindcast outputs and ERA-Interim except for the solar irradiance and CO2 mixing ratio, where values from the historical time-dependent external forcings are used (Taylor et al. 2012). We consider the time-mean state of 1981–85 (denoted as year 1983 hereafter) as the base state, and subsequent segments (with 1-yr increments) of 5-yr time mean state as the climate states to contrast with the base state. In other words, 1981–85 is the base state (climate state A, as discussed previously) with multiple cases of climate state B starting from 1982–86 (denoted as year 1984 hereafter), 1983–87 (year 1985), and so on until 2001–05 (year 2003). For each climate state, composite 3D and 2D fields used in radiative transfer calculations are constructed separately with the ERA-Interim data (observation) and the ensemble mean of the CCSM4 decadal hindcast outputs (hindcast). Partial temperature differences between multiple instances of state B and the base state are obtained through Eq. (3) for both the observation and the model hindcast. The completion of the CFRAM analysis thus provides temporal evolutions of the atmospheric and surface temperatures associated with individual physical and dynamical processes (i.e., partial temperature changes) throughout the period 1981–2005. The agreement (discrepancy) of these partial temperature changes between the model hindcast and the observation serves as direct measures of the model fidelity (errors) in representing climate processes that are critical in decadal-scale climate variations and changes.

3. Results

a. Validation of the CFRAM calculations

Figure 1 shows the net downward radiation differences (W m−2) at the top of the atmosphere (TOA) and the surface between the period 2001–05 and the base state (1981–85) based on the ERA-Interim surface flux data and CFRAM calculations. Figures 1a and 1d are constructed using the surface flux data from the ERA-Interim. The middle and right columns of Fig. 1 are the same quantities obtained respectively through the CFRAM offline radiative transfer calculation, and as the sum of partial radiative energy perturbations following

 
formula
 
formula

where represents the offline radiative transfer calculation utilizing the time-mean atmospheric and surface properties over the two periods (2001–05 and 1981–85) and is the sum of partial radiative energy perturbations associated with six individual processes obtained through linearization (see section 2 for details). The difference between the middle and left column in Fig. 1 therefore measures errors associated with the use of offline calculation while the difference between the right and middle column quantifies errors related to the linearization process. It is evident that the spatial patterns of the TOA and surface net radiation differences obtained in these three ways generally resemble each other (i.e., Fig. 1a vs. Figs. 1b,c, and Fig. 1d vs. Figs. 1e,f). Larger magnitudes of offline calculation errors (differences between Figs. 1b and 1a, or between Figs. 1e and 1d) can be seen in the tropics (the central-eastern Pacific in particular), southeastern Atlantic, and South America. These errors are primarily due to the use of time-mean cloud properties in radiative transfer calculations, which tend to overestimate cloud effects (Sejas et al. 2014; Song et al. 2014a,b; Hu et al. 2017). The errors associated with linearization (differences between Figs. 1c and 1b, or between Figs. 1f and 1e) are much smaller compared to the errors associated with the offline calculation, exhibiting slightly positive values over the high latitudes. This error evaluation suggests that the use of time-mean atmospheric and surface properties in radiative transfer calculations and the linearization of radiative energy perturbations in CFRAM are reasonable approximations to make for the purpose of this analysis.

Fig. 1.

(top) Top of atmosphere and (bottom) surface net downward radiation differences (W m−2) between 2001–05 and 1981–85 (a),(d) obtained directly from the ERA-Interim, (b),(e) obtained from the CFRAM offline calculation, and (c),(f) calculated as the sum of the six radiative energy perturbations associated with the linearization of the CFRAM.

Fig. 1.

(top) Top of atmosphere and (bottom) surface net downward radiation differences (W m−2) between 2001–05 and 1981–85 (a),(d) obtained directly from the ERA-Interim, (b),(e) obtained from the CFRAM offline calculation, and (c),(f) calculated as the sum of the six radiative energy perturbations associated with the linearization of the CFRAM.

