Surface albedo feedbacks from climate variability and change


Corresponding author: R. Colman, Centre for Australian Weather and Climate Research, Bureau of Meteorology, GPO Box 1289, Melbourne, Vic 3001, Australia. (


[1] Snow and sea ice surface albedo feedback is evaluated in models from the World Climate Research Program Coupled Model Intercomparison Project phase 3 using the radiative “kernel” technique. A comparison is carried out between feedbacks at climate change timescales and those operating under seasonal, interannual, and decadal variability. A primary motivation is to examine possible relationships across these timescales. Similarities are found in the mean model meridional distribution of feedback, although uncertainties are large due to model spread and uncertainties from regressions. Climate change feedback appears stronger over the Arctic north of 75°N. Consistent with findings elsewhere, climate change feedback strengths over Northern Hemisphere (NH) snow and sea ice are uncorrelated, although a weak correlation is found between NH and Southern Hemisphere (SH) sea ice. By contrast, at other timescales, moderate positive correlations are found between NH snow and sea ice albedo feedbacks, as well as between NH and SH sea ice albedo feedbacks. Statistically significant correlations are not found between surface albedo feedbacks at climate change and other timescales, either for global feedback or for NH and SH sea ice. Notably, however, correlations are found for NH snow albedo feedback against both seasonal and interannual feedbacks. This suggests that NH snow albedo feedback is a “special case,” potentially revealing short-term analogues appropriate for close comparison with climate change, with such relationships not holding for other components of the surface albedo feedback.

1 Introduction

[2] Surface albedo feedback (SAF) contributes to the significant uncertainty which remains in global climate change projections [Soden and Held, 2006; Randall et al., 2007]. Although weaker than the combined water vapor/lapse rate feedback, surface albedo changes represent an important positive feedback at global scales, but one which varies by nearly an order of magnitude among models [Colman, 2003; Soden and Held, 2006]. In addition, SAF is the most spatially inhomogeneous of the principal radiative feedbacks, playing an important role in the middle- to high-latitude amplification of temperature changes [Holland and Bitz, 2003; Hall, 2004; Winton, 2006b]. Indeed, substantial reductions of Arctic sea ice [Stroeve et al., 2007] and the Northern Hemisphere (NH) continental snow cover [Lemke et al., 2007] have been already observed in recent decades, with potential major impacts on climate and ecology [Anisimov et al., 2007]. SAF is thus of particular importance both for global climate change sensitivity and for regional climate change impacts over heavily populated NH continental regions.

[3] As with other climate change feedbacks, SAF presents particular evaluation challenges due to its long (decadal to centennial) timescale, which makes direct evaluation against observations impossible. To overcome this, attempts have been made to understand and evaluate SAF associated with shorter-term variability and to understand the relationship with climate change timescale feedbacks, echoing a similar approach to water vapor, lapse rate, and cloud feedbacks [Dessler et al., 2008; Dessler and Wong, 2009; Colman and Power, 2010; Dessler, 2012; Colman and Hanson, 2012].

[4] By running a model with and without suppressed albedo feedback, Hall [2004] showed that snow-covered NH land responded similarly to temperature perturbations on interannual and climate change timescales, implying similar feedbacks. However, sea ice regions, particularly in the Southern Hemisphere (SH), responded more weakly on interannual timescales, a feature Hall [2004] attributed in part to the less geographically coherent temperature and albedo variations. A further complicating factor was the presence of an “ice thickness” feedback, whereby ice thickness changes affect the seasonal or longer-term balance of warming—for example, causing a winter maximum in NH high-latitude warming under climate change, despite there being no solar insolation in this season. Hall [2004] found that interannual feedbacks from ice thickness were considerably weaker for climate variability compared to climate change. Despite these complications, Dessler [2012] found a similar global SAF (of around 0.3 W/m2/K) for the two timescales using 10 years of reanalysis but did not elaborate on differences between snow and sea ice, or between hemispheres.

[5] Winton [2006a] found that climate change albedo feedbacks from NH land, NH sea ice, and SH sea ice are essentially uncorrelated across the Coupled Model Intercomparison Project phase 3 (CMIP3) models, despite the fact that temperature changes in these three regions are correlated. This is the case, regardless of whether radiation changes (to determine feedback) are normalized by local or global temperature changes. This implies that these regions indeed need to be considered separately when evaluating albedo feedback, and Winton noted differences in the behavior of these regions when comparing interannual and climate change timescale feedbacks.

