This study investigates the annual cycle of radiative contributions to global climate feedbacks. A partial radiative perturbation (PRP) technique is used to diagnose monthly radiative perturbations at the top of atmosphere (TOA) due to CO2 forcing; surface temperature response; and water vapor, cloud, lapse rate, and surface albedo feedbacks using NCAR Community Climate System Model, version 3 (CCSM3) output from a Special Report on Emissions Scenarios (SRES) A1B emissions-scenario-forced climate simulation. The seasonal global mean longwave TOA radiative feedback was found to be minimal. However, the global mean shortwave (SW) TOA cloud and surface albedo radiative perturbations exhibit large seasonality. The largest contributions to the negative SW cloud feedback occur during summer in each hemisphere, marking the largest differences with previous results. Results suggest that intermodel spread in climate sensitivity may occur, partially from cloud and surface albedo feedback seasonality differences. Further, links between the climate feedback and surface temperature response seasonality are investigated, showing a strong relationship between the seasonal climate feedback distribution and the seasonal surface temperature response.
A large intermodel spread in climate sensitivity exists among state-of-the-art general circulation models (GCMs) (Solomon et al. 2007). A number of model intercomparison studies have shown this spread to be a result of different model feedback strengths (Colman 2003a; Soden and Held 2006; Soden et al. 2008). Recent studies (Gregory and Webb 2008; Andrews and Forster 2008) have also suggested that a portion of the intermodel spread is due to “fast feedbacks,” which are changes in atmospheric variables that respond directly to the forcing and not global mean surface temperature change. Thus, detailed diagnostic climate model forcing and feedback analysis remains a necessary activity to increase understanding of climate sensitivity.
Feedback strengths are typically characterized in terms of top-of-atmosphere (TOA) global, annual mean radiative perturbations normalized by the global mean surface temperature response assuming a quasi equilibrium (e.g., Wetherald and Manabe 1988). However, as shown by Colman (2002), Boer and Yu (2003), Winton (2006), Soden et al. (2008), and Taylor et al. (2011), geographical distributions of feedback strengths have a rich spatially varying structure. Further investigation of the feedback geographical structure was performed by Taylor et al. (2011) and Colman (2002), who investigated possible links between feedbacks and model physical processes, for example, between convection precipitation response and cloud feedbacks. Such analyses suggest strong relationships between spatial feedback variations and the model convective and boundary layer height responses; however, causality has not been shown.
Colman (2003b) first took a detailed look at the annual cycle of climate feedbacks using the Bureau of Meteorology Research Centre (BMRC) GCM, finding large seasonality in global mean surface albedo and cloud feedbacks and smaller seasonality of global mean water vapor and lapse rate feedbacks. Considering that climate response is dependent upon model processes, the seasonality of feedback strengths, as with the geographic distributions, indicates the local importance of seasonal-scale physical processes. Fundamentally, these local model responses drive individual global mean climate feedbacks and the climate response. Moreover, the seasonal structure of climate feedbacks likely explains a portion of the spread in climate sensitivity and may be used to diagnose model differences. A detailed model intercomparison of feedback seasonality would likely provide an indication of the culpable model physical processes.
Bony et al. (2006) presented a path toward understanding the intermodel spread in climate sensitivity that involves a better understanding of physical mechanisms driving climate feedbacks. It is argued here that the analysis of monthly contributions to climate feedbacks is a necessary component of a physical mechanism approach that allows for the investigation of potential feedback–physical process relationships associated with interseasonal and intermonth variability—for example, midlatitude cyclones, seasonal variations in the intertropical convergence zone (ITCZ), and the Madden–Julian oscillation. Along these lines, Lu and Cai (2009b) investigated the seasonality of polar warming under a 2 × CO2 forcing. They quantified the partial temperature contributions from individual feedbacks to the asymmetric polar warming response (strong winter warming, weaker summer warming). They found that clear-sky longwave (LW) downwelling radiation changes explained much of the polar warming seasonality, emphasizing the importance of atmospheric processes that increase the polar moist static energy. This type of analysis begins to link feedbacks to physical processes, illustrating the need for increased understanding of feedback seasonality to reduce climate sensitivity uncertainty.