b. Temporospatial structure of the process-based temperature attributions

With the CFRAM calculations validated, next we use the derived partial temperature differences to examine the relative contributions of the major physical and dynamical processes to the evolutions of the surface temperature in the period 1981–2005 at both global and regional scales. Figure 2a displays the observed temporal evolution of the globally averaged surface temperature [as departures from the average of the base state (1981–85)] and Figs. 2b–i are the corresponding partial temperatures in association with changes in solar forcing, ozone, CO2, water vapor, surface albedo, clouds, surface dynamics, and atmospheric dynamics. An overall warming trend is clear in the temperature record with a minor “transient” cooling occurring from 1989 to 1994 reflecting likely the effect of the Pinatubo eruption in the early 1990s (Meehl and Teng 2012). The partial temperature results suggest significant contributions to this cooling from changes in atmospheric dynamics (Fig. 2i) and water vapor (Fig. 1e). Note that the radiative energy perturbations due to aerosol changes are not explicitly isolated in the current CFRAM calculation, but they are included in the atmospheric dynamics term, which is estimated as residuals from other radiative energy perturbations listed in Eq. (3). Atmospheric dynamics as the main contributor to the cooling thus further supports the connection between the cooling and the Pinatubo eruption. The overall warming trend is largely associated with changes in CO2 (Fig. 2d), surface albedo (Fig. 2f, relatively small magnitudes), clouds (Fig. 2g), and surface dynamics (Fig. 2h, mainly reflecting changes in the ocean heat content). Among these four processes, only CO2 maintains positive contributions to the warming trend throughout the period while the other three switch signs of contributions (from positive to negative) in the middle to late 1990s. The sign switch is compensated by a sign switch of opposite direction (from negative to positive) in atmospheric dynamics (Fig. 2i) such that a positive temperature trend is maintained in the early 2000s. This, however, does not change the general role of atmospheric dynamics that tends to “cool” the surface through turbulent and convective heat transport as evidenced by the negative contributions of this term to the warming trend prior to the late 1990s. The changes in ozone and solar forcing, largely associated with the quasi-11-yr solar cycle, have relatively weak impacts on the surface temperature evolutions during this period (notice the small magnitudes in Figs. 2b,c). An interesting observation based on the partial temperature evolutions shown in Fig. 2 is that the contribution of an individual physical or dynamical process to the global mean temperature change could change signs, suggesting that the nature of a radiative feedback could be time-dependent (e.g., Andrews et al. 2015; Andry et al. 2017; Armour et al. 2013; Colman and Power 2010; Klocke et al. 2013; Rugenstein et al. 2016; Senior and Mitchell 2000). For example, water vapor change contributes to warming (positive partial temperature differences) in the late 1980s (with respect to 1981–85) but cooling (negative partial temperature differences) in the 1990s and 2000s (Fig. 2e).

Fig. 2.

Temporal evolutions of the observed global surface temperature (K) from 1984 to 2003 shown as the differences between the averages over individual 5-yr segments and the average over the base state (1981–85): (a) the sum of partial temperature differences (K) derived from the CFRAM, and partial temperature differences (K) due to changes in (b) solar irradiance, (c) ozone, (d) CO2, (e) water vapor, (f) albedo, (g) clouds, (h) surface dynamics, and (i) atmospheric dynamics. The values in the upper right of the figure panels denote the mean values of the corresponding curves.

Fig. 2.

Temporal evolutions of the observed global surface temperature (K) from 1984 to 2003 shown as the differences between the averages over individual 5-yr segments and the average over the base state (1981–85): (a) the sum of partial temperature differences (K) derived from the CFRAM, and partial temperature differences (K) due to changes in (b) solar irradiance, (c) ozone, (d) CO2, (e) water vapor, (f) albedo, (g) clouds, (h) surface dynamics, and (i) atmospheric dynamics. The values in the upper right of the figure panels denote the mean values of the corresponding curves.

The counterpart results based on the CCSM4 decadal hindcasts are shown in Fig. 3. The overall warming trend and the transient cooling from 1989 to 1994 are clearly captured by the model although the temperature drop during the cooling period is of greater magnitude and the interannual fluctuations in the surface temperature are considerably smaller compared to the observation (cf. Figs. 3a and 2a). The temperature drop in the model during the transient cooling period is largely driven by changes in water vapor and surface dynamics (Figs. 3e,h). The increase of aerosol concentrations during this period clearly cools the surface as indicated by negative partial temperature differences associated with atmospheric dynamics (Fig. 3i). However, the net contribution of atmospheric dynamics to the cooling trend is negative since the amplitude of the negative partial temperature difference decreases from 1989 to 1994. Similar to the observation, CO2, surface albedo and clouds are the major contributors to the overall warming trend. Surface dynamics (mainly ocean heat content change) plays a key role in the warming in the observation, but exhibits quite different behavior in the model (cf. Figs. 3h and 2h), suggesting potential problems in representing long-term changes of ocean processes in the model. Another striking difference between the observation and the model is that in the observation water vapor effect changes signs (Fig. 2e) while in the model water vapor effect maintains positive values (i.e., net warming contributions) of much greater amplitudes throughout the analysis period (Fig. 3e). The surface temperature evolution in the hindcast is largely determined by the effect of water vapor change (cf. Figs. 3a and 3e). This suggests that the increase of atmospheric water vapor content from the base state (1981–85) to each of the 5-yr segments is much greater in the hindcast compared to that in the observation. Furthermore, the interannual variation in atmospheric water vapor content is significantly underestimated in the model. This underestimation is confirmed by examining the atmospheric water vapor field in other modern reanalysis data [e.g., the NASA Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2)] and satellite observations [e.g., the Remote Sensing Systems (RSS) Merged Total Precipitable Water version 7]. The temporal evolutions of atmospheric moisture content in both the ERA-Interim and MERRA-2 share similar features with the RSS observation while the CCSM4 hindcast shows a significant systematic bias as well as a much stronger positive trend since 1988 (figures not shown). Several earlier studies also demonstrated that the relationship between temperature and water vapor is weaker in radiosonde observations and reanalysis data than in atmospheric models, indicating that the water vapor assimilation is working against a moist bias in the model atmosphere and the increasing surface evaporation caused by surface warming (Sun and Held 1996; Mears and Wentz 2009; Bosilovich et al. 2017). The cloud effects in both the model and the observation are positive (i.e., net warming contributions with respect to the base state), although the magnitude of cloud effect in the model is about 4 times smaller compared to that in the observation. This difference, however, does not directly point to any problems in model representations of clouds since cloud fields in the ERA-Interim are also subject to substantial uncertainties in model parameterizations. Furthermore, cloud effects tend to be overestimated when time-mean cloud properties are used in the offline radiative transfer calculations. This implies that differences of cloud forcing between the CCSM4 hindcast and the ERA-Interim might have been slightly amplified by the CFRAM analysis. As in the observation, the effects of solar forcing and ozone on the surface temperature evolution are much weaker compared to other processes (Figs. 3b,c).