[6] For seasonal timescale variation, an important result from Hall and Qu [2006] is that April–May seasonal albedo feedback is highly correlated with April climate change feedback for NH continental regions poleward of 30° of latitude. This implies that albedo responses to the seasonal cycle could potentially form a surrogate for climate change feedbacks, opening the possibility of a convergence of model climate change feedback results based on current climate constraints and evaluation. Sea ice changes, however, were not explored, so the implications for evaluating global scale feedbacks remain unclear.

[7] Many questions remain on the potential of evaluation of climate change timescale SAF from shorter-term variability. For example, do the promising results of Hall and Qu [2006] apply to other timescales, over sea ice, or over the SH? Can decadal variability of surface albedo provide analogues for climate change? Are there relationships between albedo feedbacks of snow and sea ice at interannual (or other) timescales? The present short note seeks to address these questions by performing a systematic investigation of timescale variation of SAF across the CMIP3 models, including seasonal, interannual, decadal and climate change. The strength and zonal distributions of these will be documented, and relationships will be explored for snow and sea ice responses. This is not an exhaustive investigation of this complex area but aims to provide a benchmark for evaluation and comparison across these timescales and to identify significant correlations. Note that the purpose of the study is not to directly evaluate the models but to explore whether models themselves suggest that evaluation of components of the climate change albedo feedbacks is possible through the establishment of relationships with feedbacks at shorter timescales. If, for example, relationships were found across timescales, then this suggests constraint may be possible, e.g., along the lines in Hall and Qu [2006], by comparison of model feedbacks with observed feedbacks at such timescales.

[8] The layout of the paper is as follows: Section 2 will discuss the experiments examined and describe the analysis technique, section 3 will present results and discussion, and section 4 will provide concluding comments.

2 Evaluation of Climate Feedbacks

[9] The radiative feedback from surface albedo changes αs can be expressed as

display math(1)

where Q is the top-of-atmosphere (TOA) short-wave radiation, αp is the TOA albedo, Ts is the surface temperature, and S is the TOA incoming solar radiation [Hall and Qu, 2006]. The S term is assumed not to vary on interannual to climate change timescales but must be explicitly included on seasonal timescales to remove the effect of varying incident solar radiation on the feedback term [Cess and Potter, 1988; Tsushima et al., 2005]. The method of evaluating SAF follows the “radiative kernel” method [Soden et al., 2008], with the second term on the right-hand side of equation ((1)) specified as a model independent kernel Kα, being the radiative response resulting from a “standard” perturbation of albedo (here specified as 1%), calculated using a single global climate model (GCM). The third term is derived from monthly mean temperatures taken from the CMIP3 GCMs, with albedo calculated from the ratio of monthly mean values of downward and upward surface short-wave radiation. The kernel technique has proved to be a powerful tool for the evaluation of model feedbacks, including surface albedo [e.g., Soden and Held, 2006; Shell et al., 2008; Sanderson et al., 2009; Dessler, 2012]. Kernels in this study were derived from the Bureau of Meteorology Research Centre (now Centre for Australian Weather and Climate Research) model, using the radiation parameterization of Sun-Edwards-Slingo (SES) [Sun and Rikus, 1999] and are identical to the “SES” kernels discussed in Soden et al. [2008]. Kernels were specified on a 2.5° × 2.5° grid, with model fields interpolated to this grid before processing.

[10] Kernels generated using different radiation schemes and models were compared in Soden et al. [2008]. Results for SAF were found to be a little more sensitive to the choice of model kernel than for other feedbacks, a result attributed by the authors to differences in high-latitude cloud cover. This provides a necessary caveat for the present analysis. Furthermore, although the primary cause of year-to-year variability in high-latitude planetary albedo is surface albedo changes [Qu and Hall, 2007], atmospheric effects can strongly attenuate this effect [Qu and Hall, 2007; Donohoe and Battisti, 2011]. Since the degree of attenuation will be model dependent (in particular depending on the amount of local cloud), this provides an important caveat on results derived from the use of a single kernel.