Zhang et al. (2005) provide further indication of the importance of investigating the annual cycle. In the analysis, Zhang et al. (2005) investigated the seasonal cycle of clouds within GCMs against satellite observations, finding that models tend to simulate seasonal variations of high clouds better than low clouds. Additionally, Zhang et al. (2005) showed that models overestimating (underestimating) the frequency and optical thickness of a particular cloud regime tend to produce too strong (weak) of an annual cycle of those cloud regimes. This result links model mean state cloud deficiencies to model deficiencies in cloud seasonality. With this relationship in mind, it is likely that mean state deficiencies will carry over to the model mean state response and also influence the seasonality of the response. Thus, if feedback seasonality is important to equilibrium climate sensitivity, then the response seasonality must be correct to correctly simulate sensitivity. However, the influence of feedback seasonality on climate sensitivity has not been adequately investigated.
In this study, the seasonal contributions to climate feedbacks are investigated in the National Center for Atmospheric Research (NCAR) Community Climate System Model, version 3 (CCSM3). The outcomes of this study include relationships between the seasonal cycle and feedback radiative perturbations and suggestions about the importance of the seasonal cycle in projecting future climate change. The methodology illustrated by Taylor et al. (2011), discussed in section 2, is used to determine the monthly radiative perturbations. Section 3 contains a presentation and discussion of the monthly contributions to the individual feedbacks. Section 4 discusses the relationship between climate feedback and surface temperature response seasonality. Lastly, a summary and conclusions are presented in section 5.
a. Model characteristics and forcing
Model output from the NCAR CCSM3 forced with the Special Report on Emissions Scenarios (SRES) A1B was used to quantify climate feedback seasonality. The SRES A1B forcing is characterized by an 86% CO2 increase over year 2000 concentration by 2100, and it includes additional radiative forcing due to other greenhouse gases and sulfate aerosols. The global mean surface temperature response after 100 yr was +2.11 K. Collins et al. (2004) describes the NCAR CCSM3 components, and interested readers are referred there for a technical discussion. Collins et al. (2006) evaluated the model mean state against European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis documenting model deficiencies but did not extensively evaluate the model annual cycle.
As an evaluation of the CCSM3-simulated annual cycle, Fig. 1, the model longwave (bottom-left panel) and shortwave (SW, bottom-right panel) zonal mean cloud radiative effect (CRE) is compared with Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) (Loeb et al. 2009) zonal mean LW (top-left panel) and SW (top-right panel) CRE seasonal cycle. The LW and SW CREs are defined as the difference between all-sky and clear-sky TOA radiative fluxes. We note that the clear-sky flux definition from satellite data is different from climate models (Zhang et al. 1994). In climate models, clear-sky flux is calculated by instantaneously setting the cloud occurrence to zero, retaining the same temperature and water vapor profiles. However, CERES satellite observations only report a clear-sky flux when the field of view is determined to be 99% clear [for further details see Loeb et al. (2009)]. Thus, the CERES EBAF clear-sky flux is defined for cloud-free scenes and is likely to represent different temperature and water vapor vertical distributions than the model clear-sky flux. As a result, the model is expected to have a positive LW CRE bias and a negligible effect on the SW CRE. Both the LW and SW CREs show very similar seasonal distributions, with magnitude differences in most locations less than ±10 W m−2. The largest differences between the model and observations, larger than −30 W m−2, occur in the boreal summer over the pole in the SW CRE, indicating that model clouds are too reflective. This SW CRE bias is not likely a result of different clear-sky definitions. Despite differences in LW and SW CRE magnitudes, the CCSM3 simulates the zonal mean annual cycle reasonably well.
b. Feedback calculation
Climate feedback radiative perturbations are calculated using a partial radiative perturbation (PRP) technique (Wetherald and Manabe 1988), specifically following Taylor et al. (2011). Many of the details are repeated here for convenience, and the reader is urged to review Taylor et al. (2011) for additional details. Monthly-mean model output is commonly used to diagnose model feedback strengths and is used here (e.g., Shell et al. 2008; Wetherald and Manabe 1988). Previous studies—for example, Colman (2002) and Soden and Held (2006)—have also used daily mean or every-time-step cloud information for these calculations. Shell et al. (2008) compared the feedback strengths diagnosed using every-time step and monthly-mean output, finding differences on the order of 10%. Wetherald and Manabe (1988), however, cautioned the use of mean cloud properties in these calculations. Thus, Taylor et al. (2011) presented a method for calculating radiative perturbations and climate feedback sensitivity parameters using monthly-mean cloud information, discussed below.