Fig. 3.

As in Fig. 2, but for the CCSM4 decadal hindcast results.

Fig. 3.

As in Fig. 2, but for the CCSM4 decadal hindcast results.

The global surface temperature differences averaged over all the 5-yr segments except for the 1981–85 are shown in Fig. 4a (observation) and Fig. 5a (model) as departures from the corresponding base state (1981–85) temperature. The most pronounced feature is the spatially nonuniform warming with land experiencing stronger temperature increase compared to the ocean, especially over the Greenland, North America, East Asia, North Africa, and the Middle East. As for the oceans, significant warming occurred over the western Pacific, the northern Indian Ocean, and the North Atlantic while cooling is found near the west coasts of South America and in the Southern Ocean. The CCSM4 hindcast (Fig. 5a) captures to a lesser degree the warming contrast between land and ocean. However, the warming pattern in the model hindcast is more spatially homogenous compared to the observation and characterized by regional temperature changes of smaller amplitudes. The spatial distributions of the partial temperature differences associated with individual radiative and nonradiative processes are given in Figs. 4b–i (observation) and Figs. 5b–i (model). The rise of atmospheric CO2 concentration provides nearly uniform warming over the globe in both the observation and the model hindcast with a globally averaged partial temperature difference of approximately 0.08 K (upper right corners of Figs. 4d and 5d). Solar forcing has a net warming effect during this period but it is nearly two magnitudes smaller compared to the CO2 (Figs. 4b and 5b). The effect of ozone is also weak and varies spatially, but Figs. 4c and 5c are not directly comparable since different ozone profiles have been used in the production of the ERA-Interim and the CCSM4 hindcast. Although the surface albedo effect on a global average is weakly negative in the model (−0.01 K) and positive in the observation (0.05 K), the model does capture the large positive contributions (warming effect) of albedo changes in the polar regions (Figs. 4f and 5f).

Fig. 4.

Observed global surface temperature differences (K) (with respect to the 1981–85 base state) averaged over all the 5-yr segments except for the 1981–85 one: (a) the sum of partial temperature differences (K) derived from the CFRAM, and partial temperature differences (K) due to changes in (b) solar irradiance, (c) ozone, (d) CO2, (e) water vapor, (f) albedo, (g) clouds, (h) surface dynamics, and (i) atmospheric dynamics. The values in the upper right of the figure panels denote the global mean values and the curves on the right of the panels denote the meridional profiles of the zonally averaged values.

Fig. 4.

Observed global surface temperature differences (K) (with respect to the 1981–85 base state) averaged over all the 5-yr segments except for the 1981–85 one: (a) the sum of partial temperature differences (K) derived from the CFRAM, and partial temperature differences (K) due to changes in (b) solar irradiance, (c) ozone, (d) CO2, (e) water vapor, (f) albedo, (g) clouds, (h) surface dynamics, and (i) atmospheric dynamics. The values in the upper right of the figure panels denote the global mean values and the curves on the right of the panels denote the meridional profiles of the zonally averaged values.

Fig. 5.

As in Fig. 4, but for the CCSM4 decadal hindcast results.

Fig. 5.

As in Fig. 4, but for the CCSM4 decadal hindcast results.