[11] A useful method of removing the impact of clouds on the changes is to recalculate the kernels for “clear-sky” conditions, i.e., with clouds removed in the determination of the kernels [Soden et al., 2008]. These kernels are then applied in the same way as in equation ((1)), evaluating the impact of surface albedo changes in each model without cloud attenuation (although with small residual attenuation from water vapor). These clear-sky calculations are compared with the “all-sky” results below.

[12] The models analyzed in this study are those submitted to the model archive of CMIP3 [Meehl et al., 2007] and which included the required surface radiation terms for the calculation of albedo (BCCR-BCM, CCSM3, CGCM_T47, CGCM_T63, CNRM, CSIRO-3.0, CSIRO-3.5, ECHAM5/MPI, FGOALS, GFDL-CM2.0, GFDL-CM2.1, GISS-AOM, GISS-EH, GISS-ER, INGV-SXG, INM-CM3.0, IPSL-CM4, MIROC3.2_hires, MIROC3.2_medres, MRI-CGCM, PCM, UKMO-HadCM3, and UKMO-HadGEM1). These models were chosen to permit comparison with the earlier SAF analyses cited above and to complement the cross-timescale investigation for other CMIP3 feedbacks carried out in Colman and Hanson [2012]. Extending these findings to the more recent CMIP5 [Taylor et al., 2012] models is underway and will be the subject of a further study. Feedbacks were calculated for climate change, as well as for interannual variability, decadal variability, and the seasonal cycle. The overall analysis methods for the differing timescales were similar to Colman and Hanson [2012] and so are only briefly described here.

[13] Following Soden and Held [2006], climate change feedbacks were determined from corresponding pairs of months from the beginning and the end of the A1B 21st century scenario experiments (i.e., differencing months in the periods 2000–2010 and 2090–2100). This was followed by land/sea masking, area averaging, and normalization by global or regional mean surface temperature change. The different normalizations highlight the contribution to global feedback and the sensitivity to regional temperatures responses, respectively.

[14] At interannual (decadal) timescales, 100 years of the preindustrial experiment were used, with kernels in equation (1) applied to corresponding months in adjacent pairs of years (decadal means). Annual mean radiation perturbations were calculated from the monthly means and then regressed against the global or area mean temperature. This approach is analogous to that used by Dessler and Wong [2009] for water vapor feedback and by Dessler [2012] for a range of feedbacks, including surface albedo, except that individual pairs of years are used rather than deviations from a long-term mean. The use of pairs of years follows the approach used in Dessler et al. [2008] and Colman and Power [2010] and has the advantage of minimizing the impact of any climate drift which may be present in the long preindustrial integrations.

[15] For calculation of seasonal feedbacks, extending the method used previously in Colman and Hanson [2012], two-monthly differences were determined for temperature and surface albedo, followed by application of the kernel from the intervening month. Thus, for example, the February radiation perturbation was calculated by considering (half) the change in surface albedo between March and January, multiplied by the February kernel. Since the radiation change depends on incident radiation as well as change in surface albedo, following Tsushima et al. [2005], the radiative perturbations were first normalized prior to regression against temperature perturbation, by multiplying by [SA]/[S], where S (SA) is the monthly (annual) mean in solar radiation, and the square brackets represent area averaging. As in Tsushima et al. [2005], effects on radiation changes of subregional variation in amount and angle of radiation were not considered, due to the complexity of such factors.

[16] Three different regions were used for area averaging: global, NH cap, and SH cap, from 30° of latitude to the pole. These caps cover the regions of significant albedo change (see below) and allow an assessment of albedo sensitivity to either global or extra tropical hemispheric temperature anomalies. Hemispheric caps were further divided into land and ocean subregions to separate the contributions of changes in snow and sea ice. The NH land cap is therefore the same region specified by Hall and Qu [2006] for their evaluation of seasonal feedback over northern continents. Note that area average incident radiation, radiation perturbation, and temperature changes were first determined prior to the application of equation ((1)). This is consistent with the common definition of large-scale feedbacks [e.g., Bony et al., 2006] and permits calculation of the contribution of separate regions to the global albedo feedback [e.g., Winton, 2006a]. The approach of Hall and Qu [2006] differed slightly from this by weighting the albedo change by local incident radiation prior to area averaging.