Net TOA radiative flux R can be expressed as R = R(T, x, q, C, α), where T, x, q, C, and α refer to the temperature profile, CO2 concentration, atmospheric specific humidity profile, cloud property profile, and surface albedo, respectively. Considering a small radiative forcing, the global mean net TOA radiative perturbation may be written as
In (1), refers to the global mean net TOA radiative imbalance and refers to the global mean radiative perturbations from the forcing and individual feedbacks, where y represents x, T, q, C, and α. Here, can be separated into a surface temperature response and lapse rate feedback term (e.g., Hansen et al. 1984; Wetherald and Manabe 1988; Colman 2003b) as follows:
In (2), Γ represents the radiative perturbation due to a nonuniform vertical temperature response, called the lapse rate feedback. Considering equilibrium conditions, (1) is rewritten to represent the forcing-response-feedback paradigm using (2), yielding
In (3) the angled brackets represent the global, long time average, where 〈ΔR〉 = 0. Equation (3) demonstrates that the climate system net TOA radiative response due to ΔTS must balance the net TOA radiative perturbations because of the external forcing and climate feedbacks.
The global mean, long time average 〈ΔRy〉 is determined by averaging gridpoint radiative perturbations δRy calculated monthly. Here, δRy is defined as the difference between the perturbed net TOA flux and the control climate net TOA flux, where the perturbed net TOA flux is determined by substituting one variable at a time from the perturbed climate into the control climate (e.g., Wetherald and Manabe 1988; Colman 2002, 2003b). Specifically, the control climate state is defined as the first decade of the twenty-first century (years 2000–10), and the perturbed climate state is defined as the first decade of the twenty-second century (years 2100–10) of the model simulation following Soden and Held (2006). Further, the NCAR CCSM3 radiation code (Collins et al. 2004) is used in these calculations. Gridpoint, monthly-mean radiative perturbations are calculated according to
In (4)–(8), subscript 1 refers to the control climate state, subscript 2 refers to the perturbed climate state, i represents the model grid point, mon represents the particular month, j represents the year index, and N represents the 11 yr used to define the control and perturbed climates. The first and second terms on the rhs of (4)–(8) represent the perturbed and control TOA radiative fluxes, respectively.
Only monthly-mean model cloud information is archived; thus, this study follows Taylor et al. (2011) and employs a stochastic cloud generator (Raisanen et al. 2004) to sample the monthly-mean cloud vertical profile. This methodology is a variation of the Monte Carlo independent column approximation (MCICA) methodology presented by Pincus et al. (2003). The terms inside the summation operator on the rhs of (4)–(8) are calculated according to
In (9)–(14), the new index, k, refers to the kth-generated subcolumn, representing the cloud field. The perturbed and control net TOA fluxes for a given month are defined as the average calculated using M stochastically generated subcolumns; M is chosen to be 100 as a compromise between convergence and computational costs (Taylor et al. 2011). Taylor et al. (2011) showed that global, annual mean radiative perturbations and climate feedback sensitivity parameters could be accurately diagnosed using this methodology within the uncertainty range of other methodologies, for example, the radiative kernel technique (Soden and Held 2006).
Linearity, assumed in (1), is an important test for a diagnostic feedback methodology. Figure 2 is a latitude–time plot of a zonal mean net TOA radiative response calculated in two manners: 1) by simultaneously perturbing all climate variables and 2) by summing individually perturbed climate variables, representing the lhs and rhs of Eq. (1), respectively. Figure 2 (top panel) shows the simultaneous perturbation of all variables in color contours, and the sum of individual calculated radiative perturbations contours are black lines. Overall, there is good agreement between the two calculations, indicating linearity at the monthly, zonal mean scale. This result implies that 1) the contributions to TOA flux anomalies by individual climate feedbacks can be determined outside of strict quasi-equilibrium assumption and 2) the implied decorrelation assumption in Eq. (1) does not influence the qualitative feedback seasonality.