The main discrepancies between the model hindcast and the observation are in the effects of water vapor, surface dynamics, and atmospheric dynamics. As discussed previously, the model carries an overly strong water vapor effect, evidenced by the uniformly warming contributions (Fig. 5e; 0.18 K) while in the observation the water vapor effect exhibits substantial spatial variations (e.g., cooling contributions to the equatorial central and eastern Pacific) (Fig. 4e; −0.05 K). The surface dynamics (mainly ocean circulation and heat continent change) is a major contributor to surface warming in the observation with a globally averaged value of 0.83 K (Fig. 4h). This process is characterized by positive partial temperature differences over a large portion of global oceans, especially in the equatorial Indian and Pacific oceans and in the Southern Hemisphere (Fig. 4h; see the meridional profile of the zonally averaged partial temperature differences to the right). In the CCSM4 hindcast, the warming effect of surface dynamics is clearly seen in the North Atlantic but less pronounced and weaker in other ocean basins, leading to a globally averaged value of −0.01 K (Fig. 5h). The meridional profile of this effect in the model indicates warming contributions in the southern and northern midlatitudes—a feature distinctly different from that in the observation. Another notable difference between the observation and the CCSM4 hindcast is that surface dynamics shows relatively larger values over land in the former. It has been emphasized before that the surface dynamics term here includes inherent errors associated with the use of offline radiative transfer calculations and linearization of radiative energy perturbations. However, the large values over land turn out to be mainly related to the unbalanced surface energy budget in the ERA-Interim and the fact that surface dynamics is estimated as a residual in the equation of surface energy balance. Repeating the CFRAM analysis with a reanalysis product that closes its surface energy budget (e.g., the NASA MERRA-2) confirms that this issue in the surface flux data of the ERA-Interim does not affect qualitatively the relative importance of the surface dynamics term in contributing to the overall interannual-to-decadal variation of the global mean surface temperature (figures not shown). In the observation, atmospheric dynamics act as the main process to cool the surface at locations where surface dynamics exhibits warming effects (see the out-of-phase relationship between the two meridional profiles to the right of Figs. 4i and 4h), thus the global averaged value of atmospheric dynamics is about −0.94 K (Fig. 4i). Because of the differences in surface dynamics between the model and the observation, the globally averaged value of the partial temperature differences associated with atmospheric dynamics is −0.11 K (Fig. 5i). The magnitude of the cloud-associated partial temperature differences is smaller than that in the observation (0.05 vs 0.22 K; cf. Figs. 5g and 4g), in agreement with the earlier findings. The spatial distribution of cloud effect is also significantly different between the observation and the model, exemplified by the difference over the equatorial Pacific. The implications of these cloud-related differences are unclear, however, because of uncertainties in the ERA-Interim cloud fields.

c. Assessment of the overall contributions of individual processes to the decadal-scale temperature evolutions

To further quantify the overall contribution of an individual process to the temporal evolution of the surface temperature around the globe, we adopt the measure of the “pattern-amplitude projection” (PAP) coefficient used in Park et al. (2012) and Deng et al. (2013). The PAP coefficient is defined as

 
formula

where and are the observed/modeled total surface temperature differences and partial surface temperature differences related to the ith process, respectively, and the angle brackets denote the temporal average (for temporal PAP) or area average (for spatial PAP). The sum of all of converges to the temporal average or area average of the observed/modeled total surface temperature differences for the temporal PAP or spatial PAP, respectively.

The temporal PAP coefficients for all processes being considered are obtained first utilizing the partial temperature results (corresponding to individual 5-yr segments) reported in Fig. 2 (observation) and Fig. 3 (model). These temporal PAPs emphasize the contribution of a process to the interannual–decadal fluctuations in the regionally averaged surface temperature, not just the trend. The results are shown separately for “globe,” “ocean,” and “land” in the left panels in Figs. 6 and 7. Note that the sums of the PAP coefficients in Figs. 6a and 7a equal the global mean temperature difference reported in Figs. 4a and 5a, respectively. In a global sense, changes in surface dynamics (mainly ocean heat content change once all ocean basins are averaged) is the main positive contributor to the surface temperature evolution in the observation with clouds, CO2, and surface albedo change providing secondary contributions at a decreasing order of magnitude (Fig. 6a). The result over ocean is similar to that over the globe as it should be (Fig. 6c). Over land, surface dynamics and clouds remain the two main positive contributors but the relative contributions of CO2 and water vapor both increase (Fig. 6e). Atmospheric dynamics provides a net negative contribution to the surface temperature evolution since it tends to cool (warm) the surface when it gets warmer (colder), consistent with the earlier discussions. In contrast to the observation, the CCSM4 hindcast indicates a positive contribution of surface dynamics that is significantly weaker, even becoming negative over ocean (Figs. 7a,c). The reduced amplitude of the surface dynamics contribution is compensated by overly strong positive contributions from water vapor (Figs. 7a,c,e). The net effects of CO2, clouds, and albedo on the global surface temperature evolution are all consistent with the observations despite the differences in the relative significance.

Fig. 6.

(left) Temporal pattern-amplitude projection (PAP) coefficients (K) associated with the eight partial temperature differences shown in Figs. 2b–i and their sums over the (a) globe, (c) ocean, and (e) land. (right) Temporal evolution of the spatial PAP coefficients (K; refer to the left y coordinate) associated with the same eight processes and their sums (solid black curves; refer to the right y coordinate) over the (b) globe, (d) ocean, and (f) land. The labels SR, O3, CO2, WV, AL, CLD, SUR, ATM, and SUM stand for solar irradiance, ozone, CO2, water vapor, surface albedo, cloud, surface dynamics, atmospheric dynamics, and the sum, respectively. The results are from the ERA-Interim reanalysis.

Fig. 6.

(left) Temporal pattern-amplitude projection (PAP) coefficients (K) associated with the eight partial temperature differences shown in Figs. 2b–i and their sums over the (a) globe, (c) ocean, and (e) land. (right) Temporal evolution of the spatial PAP coefficients (K; refer to the left y coordinate) associated with the same eight processes and their sums (solid black curves; refer to the right y coordinate) over the (b) globe, (d) ocean, and (f) land. The labels SR, O3, CO2, WV, AL, CLD, SUR, ATM, and SUM stand for solar irradiance, ozone, CO2, water vapor, surface albedo, cloud, surface dynamics, atmospheric dynamics, and the sum, respectively. The results are from the ERA-Interim reanalysis.

