Consequences of poor representation of Arctic sea-ice albedo and cloud-radiation interactions in the CMIP5 model ensemble



[1] Clouds significantly influence the Arctic surface energy budget and a realistic representation of this impact is a key for proper simulation of the present-day and future climate. Considerable across-model spread in cloud variables remains in the fifth phase of Coupled Model Intercomparison Project ensemble and partly explains the substantial across-model spread in the surface radiative effect of the clouds. In summer, the extensive model differences in sea-ice albedo, which sets the potential of the cloud-albedo effect, are strongly positively correlated to their cloud radiative effect. This indicates that the model's sea-ice albedo not only determines the amount, but also the sign of its cloud radiative effect. The analysis further suggests that the present-day annual amplitude of sea-ice cover depends inversely on the model's sea-ice albedo. Given the present-day across-model spread in sea-ice albedo and coverage, a transition to a summer ice-free Arctic ocean translates to a model-span of increased surface shortwave absorption of about 75 W m−2.

1 Introduction

[2] The Arctic region is warming at a rate almost double that of the global average [Screen and Simmonds, 2010; Serreze and Barry, 2011], manifested, for instance, by the rapid retreat and thinning of the sea-ice [Meier et al., 2007; Stroeve et al., 2012; Kwok and Untersteiner, 2011]. According to simulations of future climate scenarios, this development will carry on in the coming decades [e.g., Winton, 2006] and produce nearly sea-ice free summers in the 21st century [Wang and Overland, 2012]. Several plausible processes to physically explain the observed surface temperature amplification have been proposed in the literature [see Serreze and Barry, 2011, for a review]. The albedo feedback [e.g., Manabe and Stouffer, 1980] and cloud changes [e.g., Francis and Hunter, 2006] are two processes commonly suggested playing a role for the temperature amplification. Diagnosing the relative role of the various factors is, however, complicated even in models [e.g., Kay et al., 2012].

[3] The evaluation of the present-day Arctic climate in global climate models (GCMs) is hampered by atmospheric conditions that are challenging for remote sensing techniques and by a lack of in situ observations. Present-day sea-ice extent, one of the variables best constrained by observations in the Arctic, is reasonably well simulated in GCMs [e.g., Parkinson et al., 2006; Stroeve et al., 2012]. Cloud variables, for which observations are less abundant, show substantial across-model spread [e.g., Eisenman et al., 2007; Karlsson and Svensson, 2011]. Consequently, models also disagree in the radiative surface fluxes. In winter, models in general have a tendency to underestimate the amount of longwave radiation reemitted back to the surface [e.g., Svensson and Karlsson, 2011]. Eisenman et al. [2007] argued that the simulated model spread in surface longwave radiation has to be compensated by across-model sea-ice albedo differences, considering that sea-ice extent is reasonably well simulated in models. In a response, DeWeaver et al. [2008] argued that tuning of sea-ice albedo is not straightforward in a GCM since it will result in a number of chain reactions that feed back on the albedo in a nontrivial way. In their fully coupled model, changes to cloudiness and to multiple scattering strongly dampened the efficiency of the albedo change. However, in other models, sea-ice albedo is effectively used as a tuning parameter [e.g., Uotila et al., 2012].

[4] The main focus of this study is on how model differences in the parameterization of sea-ice albedo in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) [Taylor et al., 2012] influence the cloud radiative effect on the surface energy budget and the annual cycle of sea-ice concentration. Additionally, the simulation of Arctic clouds in CMIP5 are evaluated and the results are compared with those of the CMIP3 model ensemble [Karlsson and Svensson, 2011].

2 Method and Data

[5] In the analysis, the Arctic is defined as the region north of the Arctic circle (66.7°N). Considering the large contrasts in surface energy budget between ice-covered and open ocean, the analysis is separated with respect to surface type. Following Karlsson and Svensson [2011] and Svensson and Karlsson [2011], grid points with monthly averaged sea-ice concentration >80% are considered sea-ice covered. To use >90% as sea-ice cover threshold does not change our main conclusions. The sea-ice concentration is bilinearly interpolated to the atmospheric grid if necessary.