The calculated TOA radiative perturbations show general agreement with the model-simulated net TOA radiation response (Fig. 2, bottom). Large differences between diagnosed and model-simulated net TOA flux responses are found in October and November near the Southern Hemisphere pole because of the handling of polar stratospheric clouds in the calculation. This difference must be considered when analyzing seasonality at the southern pole, but it will have a negligible influence on the global mean feedback magnitude because of the small spatial area. Generally, differences between diagnosed net TOA radiative perturbations and model-simulated net TOA radiation response can be ~2–4 W m−2 in the monthly, zonal mean. Some of these differences are due to not considering additional radiative forcings, for example, sulfate aerosols, ozone, and other greenhouse gasses (GHGs). These flux differences do not influence the conclusions of this study, since we are mainly investigating relationships between the seasonal response and the seasonality of climate feedbacks. However, this result indicates the need to further refine feedback diagnostic calculations.
a. Global mean
The global mean monthly LW and SW TOA radiative perturbations behave very differently (Fig. 3). LW radiative perturbations (Fig. 3, top panel) 〈ΔRr〉, 〈ΔRx〉, 〈ΔRΓ〉, and 〈ΔRC_LW〉 show little variation, less than ±10% of mean value, throughout the annual cycle. However, 〈ΔRTOT_LW〉 exhibits more seasonality than any of the individual components, indicating either some fundamental aspects of LW feedback seasonality are common across all LW feedbacks or interaction between LW feedbacks. The previous two possibilities are not mutually exclusive. Colman (2003b) found a lack of LW radiative perturbation seasonality as well. The individual LW feedback strengths are largely invariant throughout the annual cycle and the same in both models considered. The global mean LW feedback seasonal invariance is intriguing, since there is no energetic constraint that requires this behavior.
SW radiative perturbations, however, show large seasonality (Fig. 3, top panel). A broad boreal spring and early summer maximum in 〈ΔRα〉, represented by a black dashed line with plus signs in Fig. 3, is a result of large positive contributions associated with earlier spring snowmelt and reduced Northern Hemisphere sea ice. The Colman (2003b) surface albedo results are shown with a red dashed line with plus signs (Fig. 3, bottom panel), and the boreal spring maximum is similar between the two models. However, the Colman (2003b) results show a sharp decrease in June, then a sharp increase again in July, and a gradual decrease through the remainder of the year. This intermodel difference could result from a number of different processes, and more detailed diagnostics are necessary to reconcile these model differences. A few possible explanations are model parameterizations of sea ice and snow cover albedo, different realizations of oceanic and atmospheric dynamical heat transport or snow and sea ice melt seasonality, or different cloud cover seasonality modulating the solar insolation. Despite the 〈ΔRα〉 seasonality differences, the two models simulate global, annual mean surface albedo feedback strengths that are similar, ~0.3 W m−2 K−1 (Colman 2002; Soden and Held 2006; Taylor et al. 2011), further indicating the need to understand the model feedback seasonal structure.
Differences in the SW global mean cloud radiative perturbation 〈ΔRC_SW〉 annual cycle representations are found between the models (Fig. 3, bottom panel). The CCSM3 〈ΔRC_SW〉 is negative throughout the year, characterized by two negative maxima: austral and boreal summer. This CCSM3 feedback results from increased global mean low and high cloud amounts (Fig. 4, top panel). It becomes difficult to separate the low and high cloud effects because of strong correlations between the zonal mean low and high cloud seasonal responses (Fig. 4, bottom panel). Significant positive and negative correlations, greater than 0.5 or less than −0.5 signified by the thick black vertical line in Fig. 4 (bottom panel), are found in a number of zonal bands and indicate a complicated relationship between the low and high cloud seasonal response patterns. Colman (2003b) shows a 〈ΔRC_SW〉 annual cycle with an austral summer maximum (negative) and a boreal summer minimum (positive), and speculates that the reduction in cloud amount during boreal summer is due to a cloud–soil moisture feedback. The 〈ΔRC_SW〉 annual cycle differences will be revisited when considering the meridional distribution.