Fig. 7.

As in Fig. 6, but for the CCSM4 decadal hindcast results.

Fig. 7.

As in Fig. 6, but for the CCSM4 decadal hindcast results.

In the last step, we examine the temporal evolutions of the spatial PAP coefficients computed for an individual 5-yr segment. For each segment, the sum of the PAP coefficients equals the total surface temperature differences (with respect to the 1981–85 base state) averaged over the globe, ocean or land (black curves in the right panels of Figs. 6 and 7). Note that the spatial PAP here emphasizes the contribution of an individual process to the spatial distribution of the total temperature change through pattern projection. This gives PAP coefficients slightly different physical meanings compared to the original partial temperature differences they are based off, even though sum of the spatial PAPs over all the processes gives the same total temperature difference as the sum of the partial temperature differences over all the processes.

In the observation, surface dynamics clearly dictates the temperature difference between each 5-yr segment and the base state (1981–85) over the globe, ocean, or land (Figs. 6b,d,f). The overall temperature evolution also receives secondary positive contributions from processes related to clouds, albedo, and CO2 changes although their magnitudes are much smaller compared to that related to surface dynamics. As expected, atmospheric dynamics tends to work against surface temperature fluctuations, evidenced by negative values throughout the period. In the CCSM4 hindcast, surface and atmospheric dynamics act respectively as the main positive and negative contributor to the temperature evolution in a way similar to the observation (Figs. 7b,d,f). The main difference between the model and observation is that water vapor effect replaces cloud effect to become the second important positive contributor. In addition, the importance of both CO2 and albedo relative to surface dynamics is amplified in the model.

4. Concluding remarks

In this study, we make an initial effort to partition the temporal evolution of the global surface temperature into individual radiative and (nonradiative) dynamical processes in the context of the CFRAM method. This partitioning is done for the NCAR CCSM4 decadal hindcasts and the results are contrasted with those based on the ERA-Interim. Getting back to the questions raised toward the end of the introduction, CCSM4 is able to predict an overall warming trend through the period 1981–2005 as well as the transient cooling occurring during 1989 to 1994. While the rising CO2 concentration provides consistent positive contributions to the warming trends in both the observation and the model, the two differ substantially in the relative importance of other individual processes in contributing to the total temperature evolution. Most noticeable is the difference in the most critical process dictating the surface temperature evolution: it is the surface dynamics (i.e., ocean circulation and heat content change) in the observation but water vapor in the model. What makes this contrast even more interesting is that surface dynamics clearly determines the spatial pattern of the surface temperature change in the model just like that in the observation (see discussion of the spatial PAPs), but the net effect (i.e., when globally averaging the partial temperature differences) is much smaller compared to an overly strong water vapor effect characterized by nearly all positive partial temperature differences around the globe. The root of this difference seems to lie in the fact that CCSM4 overpredicts the increase of atmospheric water vapor during the study period; in other words, the water vapor feedback seems to be too strong in the model. This strong positive trend and suppressed interannual variations of water vapor in the CCSM4 are further verified by comparing with the NASA MERRA-2 reanalysis and satellite total precipitable water products, suggesting potential problems in modeling moisture sources and sinks (i.e., evapotranspiration and precipitation) in the CCSM4.

Impacts of solar forcing and ozone change are relatively weak in both the observation and the CCSM4 hindcasts and the model representation of surface albedo and atmospheric dynamics are relatively reasonable despite differences in the relative magnitudes. The cloud effect appears weaker compared to that in the observation and the spatial distributions of cloud-related partial temperature differences also differ substantially between the model and the observation. However, no clear conclusions can be drawn regarding the cloud effects due to the fact that cloud fields in the ERA-Interim are also subject to significant model uncertainties.

A final cautionary note on the results reported here is that the temperature partitioning carried out in this study is based upon the principle of energy balance in an atmosphere–surface column and the partial temperature difference associated with a particular process (e.g., clouds) obtained through CFRAM are effectively those (atmospheric/surface) temperature differences that, when added to the temperature of the base state, will produce a longwave radiative heating difference that balances the radiative energy perturbations induced by the changes in clouds. In this sense, CFRAM does not suggest “causality” by definition. In other words, the partial temperature differences do not indicate what processes initiate the change and what processes follow. It represents only an efficient offline diagnostic tool to produce addable, process-based partial surface temperature differences that can be utilized to provide a quick scan of what processes appear to be different between the model and the reality. Additional model experiments where individual processes are manipulated are necessary to complete the attribution of any existing model biases. Finally, other observational datasets and reanalysis products (e.g., the NOAA CFSR) and the new decadal hindcasts from phase 6 of the Coupled Model Intercomparison Project (CMIP6; Eyring et al. 2016) will be used in the near future to further verify the conclusions drawn here, and in particular to further constrain the cloud effects and understand the origins of the observation–hindcast differences in the magnitude of the water vapor effect.

Acknowledgments

This study is supported by the National Science Foundation under Grants AGS-1147601, AGS-1354402, and AGS-1445956.