[6] Two retrieval product data sets based on Advanced Very High Resolution Radiometer (AVHRR) data are used (for the period 1982–2004). Observations of total cloud cover and surface radiative fluxes are from the Extended AVHRR Polar Pathfinder (APP-x) product [Wang and Key, 2005]. The surface albedo retrievals used are from APP-x [Key et al., 2001] and from the CM SAF (Climate Monitoring Satellite Application Facility project) CLouds, Albedo and RAdiation dataset from AVHRR data [CLARA-A1, Riihelä et al., 2013; Karlsson et al., 2013]. Both albedo data sets have been validated against summer in situ data from the 1 year Surface Heat Budget of the Arctic Ocean (SHEBA) ice campaign [Persson et al., 2002] showing similar accuracy. Root mean square errors for the SHEBA summer were 0.08 and 0.07 for CLARA-A1 and APP-x, respectively [Riihelä et al., 2013; Key et al., 2001]. The albedos reported by the two data sets are not identically defined. In CLARA-A1, it is an inherent surface reflectance (independent of atmospheric conditions), while the retrieved APP-x surface albedo represents the apparent albedo for all-sky conditions. Over snow and sea-ice, in clear-sky conditions, the apparent albedo is expected to be higher than the inherent albedo [Key, 2002] and cloudy conditions will further increase the apparent albedo [Key et al., 2001].

[7] Model output is from the CMIP5, which underlies the forthcoming Intergovernmental Panel on Climate Change's Fifth Assessment Report (IPCC-AR5). We analyze monthly mean output for the present-day period (1980–2004) of the historical experiment, a simulation where all known forcings are applied in fully coupled GCMs. In Table S1 (supporting information), the 17 GCMs included in the analysis are listed. The model selection criteria was that all the relevant variables for the analysis had to be available in the CMIP5 database. One ensemble member from each model has been used.

[8] Since the surface albedos of the GCMs are derived from the surface shortwave fluxes, they represent the apparent all-sky albedo. For summer average sea-ice albedo, the variations in the area of the sea-ice are also considered. We call this albedo the effective summer albedo:

display math(1)

where the sums are taken over the summer months (i={May,June,July,August}) and F represents the fractional area of the sea-ice.

[9] Sea-ice cover masks for the observational records are derived from the National Snow and Ice Data Center's monthly mean sea-ice concentration from Nimbus-7 Scanning Multichannel Microwave Radiometer and Defense Meteorological Satellite Program Special Sensor Microwave Imager passive microwave data set [Cavalieri et al., 1996, updated yearly] by bilinearly interpolating the sea-ice concentrations to the observational grid before applying the >80% sea-ice concentration threshold.

[10] Besides GCMs and observations, we also include the European Centre for Medium-Range Weather Forecasts Interim reanalysis data (hereon ERA-Interim) [Dee et al., 2011] in the analysis.

3 Annual Cycles

[11] Figures 1a and 1b show the average annual cycles of total cloud cover and surface cloud radiative effect (SCRE, defined as the difference between net all-sky and net clear-sky radiative fluxes) [Ramanathan et al., 1989] over sea-ice covered ocean in the Arctic region in GCMs, reanalysis, and observations. Although the CMIP5 model ensemble median total cloud cover (Figure 1a) agree well with the observed annual cycle of the total cloud fraction, the across-model spread is distressingly large. The spread remains at least as large as it was in the CMIP3 model ensemble [Eisenman et al., 2007; Karlsson and Svensson, 2011]. Most GCMs show an annual cycle in phase with observations, but four models have peak cloudiness during winter. The difference in phase and amount of cloud condensate also remain large. For example, the model ensemble winter spread (and ensemble median) in total water path (ice + liquid condensate) and liquid-to-total water path ratio ranges from 7.1 to 95.9 (33.3) gm−2 and from 0.02 to 0.68 (0.19), respectively (Figure S1). During winter, considerable amounts of liquid condensate has been observed in clouds with cloud top temperatures down to −40° [Shupe et al., 2006].