b. Meridional distribution
Climate feedback seasonality δRy, where y can be T, q, C, or α, can be caused by three different mechanisms: 1) annual cycle of solar insolation, 2) annual cycle of temperature, and 3) seasonality of the model climate response. The third mechanism, model response, is only one indicative of climate feedback. The first and second mechanisms represent radiative perturbations related to known mean state climate seasonality, and these radiative perturbations should be considered separately from feedback seasonality. Control climate seasonality generally enhances the zonal mean δRy annual cycle. Colman (2003b) normalized the zonal mean δRy annual cycle using solar insolation for SW feedbacks and the TOA outgoing longwave radiation (OLR) for LW feedbacks to remove control climate seasonality influences. Here, the same normalization was performed, finding only small differences in the δRy annual cycle, resulting in little change to the interpretation. For brevity, the normalized results are not presented; however, situations where the interpretation of the results changes significantly are noted.
1) CO2 forcing
Illustrated in Fig. 3 (top panel), 〈ΔRx〉 shows little variation throughout the year. The largest zonal mean δRx annual cycle (Fig. 5, top-left panel) occurs in the tropics and subtropics, where the amplitude can be 1.0–1.5 W m−2, depending on latitude. Weak-amplitude δRx seasonality is exhibited in the mid- and high latitudes, and most of this seasonality is removed when performing OLR normalization. Local δRx minima are found off the equator in the summer hemisphere. Comparing all-sky and clear-sky δRx (not shown), it is evident that these minima are caused by control climate cloud amount seasonality, suggesting that model control climate cloud deficiencies can influence the radiative forcing seasonal distribution.
Large spatial and seasonal variability are exhibited by δRC_LW and δRC_SW (Taylor et al. 2011). Figure 3 illustrated the large-amplitude global mean 〈ΔRC_SW〉 annual cycle and the negligible 〈ΔRC_LW〉 annual cycle. Here δRC_LW (Fig. 6, top-left panel) exhibits weak-amplitude zonal mean annual cycles. The largest amplitude zonal mean δRC_LW annual cycle occurs in the equatorial region during boreal spring due to the high cloud response seasonality with an April maximum of ~+10 W m−2 and an August minimum of ~+2 W m−2. A large-amplitude δRC_LW annual cycle is present in the southern polar region; however, as mentioned above, this region seems be a result of the calculation methodology. Small maximum δRC_LW occur in the subtropics during summer, associated with increased low cloud area fraction and liquid water path, which only slightly offset the strong negative δRC_SW (Fig. 6, top-right panel).
Large negative zonal mean δRC_SW values are found during austral summer at 60°, 20°, and 5°S, accounting for the 〈ΔRC_SW〉 negative peak. Further, the 〈ΔRC_SW〉 boreal summer negative peak is accounted for by δRC_SW contributions from 5°, 30°, and poleward of 60°N. The large δRC_SW values, such as δRC_LW, result from combined increases in low and high cloud amounts (Fig. 7) and liquid water path (not shown) concurrent with maximum local solar insolation. However, normalization by the monthly-mean solar insolation weakens but does not remove the subtropical annual cycle. The Northern Hemisphere high latitudes exhibits a zonal mean δRC_SW annual cycle with −8 W m−2 values during summer; however, after normalization, this seasonality is completely removed. The zonal mean δRC_NET (Fig. 6, bottom-left panel) annual cycle is the sum of δRC_LW and δRC_SW, shown for completeness.
The zonal mean δRC_LW and δRC_SW annual cycles diagnosed for the CCSM3 illustrate very different patterns from the BMRC GCM (Colman 2003b). Since the δRC_LW and δRC_SW seasonality is associated with seasonality of cloud feedback, indicated here and by Colman (2003b), this suggests that a portion of the intermodel spread in cloud feedback may result from differences in the cloud response seasonality. The largest discrepancy in zonal mean δRC_SW seasonality between CCSM3 and BMRC GCM occurs during boreal summer between 30° and 50°N. It is this region where large negative contributions are found during boreal summer, strongly contributing to the CCSM3 negative 〈ΔRC_SW〉. Colman (2003b) shows strong positive zonal mean δRC_SW between 30° and 50°N during boreal summer, which results from a decrease in cloud fraction.