REFERENCES

REFERENCES
Andrews
,
T.
,
J. M.
Gregory
, and
M. J.
Webb
,
2015
:
The dependence of radiative forcing and feedback on evolving patterns of surface temperature change in climate models
.
J. Climate
,
28
,
1630
1648
, doi:.
Andry
,
O.
,
R.
Bintanja
, and
W.
Hazeleger
,
2017
:
Time-dependent variations in the Arctic’s surface albedo feedback and the link to seasonality in sea ice
.
J. Climate
,
30
,
393
410
, doi:.
Armour
,
K. C.
,
C. M.
Bitz
, and
G. H.
Roe
,
2013
:
Time-varying climate sensitivity from regional feedbacks
.
J. Climate
,
26
,
4518
4534
, doi:.
Bellucci
,
A.
, and Coauthors
,
2013
:
Decadal climate predictions with a coupled OAGCM initialized with oceanic reanalyses
.
Climate Dyn.
,
40
,
1483
1497
, doi:.
Bellucci
,
A.
, and Coauthors
,
2015a
:
Advancements in decadal climate predictability: The role of nonoceanic drivers
.
Rev. Geophys.
,
53
,
165
202
, doi:.
Bellucci
,
A.
, and Coauthors
,
2015b
:
An assessment of a multi-model ensemble of decadal climate predictions
.
Climate Dyn.
,
44
,
2787
2806
, doi:.
Bombardi
,
R. J.
, and Coauthors
,
2015
:
Evaluation of the CFSv2 CMIP5 decadal predictions
.
Climate Dyn.
,
44
,
543
557
, doi:.
Bosilovich
,
M.
,
F.
Robertson
,
L.
Takacs
,
A.
Molod
, and
D.
Mocko
,
2017
:
Atmospheric water balance and variability in the MERRA-2 reanalysis
.
J. Climate
,
30
,
1177
1196
, doi:.
Branstator
,
G.
, and
H.
Teng
,
2012
:
Potential impact of initialization on decadal predictions as assessed for CMIP5 models
.
Geophys. Res. Lett.
,
39
,
L12703
, doi:.
Cai
,
M.
, and
J.
Lu
,
2009
:
A new framework for isolating individual feedback processes in coupled general circulation climate models. Part II: Method demonstrations and comparisons
.
Climate Dyn.
,
32
,
887
900
, doi:.
Chikamoto
,
Y.
,
M.
Kimoto
,
M.
Watanabe
,
M.
Ishii
, and
T.
Mochizuki
,
2012
:
Relationship between the Pacific and Atlantic stepwise climate change during the 1990s
.
Geophys. Res. Lett.
,
39
,
L21710
, doi:.
Choi
,
J.
,
S.-W.
Son
,
Y.-G.
Ham
,
J.-Y.
Lee
, and
H.-M.
Kim
,
2016
:
Seasonal-to-interannual prediction skills of near-surface air temperature in the CMIP5 decadal hindcast experiments
.
J. Climate
,
29
,
1511
1527
, doi:.
Colman
,
R. A.
, and
S. B.
Power
,
2010
:
Atmospheric radiative feedbacks associated with transient climate change and climate variability
.
Climate Dyn.
,
34
,
919
933
, doi:.
Corti
,
S.
, and Coauthors
,
2015
:
Impact of initial conditions versus external forcing in decadal climate predictions: A sensitivity experiment
.
J. Climate
,
28
,
4454
4470
, doi:.
Dee
,
D. P.
, and Coauthors
,
2011
:
The ERA-Interim reanalysis: Configuration and performance of the data assimilation system
.
Quart. J. Roy. Meteor. Soc.
,
137
,
553
597
, doi:.
Deng
,
Y.
,
T.-W.
Park
, and
M.
Cai
,
2012
:
Process-based decomposition of the global surface temperature response to El Niño in boreal winter
.
J. Atmos. Sci.
,
69
,
1706
1712
, doi:.
Deng
,
Y.
,
T.-W.
Park
, and
M.
Cai
,
2013
:
Radiative and dynamical forcing of the surface and atmospheric temperature anomalies associated with the Northern Annular Mode
.
J. Climate
,
26
,
5124
5138
, doi:.
Doblas-Reyes
,
F. J.
, and Coauthors
,
2013
:
Initialized near-term regional climate change prediction
.
Nat. Commun.
,
4
,
1715
, doi:.
ECMWF
,
2009
: ERA-Interim project. National Center for Atmospheric Research Computational and Information Systems Laboratory Research Data Archive, accessed 30 March 2014, doi:.
Eyring
,
V.
,
S.
Bony
,
G. A.
Meehl
,
C. A.
Senior
,
B.
Stevens
,
R. J.
Stouffer
, and
K. E.
Taylor
,
2016
:
Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization
.
Geosci. Model Dev.
,
9
,
1937
1958
, doi:.
Fu
,
Q.
, and
K. N.
Liou
,
1992
:
On the correlated k-distribution method for radiative transfer in nonhomogeneous atmospheres