[17] Regression standard errors were used to determine confidence limits of the feedbacks, except for the climate change case where errors were calculated from spread in the 10 years of the (annual) feedback values.

3 Results and Discussion

[18] “Box-and-whisker” plots of global feedbacks are shown in Figure 1, along with the separate contributions from snow and sea ice NH/SH “caps” (radiation perturbations over these areas having been regressed against global mean change and then scaled by their fractional contribution to the total global area). Note that what is categorized as “sea ice” here includes all changes over nonland points and will therefore also include any albedo effects from changes in snow overlying sea ice. Multimodel average global feedback strength is similar in climate change and decadal and slightly weaker/stronger for interannual/seasonal. Two models (CGCM_T47 and ECHAM5/MPI) show negative albedo feedback on both interannual and decadal timescales, although these are not statistically different from zero at the 10% level. A third model, CSIRO-3.0, has negative decadal feedback, this time significant. No model has negative global feedback for climate change or seasonal, although the spread, particularly for seasonal, is large.

Figure 1.

Box-and-whisker plots of (all-sky) feedback strength (W/m2/K) for (a) climate change, (b) decadal, (c) interannual, and (d) seasonal albedo feedbacks. Shown are the global feedback, then the contributions to the global feedback from the four regions (NH/SH, snow/sea ice). Each box represents the 25th and 75th percentiles (the interquartile range), the white band represents the median, and the ends of the whiskers represent the range of values falling within 1.5 interquartiles from the median. Outliers beyond this are shown as red stars. Diamond depicts the average.

[19] NH and SH sea ice, and NH snow make roughly equal contributions to the global feedback strength for climate change, decadal, and interannual timescales, with a marginally increased contribution from northern sea ice for decadal. All timescales show only a small contribution from SH snow, consistent with the small area of extra polar snow and implying minimal changes to Antarctic albedo. Because of the method of calculation of the feedback (regressions against temperature), and because of small feedback contributions equatorward of 30°, individual area contributions are not guaranteed to sum to the global feedback. Nevertheless, mean model areas do sum to within around 10% of the global values for all timescales (not shown), making this a useful division. For seasonal variation, SH contributions are negative in all models due to the dominating role of NH temperatures in the annual cycle. NH snow plays a disproportionately strong role in global feedback on seasonal timescales, consistent with very large changes in seasonal continental snow cover.

[20] The zonal mean profiles of SAF at the four timescales for each of the models are shown in Figure 2. Climate change feedbacks were calculated as zonal mean radiative perturbations normalized by global temperature change. Other timescales show regression coefficients of zonal mean perturbations against global temperatures. There are significant uncertainties from these regressions (not shown on the plot), particularly for interannual and decadal timescales, due to small global temperature changes and limited sampling due to lengths of available time series [Colman and Hanson, 2012]. For example, the mean uncertainty for the interannual regressions is around 2 W/m2/K at 60°, increasing to 5 W/m2/K at the pole. Model “spread” is lowest for climate change and seasonal and is particularly large for decadal. As expected, contributions occur overwhelmingly poleward of 30°, although decadal and interannual timescales show some low latitude contributions. The negative SH seasonal contributions reflect the fact that the global mean temperature deviations used in the regression are dominated by NH seasonality. Multimodel means are compared in Figure 2e (although note that this is for qualitative discussion only as there are significant uncertainties on these means).

Figure 2.

Zonal mean albedo feedbacks (W/m2/K) for the CMIP3 models (identification key right) for (a) climate change, (b) decadal, (c) interannual, and (d) seasonal timescales. For the climate change case, values are calculated by normalizing zonal mean radiative change by global mean temperature change. For all other timescales, results are from regression of zonal mean against global mean changes. Refer to text for details. Thick black lines represent the mean of the regressions (normalizations) for the 23 models. (e) Summary of the means for Figures 2a–2c and, for comparison, the absolute value of the mean for Figure 2d. Note that the y axis scales differ.