3) Water vapor
More structure is found when investigating the zonal mean δRr annual cycle (Fig. 5, top-right panel) than in 〈ΔRr〉. Weak-amplitude zonal mean δRr annual cycles are illustrated in Fig. 5, the largest amplitude variations occur in Northern Hemisphere high latitudes and along the equator. In these cases, normalization δRr does not affect the interpretation. The strongest contributions to 〈ΔRr〉 occur in April and May in the equatorial region and coincide with the column precipitable water response seasonality.
4) Lapse rate
Evident from Fig. 5 (bottom-left panel), a strong annual cycle exhibited in the Northern Hemisphere polar region, changing from δRΓ greater than +20 W m−2 for most of the winter to approximately −5 W m−2 in boreal summer. A similar behavior is apparent for the Southern Hemisphere polar region, however, with weaker amplitude. Interestingly, normalizing δRΓ by the seasonal OLR distribution leads to an enhancement of the annual cycle. A small amplitude zonal mean δRΓ annual cycle is evident in the tropics following the ITCZ convective maximum. This feature is not unexpected, as it is well known that the negative lapse rate feedback in the tropics is linked to the nature of tropical convection (e.g., Hansen et al. 1984) and is further verified by the analyses of Colman (2002) and Taylor et al. (2011).
The largest contributions to the 〈ΔRΓ〉 annual cycle occur at high latitudes in austral and boreal winter. This high-latitude lapse rate feedback behavior leads to a strong positive correlation with polar warming amplification seasonality, shown by Lu and Cai (2009b) to be related to poleward dynamical moist static energy transport seasonality. Lu and Cai (2009a) demonstrate that the lapse rate feedback is an artificial feedback resulting from TOA feedback diagnosis. They further show that the lapse rate feedback can be attributed to other radiative feedbacks, such as the water vapor and cloud feedbacks as well as nonradiative dynamical feedbacks. Considering the results of Lu and Cai (2009b), the results support the assertion that the high-latitude lapse rate feedback seasonality is related to the seasonality of the polar warming amplification and driven by the seasonality of atmospheric processes related to the dynamical transport of moist static energy.
5) Surface albedo
A large amplitude zonal mean δRα annual cycle is found in the northern and southern high latitudes (Fig. 5, bottom-right panel); however, normalization by monthly-mean solar insolation removes most of this seasonality. The largest contributions to 〈ΔRα〉 occur at high latitudes during summer months. The largest δRα contributions to the 〈ΔRα〉 annual cycle in April originate from 45°N, associated with spring snowmelt seasonality and progress northward through August from sea ice melt. Comparing the CCSM3 results with the BMRC GCM results (Colman 2003b), the model zonal mean δRα distributions are very similar; however, the local magnitudes are different. These discrepancies are likely related to different seasonal snow and sea ice melt rates. The results suggest that the magnitude of δRα seasonality contributes to intermodel surface albedo feedback differences, not differences in seasonal distribution.
Boer and Yu (2003) investigated the relationship between annual mean surface temperature response and TOA radiative perturbation spatial patterns, showing that the surface temperature response spatial pattern is explained by the spatial distribution of feedbacks. Taylor et al. (2011) extended this analysis to examine the influence of individual feedbacks on the surface temperature response spatial pattern. Here, we will discuss the links between seasonal temperature response pattern and individual climate feedbacks, which in the zonal mean have a strong positive correlation, r = +0.75. The seasonal correlation between zonal mean surface temperature response and the total climate feedback radiative perturbation ΔRfeedback—defined as the sum of all feedbacks—shows locations and times with strong and weak correlation (Fig. 8, bottom panel). The strong correlations indicate when and where climate feedback seasonality largely determines the surface temperature response, and weak and negative correlations indicate a large influence of atmosphere and ocean transport (Fig. 9, bottom panel) discussed below.