.
J. Atmos. Sci.
,
49
,
2139
2156
, doi:.
Fu
,
Q.
, and
K. N.
Liou
,
1993
:
Parameterization of the radiative properties of cirrus clouds
.
J. Atmos. Sci.
,
50
,
2008
2025
, doi:.
Fyfe
,
J. C.
,
W. J.
Merryfield
,
V.
Kharin
,
G. J.
Boer
,
W. S.
Lee
, and
K.
von Salzen
,
2011
:
Skillful predictions of decadal trends in global mean surface temperature
.
Geophys. Res. Lett.
,
38
,
L22801
, doi:.
García-Serrano
,
J.
, and
F. J.
Doblas-Reyes
,
2012
:
On the assessment of near-surface global temperature and North Atlantic multi-decadal variability in the ENSEMBLES decadal hindcast
.
Climate Dyn.
,
39
,
2025
2040
, doi:.
García-Serrano
,
J.
,
V.
Guemas
, and
F. J.
Doblas-Reyes
,
2015
:
Added-value from initialization in predictions of Atlantic multi-decadal variability
.
Climate Dyn.
,
44
,
2539
2555
, doi:.
Gent
,
P. R.
, and Coauthors
,
2011
:
The Community Climate System Model version 4
.
J. Climate
,
24
,
4973
4991
, doi:.
Guemas
,
V.
,
F. J.
Doblas-Reyes
,
F.
Lienert
,
Y.
Soufflet
, and
H.
Du
,
2012
:
Identifying the causes of the poor decadal climate prediction skill over the North Pacific
.
J. Geophys. Res.
,
117
,
D20111
, doi:.
Guemas
,
V.
,
S.
Corti
,
J.
García-Serrano
,
F. J.
Doblas-Reyes
,
M.
Balmaseda
, and
L.
Magnusson
,
2013
:
The Indian Ocean: The region of highest skill worldwide in decadal climate prediction
.
J. Climate
,
26
,
726
739
, doi:.
Hazeleger
,
W.
,
V.
Guemas
,
B.
Wouters
,
S.
Corti
,
I.
Andreu-Burillo
,
F. J.
Doblas-Reyes
,
K.
Wyser
, and
M.
Caian
,
2013a
:
Multiyear climate predictions using two initialization strategies
.
Geophys. Res. Lett.
,
40
,
1794
1798
, doi:.
Hazeleger
,
W.
, and Coauthors
,
2013b
:
Predicting multiyear North Atlantic Ocean variability
.
J. Geophys. Res. Oceans
,
118
,
1087
1098
, doi:.
Hermanson
,
L.
,
R.
Eade
,
N. H.
Robinson
,
N. J.
Dunstone
,
M. B.
Andrews
,
J. R.
Knight
,
A. A.
Scaife
, and
D. M.
Smith
,
2014
:
Forecast cooling of the Atlantic subpolar gyre and associated impacts
.
Geophys. Res. Lett.
,
41
,
5167
5174
, doi:.
Hu
,
X.
,
S.
Yang
, and
M.
Cai
,
2016
:
Contrasting the eastern Pacific El Niño and the central Pacific El Niño: Process-based feedback attribution
.
Climate Dyn.
,
47
,
2413
2424
, doi:.
Hu
,
X.
,
Y.
Li
,
S.
Yang
,
Y.
Deng
, and
M.
Cai
,
2017
:
Process-based decomposition of the decadal climate difference between 2002–13 and 1984–95
.
J. Climate
,
30
,
4373
4393
, doi:.
Huang
,
B.
, and Coauthors
,
2015
:
Climate drift of AMOC, North Atlantic salinity and Arctic sea ice in CFSv2 decadal predictions
.
Climate Dyn.
,
44
,
559
583
, doi:.
Karspeck
,
A.
,
S.
Yeager
,
G.
Danabasoglu
, and
H.
Teng
,
2015
:
An evaluation of experimental decadal predictions using CCSM4
.
Climate Dyn.
,
44
,
907
923
, doi:.
Kim
,
H.-M.
,
P. J.
Webster
, and
J. A.
Curry
,
2012
:
Evaluation of short-term climate change prediction in multi-model CMIP5 decadal hindcasts
.
Geophys. Res. Lett.
,
39
,
L10701
, doi:.
Klocke
,
D.
,
J.
Quaas
, and
B.
Stevens
,
2013
:
Assessment of different metrics for physical climate feedbacks
.
Climate Dyn.
,
41
,
1173
1185
, doi:.
Lu
,
J.
, and
M.
Cai
,
2009
:
A new framework for isolating individual feedback processes in coupled general circulation climate models. Part I: Formulation
.
Climate Dyn.
,
32
,
873
885
, doi:.
Mears
,
C. A.
, and
F. J.
Wentz
,
2009
:
Construction of the RSS V3.2 lower tropospheric dataset from the MSU and AMSU microwave sounders
.
J. Atmos. Oceanic Technol.
,
26
,
1493
1509
, doi:.
Meehl
,
G. A.
, and
H.
Teng
,
2012
:
Case studies for initialized decadal hindcasts and predictions for the Pacific region
.
Geophys. Res. Lett.
,
39
,
L22705
, doi:.
Meehl
,
G. A.
, and
H.
Teng
,
2014a
:
CMIP5 multi-model hindcasts for the mid-1970s shift and early 2000s hiatus and predictions for 2016–2035