Figure 1.

Climatological seasonal cycles of (a) total cloud cover (%), (b) surface cloud radiative effect (W m−2), and (c) surface albedo over sea-ice covered ocean, as defined in the text, north of 66.7°N. Colored lines show individual CMIP5 models, grey envelope represents the range of the CMIP3 model ensemble [Karlsson and Svensson, 2011], and black dashed and solid lines represent APP-x and ERA-Interim, respectively. Gray solid line represents CLARA-A1 surface albedo. Periods considered are 1980–2004 and 1982–2004 for models and observations, respectively.

[12] Wintertime, in the absence of solar radiation, the ability of the clouds to reemit longwave radiation to the surface results in a positive cloud radiative effect on the surface energy budget. During the sunlit season, this positive cloud greenhouse effect is competing with a negative cloud albedo effect, which arises because the clouds decrease the amount of incident solar radiation at the surface. The actual cloud influence on the radiation budget does not only depend on the cloud properties themselves but also on the environmental settings. For example, a cloud over an dark surface (e.g., ocean) will have a much more profound influence on the amount of absorbed solar radiation at the surface than an identical cloud overlying a bright surface (e.g., sea-ice). The amount of cloud greenhouse effect depends on the vertical profile of humidity and temperature.

[13] Considering the large spread in cloud properties, the substantial spread in the SCRE is expected (Figure 1b). Note that, due to the SCRE definition, model differences in the clear-sky radiative fluxes propagate to model differences in SCRE, as will be discussed further below.

[14] Satellite-based retrievals indicate that only during peak-summer (July) does the cooling cloud albedo effect exceed the warming cloud greenhouse effect, generating a negative net SCRE over sea-ice covered ocean (Figure 1b). Similar observations were made during the SHEBA campaign [Intrieri et al., 2002]. In agreement with the observations, 6 out of the 17 GCMs show negative net cloud radiative effect only in July, six GCMs never show monthly averaged negative net SCRE, and in the remaining five models, more than 1 month are associated with negative SCRE. The CMIP5 model span in simulated SCRE is similar in winter and larger during summer compared with the CMIP3 ensemble (Figure 1b). It should be noted that the CMIP3 ensemble was smaller (nine models) [Karlsson and Svensson, 2011], this could partly explain the larger spread in the CMIP5 ensemble.

[15] In winter (December-January-February), no significant (at the 95% level) across-model correlations between the SCRE and cloud cover, liquid water path, or ice water path exist. Most likely, this indicates that models show varying amounts of optically thin clouds and/or cloud base temperatures, but different atmospheric profiles of humidity and temperature may also contribute. In summer (May-June-July-August (MJJA)), there is a relatively low but significant (at the 95% level) across-model correlation between the SCRE and total cloud cover (r=0.60). This correlation indicates that in models with higher cloud fractions, the cloud greenhouse effect exceeds the cloud albedo effect, while the opposite is true for models with less clouds. Potentially, this could be explained by a higher fraction of optically thin clouds in models with a larger cloudiness, but there is no significant across-model correlation of cloud water content and SCRE. Instead, as soon will become evident, the substantial spread in simulated summer net SCRE (Figure 1b) is related to the representation of sea-ice albedo.

4 Sea-Ice Albedo

[16] In the transition from spring to summer, the albedo of sea-ice decreases [e.g., Perovich et al., 2002]. When the sun returns from its winter absence, the sea-ice is typically covered in cold snow with high albedo. With decreasing zenith angles, the irradiance will increase, and eventually, the snow starts to melt and the albedo decreases even further when melt ponds start to emerge on the sea-ice. The seasonal large-scale trend is occasionally interrupted with cold spells and snow fall. In early fall, when zenith angles become smaller, sea-ice will freeze up, and with fresh snow, albedo increases again. Although not considered in this study, the age of the sea-ice (through its thickness) also influences the albedo [e.g., Agarwal et al., 2011].