The vertically integrated column perturbation energy budget can be written as ΔH=ΔR+ΔD, where H represents energy deposition or storage in the column, R represents the TOA net radiative flux convergence, D represents the convergence of atmospheric and oceanic energy transport, and Δ indicates a difference between the perturbed and control climate states. At the annual mean scale, ΔH is negligible and ΔR and ΔD must balance (Boer and Yu 2003; Taylor et al. 2011). The locations and times of weak correlation in Fig. 8 indicate a stronger influence from internal atmospheric processes and energy transport. Using monthly model output, ΔH can be estimated using the surface temperature response (Fig. 9, top panel), and it shows very small monthly-mean energy storage outside the polar regions. Contributions from energy transport are calculated by taking the difference between ΔH (Fig. 9, top panel) and ΔR (Fig. 2, bottom panel) and are shown in Fig. 9 (bottom panel). Two major features are 1) ΔD is nearly equal and opposite of ΔR outside the polar regions, indicating an approximate balance between the terms at the monthly-mean scale; and 2) the complicated relationship near the northern pole. These results suggest the northern polar region seasonal surface temperature response is determined by the dynamical transport characteristics, since the total northern polar region TOA climate feedback radiative perturbation is seasonally invariant.
Our view is that the first-order surface temperature response seasonality is determined by climate feedback seasonality and modulated by atmospheric and oceanic energy transport, which is supported by correlation analysis results showing a stronger correlation of ΔTsfc with ΔRfeedback (r = +0.75) than with ΔD (r = +0.46). It is possible to elucidate this important relationship between monthly-mean TOA radiative flux anomalies, surface temperature response, and implied atmospheric energy transports using observations to test GCMs. Thus, this framework provides insights into climate feedbacks and atmospheric energy transport process interactions that determine the surface temperature response.
5. Summary and conclusions
The understanding of climate sensitivity, climate feedback, and the associated intermodel differences in these quantities can be further understood through detailed comparisons beyond traditional global, annual mean considerations. Here, an analysis of the monthly radiative contributions to climate feedbacks was performed using the NCAR CCSM3 SRES A1B climate simulation. The feedback radiative perturbations and climate feedback strengths were calculated following Taylor et al. (2011) using monthly-mean model output.
Examining the LW feedback radiative perturbations, very little monthly feedback strength variability was found. One potential ramification, since January and July global mean water vapor feedback strength should be about the same, is that a shorter observational time series is necessary for detection. Global mean LW monthly radiative perturbations from the NCAR CCSM3 are very similar to Colman (2003b). Alternatively, SW radiative perturbations associated with cloud and surface albedo feedbacks were found to have large seasonality. The SW cloud feedback shows two minima, one in austral and one in boreal summer. Compared with Colman (2003b), there are large differences of seasonal SW cloud radiative perturbations. It is suggested here that the cloud seasonal response is likely partly responsible for intermodel differences in the SW cloud feedback. Large differences between Colman (2003b) and the present study were also found in surface albedo feedback, despite having very similar global, annual mean feedback sensitivity parameters. Thus, intermodel differences of the representations of the surface albedo feedback may be hidden within the global, annual mean definition of climate feedback sensitivity parameters. This is instructive for future model intercomparisons, indicating that there is a need to analyze model climate feedbacks in terms of geographic and seasonal distributions, not simply in terms of the global, annual mean.
It has been previously reported that annual mean TOA radiative feedback spatial distribution explains a majority of the surface temperature response spatial variability. This study has extended the concept to the seasonal surface temperature response, suggesting that 1) the order of the TOA radiative feedback perturbations determines the seasonal surface temperature response and that 2) it is modulated by atmospheric and oceanic energy transport. Further, we speculate that intermodel differences in the degree to which TOA radiative feedback perturbations or energy transport determine the surface temperature response may indicate reasons for the climate sensitivity spread.
The Office of Biological and Environmental Research of the U.S. Department of Energy supported this research under Grant DE-FEG02-02ER63338 as part of the Atmospheric Radiation Measurement Program. We would also like to thank three anonymous reviewers for their useful comments. We acknowledge the Florida State University shared High-Performance Computing facility and staff for contributions to results presented in this paper.