[21] Despite the enormous intermodel spread, multimodel mean distributions of climate change, decadal, and interannual surface albedo feedbacks are broadly consistent. Notably, the SH profile is similar in all, with a suggestion of extending slightly poleward in the climate change case. In the NH, the greatest differences lie north of 75°, with the climate change feedback roughly twice the magnitude of the other two timescales. This is consistent with the significant high-latitude Arctic sea ice reduction occurring under climate change, but less under shorter-term temperature variations, a feature noted by Hall [2004]. Hall postulated that this was due to land constraining the sea ice boundary, making short-term sea ice changes weaker. In contrast, the unconstrained and thin-edged SH sea ice distribution is more sensitive to short-term fluctuations. Although this can potentially account for the agreement between multimodel SH (sea ice dominated) albedo profiles, this may disguise different processes. For example, Hall [2004] also noted the compounding effect of “ice thickness feedback,” which may operate quite differently at different timescales. Furthermore, the gradient in high-latitude temperature changes per degree of global temperature increase is different across the three timescales [refer to Colman and Hanson, 2012, Figure 1], although all show amplified high-latitude warming. Finally, the extremely large disagreement between models at interannual and decadal timescales also underlines that the very close agreement across timescales (e.g., in the SH) is likely fortuitous and that strong conclusions should not be drawn from the meridional profiles. The area-derived feedbacks in Figure 1 indeed do not suggest that net sea ice albedo feedback in the NH is stronger for climate change than for other timescales. This may reflect stronger contributions from interannual and decadal sea ice responses equatorward of 75° or may be an artifact of the averaging of regression coefficients. These differences should be examined using CMIP5 models, particularly with the availability of longer preindustrial runs, and in particular the availability of ensembles, potentially reducing regression uncertainties.

[22] What relationships, if any, exist between feedbacks over these different areas and timescales? Regressions between regions and across timescales, along with 90% confidence intervals and explained variances, are shown in Tables 1a and 2a. Tables 1b and 2b show the results from equivalent regressions for “clear-sky” feedbacks. For climate change timescales, NH sea ice and snow albedo feedbacks are uncorrelated, and SH/NH sea ice albedo feedbacks are only weakly correlated, confirming the findings of Winton [2006a]. A weak positive correlation is found between SH snow and sea ice albedo feedbacks (not shown), although as shown in Figure 1, SH snow albedo feedback is negligible on global scales. By contrast, NH snow and sea ice albedo feedbacks are positively correlated at decadal, interannual, and seasonal timescales. This is the case for both all-sky and clear-sky feedbacks. Unfortunately, the current analysis does not detail the differences in the physical processes operating at these different timescales, and further research would be necessary to explore them. The strength of the snow/sea ice albedo feedback correlation depends on whether the feedbacks are calculated by regression against area average temperatures or global mean temperatures. For “area normalization,” correlation coefficients for decadal, interannual, and seasonal are 0.59, 0.67, and 0.69, respectively, for the all-sky case. In all cases, explained variance is substantially weaker when calculated from regressions against global mean temperatures (not shown), consistent with the relatively weak (compared with climate change) correlations between regional and global temperature changes at these timescales [e.g., Colman and Power, 2010]. Given the uncertainties involved in estimating regression coefficients, further evidence for a robust relationship is needed at decadal timescales in particular (despite there being a statistically significant relationship). SH and NH sea ice albedo feedbacks are also positively correlated at interannual and seasonal timescales, with coefficients of 0.59 and 0.55, respectively, for the all-sky case. Again, correlations are lower when regressions are performed against global temperature change (not shown).

Table 1a. SAF Regression Coefficients (and the 90% Confidence Interval) for NH Sea Ice Against NH Snow Feedbacks (Top Row) and SH Against NH Sea Ice Feedback (Bottom Row)a
 Climate ChangeDecadalInterannualSeasonal
  1. a

    Values statistically significantly different from zero are highlighted in bold. Explained variance (r2) is shown in brackets. In all cases, feedbacks are calculated by regression against (or normalization by) corresponding area averaged temperatures.