.
Geophys. Res. Lett.
,
41
,
1711
1716
, doi:.
Meehl
,
G. A.
, and
H.
Teng
,
2014b
:
Regional precipitation simulations for the mid-1970s shift and early-2000s hiatus
.
Geophys. Res. Lett.
,
41
,
7658
7665
, doi:.
Meehl
,
G. A.
, and Coauthors
,
2009
:
Decadal prediction: Can it be skillful?
Bull. Amer. Meteor. Soc.
,
90
,
1467
1485
, doi:.
Meehl
,
G. A.
,
H.
Teng
, and
J. M.
Arblaster
,
2014a
:
Climate model simulations of the observed early-2000s hiatus of global warming
.
Nat. Climate Change
,
4
,
898
902
, doi:.
Meehl
,
G. A.
, and Coauthors
,
2014b
:
Decadal climate prediction: An update from the trenches
.
Bull. Amer. Meteor. Soc.
,
95
,
243
267
, doi:.
Meehl
,
G. A.
,
A.
Hu
, and
H.
Teng
,
2016
:
Initialized decadal prediction for transition to positive phase of the Interdecadal Pacific Oscillation
.
Nat. Commun.
,
7
,
11718
, doi:.
Msadek
,
R.
, and Coauthors
,
2014
:
Predicting a decadal shift in North Atlantic climate variability using the GFDL forecast system
.
J. Climate
,
27
,
6472
6496
, doi:.
Park
,
T.-W.
,
Y.
Deng
, and
M.
Cai
,
2012
:
Feedback attribution of the El Niño–Southern Oscillation–related atmospheric and surface temperature anomalies
.
J. Geophys. Res.
,
117
,
D23101
, doi:.
Pohlmann
,
H.
,
D. M.
Smith
,
M. A.
Balmaseda
,
N. S.
Keenlyside
,
S.
Masina
,
D.
Matei
,
W. A.
Müller
, and
P.
Rogel
,
2013
:
Predictability of the mid-latitude Atlantic meridional overturning circulation in a multi-model system
.
Climate Dyn.
,
41
,
775
785
, doi:.
Rugenstein
,
M. A. A.
,
K.
Caldeira
, and
R.
Knutti
,
2016
:
Dependence of global radiative feedbacks on evolving patterns of surface heat fluxes
.
Geophys. Res. Lett.
,
43
,
9877
9885
, doi:.
Sejas
,
S. A.
,
M.
Cai
,
A.-X.
Hu
,
G. A.
Meehl
,
W.
Washington
, and
P. C.
Taylor
,
2014
:
Individual feedback contributions to the seasonality of surface warming
.
J. Climate
,
27
,
5653
5669
, doi:.
Senior
,
C. A.
, and
J. F. B.
Mitchell
,
2000
:
The time-dependence of climate sensitivity
.
Geophys. Res. Lett.
,
27
,
2685
2688
, doi:.
Smith
,
D. M.
,
R.
Eade
, and
H.
Pohlmann
,
2013
:
A comparison of full-field and anomaly initialization for seasonal to decadal climate prediction
.
Climate Dyn.
,
41
,
3325
3338
, doi:.
Song
,
X.
,
G. J.
Zhang
, and
M.
Cai
,
2014a
:
Quantifying contributions of climate feedbacks to tropospheric warming in the NCAR CCSM3.0
.
Climate Dyn.
,
42
,
901
917
, doi:.
Song
,
X.
,
G. J.
Zhang
, and
M.
Cai
,
2014b
:
Characterizing the climate feedback pattern in the NCAR CCSM3-SOM using hourly data
.
J. Climate
,
27
,
2912
2930
, doi:.
Sun
,
D.-Z.
, and
I. M.
Held
,
1996
:
A comparison of modeled and observed relationships between interannual variations of water vapor and temperature
.
J. Climate
,
9
,
665
675
, doi:.
Swingedouw
,
D.
,
J.
Mignot
,
S.
Labetoulle
,
E.
Guilyardi
, and
G.
Madec
,
2013
:
Initialisation and predictability of the AMOC over the last 50 years in a climate model
.
Climate Dyn.
,
40
,
2381
2399
, doi:.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
, doi:.
Thoma
,
M.
,
R. J.
Greatbatch
,
C.
Kadow
, and
R.
Gerdes
,
2015
:
Decadal hindcasts initialized using observed surface wind stress: Evaluation and prediction out to 2024
.
Geophys. Res. Lett.
,
42
,
6454
6461
, doi:.
Yeager
,
S.
,
A.
Karspeck
,
G.
Danabasoglu
,
J.
Tribbia
, and
H.
Teng
,
2012
:
A decadal prediction case study: Late twentieth-century North Atlantic ocean heat content
.
J. Climate
,
25
,
5173
5189
, doi:.
Yeager
,
S.
,
A.
Karspeck
, and
G.
Danabasoglu
,
2015
:
Predicted slowdown in the rate of Atlantic sea ice loss
.
Geophys. Res. Lett.
,
42
,
10 704
10 713
, doi:.

Footnotes

This article is included in the CCSM4 Special Collection.

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).