NH sea ice/snow0.02 ± 0.670.79 ± 0.511.62 ± 0.820.42 ± 0.28
SH/NH sea ice0.67 ± 0.600.79 ± 0.860.62 ± 0.401.19 ± 0.85
Table 1b. As in Table 1a Except Values Are Calculated Based on Clear-Sky Feedbacks
 Climate ChangeDecadalInterannualSeasonal
NH sea ice/snow-0.20 ± 0.680.79 ± 0.471.36 ± 0.670.42 ± 0.28
SH/NH sea ice0.63 ± 0.540.71 ± 0.740.63 ± 0.391.18 ± 0.86
Table 2a. SAF Regression Coefficients (and the 90% Confidence Interval) From Fits of Different Timescale Feedback Versus Climate Change Feedbacka
Regression Against Climate ChangeDecadalInterannualSeasonal
  1. a

    Values significantly different from zero are highlighted in bold. Explained variance (r2) is shown in brackets. In all cases, feedbacks are calculated by regression against (or normalization by) corresponding area averaged temperatures.

Global-0.05 ± 2.00.29 ± 1.1-0.08 ± 1.1
NH snow1.10 ± 1.21.16 ± 0.710.98 ± 0.38
NH sea ice-0.53 ± 1.3-0.10 ± 1.5-0.15 ± 0.32
SH sea ice1.20 ± 1.80.70 ± 1.10.21 ± 0.50
Table 2b. As in Table 2a Except Values Are Calculated Based on Clear-Sky Feedbacks
Regression Against Climate ChangeDecadalInterannualSeasonal
Global0.17 ± 1.20.18 ± 0.620.42 ± 0.64
NH snow1.40 ± 1.401.19 ± 0.920.97 ± 0.40
NH sea ice-0.08 ± 1.40.10 ± 1.5-0.07 ± 0.32
SH sea ice1.27 ± 1.80.87 ± 1.10.20 ± 0.54

[23] Do statistically significant relationships occur between climate change and other timescales, such as those found by Hall and Qu [2006] for April NH snow? If so, this could form an important basis for evaluating models and constraining climate change feedbacks. To test this, regressions were performed between climate change global and regional feedbacks and corresponding decadal, interannual, and seasonal values, with results shown in Table 2 for both all-sky and clear-sky feedbacks. For SH and NH sea ice, no statistically significant correlations were found at the 10% confidence level. Only for NH snow were statistical correlations found, and only for interannual and seasonal timescales. The three scatterplots for northern snow all-sky feedbacks are shown in Figure 3, along with “error bars” and lines of best fit. Positive correlations are found in all three, although not statistically different from zero for the decadal case (for either all-sky or clear-sky; see Table 2). The seasonal/climate change feedback correlation is not the same as that of Hall and Qu [2006], due to calculation differences (including regression using the full annual cycle, rather than only northern spring). Nevertheless, it confirms a strong positive correlation for northern continental snow on these timescales. Large error bars on the interannual and decadal timescale data points preclude stronger conclusions but suggest that this relationship also holds for these cases, and indeed, the relationship is statistically different from zero at the 10% level for interannual for both all-sky and clear-sky cases. The question arises whether a relationship would need to hold at all intervening timescales for it to provide a meaningful constraint for climate change feedback. This issue is not resolved here, but until greater precision is available at decadal timescales, the lack of statistical significance should not rule out the possibility of useful constraints from shorter timescales.

Figure 3.

NH snow climate change feedback strength (area normalized) versus (a) decadal, (b) interannual, and (c) seasonal NH snow albedo feedbacks. Data points represent individual models, and error bars represent the 90% confidence from regressions of area mean radiation perturbations against regional area average surface temperature (decadal, interannual, and seasonal) or from the corresponding estimate of feedback strength from the 10 year sample (climate change)—refer to text. These plots correspond to the second row of Table 2a. Red lines are regressions through data points assuming zero errors in x values and identical y standard deviations. Regression equations and explained variance are also shown. All units are W/m2/K.

[24] It thus appears that snow albedo feedback is a “special case” in that stronger feedback at one timescale implies stronger feedback at others. This may be due to consistent physical processes operating over land over these different timescales. For example, Qu and Hall [2007] noted that the sensitivity of snow albedo to climate change depended on the treatment of features such as vegetation canopy/surface interactions, with high albedo contrast implying greater climate change sensitivity. Processes such as this are likely to apply on other timescales, too. Such may not be the case for sea ice, however, due to more complex interaction with the underlying ocean, particularly from “ice thickness” feedback, giving different effects on differing timescales [Hall, 2004]. The details of these processes are not explored further here as they are beyond the scope of the present paper. Nevertheless, they suggest that without further elaboration of processes, clear constraints on climate change feedback from observations of the current seasonal cycle or from interannual or decadal variability may be readily applied to only around one third of the total SAF.

[25] It remains possible, of course, that an investigation of further models may give different conclusions. Notably, the uncertainties associated with evaluation of decadal and interannual feedbacks are very large and may be reduced by examining longer time series and multimember ensembles which are now becoming available under CMIP5 [Taylor et al., 2012]. This therefore warrants further investigation.

4 Summary and Concluding Remarks

[26] This study has analyzed climate change SAF responses from CMIP3 models and compared them with those related to the annual cycle (“seasonal”) and from “naturally occurring” interannual and decadal variability. Calculations of the feedbacks use the “kernel” technique of Soden et al. [2008]. Climate change feedbacks were determined by normalizing TOA radiative perturbations by surface temperature changes. Variability-related feedbacks were determined by regressing global radiative perturbations against corresponding surface temperature changes. Meridional distributions of feedbacks were determined using linear regression/normalization of zonal mean radiation perturbations against temperature.

[27] Mean global scale feedbacks are similar across all four timescales, albeit a little stronger for seasonal. The breakdown between contributions from northern and southern snow and sea ice areas is similar. SH snow contributions are negligible, with roughly equal contributions from the other three regions for climate change, decadal, and interannual. Seasonal is an exception to this, although the global temperature change dictates offsetting contributions from Northern and Southern Hemispheres. NH snow also provides a disproportionately strong seasonal contribution. The meridional distribution shows strong similarities between climate change, decadal, and interannual timescales for the multimodel mean, with close agreement in the SH sea ice, but suggests climate change feedback is stronger over Arctic sea ice north of 75°N. However, the spread of model feedbacks diagnosed from decadal and interannual variability are particularly large, as are uncertainties from the zonal mean regressions, so strong quantitative conclusions cannot be drawn.

[28] Relationships between feedbacks in these differing regions reveal significant positive correlations between NH snow and sea ice albedo feedbacks, and SH and NH sea ice albedo feedbacks for decadal, interannual, and seasonal timescales. These are weak for climate change feedbacks. These relationships are confirmed for “clear-sky” conditions, indicating that they are not sensitive to the cloud cover peculiarities of the radiative kernel chosen. These “climate variability” feedbacks are found to be stronger when calculated from regressions against regional temperature changes, rather than global. Regression of feedbacks across different timescales reveals no statistically significant relationships, with the exception of NH snow albedo feedback. For NH snow albedo feedback, significant, positive correlations are found between climate change and interannual and between climate change and seasonal timescales, the latter consistent with the findings of Hall and Qu [2006]. There is a suggestion of a similar relationship with decadal timescale feedback. These results imply that NH snow is a “special case” in revealing such strong relationships across timescales and indicates that the seasonal/climate change particularly remains the most promising analogue. These findings are similar when clear-sky calculations are used.

[29] Although large uncertainties remain in the calculation of feedbacks at some timescales (particularly decadal), it is hoped that the current study can serve as a benchmark for evaluation and comparison of these different aspects of SAF in models. The newer intercomparison carried out under the CMIP5 experiment [Taylor et al., 2012] provides the opportunity to repeat and update the current analysis. It would be also fruitful to examine other techniques for reducing some of the uncertainties from the large model spread. In particular, results could be stratified based on skill in reproducing aspects of the current climate. For example, weighting could be carried out depending on its ability to reproduce short-term feedbacks at appropriate latitudes, from sea ice or snow variations, or from total feedback.


[30] I am particularly grateful for the extensive script development, data handling, and data analysis carried out by Lawson Hanson. Thanks also to Josephine Brown, Martin Dix, and two anonymous reviewers for helpful comments. I acknowledge the modeling groups for making their simulations available for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP3 model output, and the WCRP's Working Group on Coupled Modelling for organizing the model data analysis activity. The WCRP CMIP3 multimodel data set is supported by the Office of Science, U.S. Department of Energy. This work has been undertaken as part of the Australian Climate Change Science Program, funded jointly by the Department of Climate Change and Energy Efficiency, the Bureau of Meteorology, and CSIRO.