[17] How sea-ice albedo is described varies among the GCMs. In some models, the sea-ice albedo depends on snow cover, ice thickness, and temperature [e.g., Briegleb et al., 2002]. In the more sophisticated sea-ice schemes applied in GCMs [e.g., Holland et al., 2012], sea-ice albedo is derived for different ice categories and albedo adjustments by melt ponds and aerosols on snow/ice are explicitly accounted for. Note that ERA-Interim sea-ice albedo is prescribed to climatological values (based on Ebert and Curry [1993]) and does not explicitly depend on any local surface properties that change in time and space.

[18] In general, regardless of the complexity of the sea-ice albedo description, the seasonal evolution of the sea-ice albedo is represented by the models (Figure 1c). The CMIP5 model ensemble median shows reasonable agreement with the retrievals. However, for individual models, the differences are substantial. For example, the drop in simulated sea-ice albedo from May to summer minimum value ranges from 0.06 (INM-CM4) to 0.32 (MRI-CGCM3) while APP-x and CLARA-A1 drops by 0.27 and 0.20, respectively. There is also large model discrepancy in the absolute values of the sea-ice albedo; the GCMs' extended summer (May, June, July, and August) effective sea-ice albedo ranges from 0.42 (GISS-E2-R) to 0.77 (MIROC5). The model median effective albedo is 0.68. Intermodel difference in mean albedo and in the summer albedo drop is not explained by differences in sea-ice concentration (Figure S2).

[19] Throughout the sunlit season, the APP-x product consistently shows lower albedo values than CLARA-A1. By comparing with in situ data from Greenland, Stroeve et al. [2001] found a negative bias in the APP-x surface albedo of about 6%, which partly could explain the discrepancy.

5 Sea-Ice Albedo - Cloud Radiative Effect Interactions

[20] The very different sea-ice albedo of the models will affect how clouds influence the surface energy budget (Figure 1b). Models with a high sea-ice albedo have a smaller cloud albedo effect than those models having less reflective sea-ice surface. Figure 2 shows a scatter plot of the SCRE normalized with total cloud cover against the sea-ice albedo. If models had identical cloud microphysical properties, cloud base temperatures, insolation, and environmental settings, the normalized SCREs and sea-ice albedos would line-up perfectly in Figure 2, in accordance with the definition of SCRE. Although this is not the case, the figure indicates the across-model variability in SCRE to primarily be related to differences in the simulated sea-ice albedo (r=0.79). In fact, the sea-ice albedo - SCRE dependence is so profound that it determines the sign of the cloudiness normalized net SCRE (the longwave component being independent of surface reflectance). The physical consequence is that a specific change in cloudiness would affect the sea-ice surface energy budget very differently in the different models. The sensitivity of the normalized cloud radiative effect evaluated at the top of the atmosphere to the surface albedo is equivalently large although the model ensemble then agrees on net negative values (not shown).

Figure 2.

Simulated summer (MJJA) net surface cloud radiative effect normalized with total cloud fraction versus summer sea-ice albedo. The gray-shaded area indicates the observed interannual range of summer sea-ice albedo (CLARA-A1 and APP-x) and normalized surface CRE (APP-x).

6 Sea-Ice Albedo and the Sea-Ice Cover

[21] Having a key influence on the surface energy budget during summer, it may be expected that the sea-ice albedo directly will influence the amount of sea-ice retreat during summer. Figure 3 shows the summer-averaged sea-ice albedo versus the annual amplitude of the Arctic sea-ice concentration (the markers indicate the intersect of the averages and the bars the interannual range). The annual amplitude in sea-ice concentration is derived, each year, as the annual maximum minus the annual minimum. The results suggest an across-model relationship such that models simulating higher sea-ice albedo also show less pronounced seasonal cycles in sea-ice coverage. The spread in Arctic sea-ice cover amplitude is attributed mainly to differences in the summer sea-ice minimum concentrations between the models (19–75%), since the maximum sea-ice concentrations show larger agreement (69% to 92%).

Figure 3.

Summer (MJJA) sea-ice effective albedo versus annual amplitude in Arctic sea-ice concentration (%). Colored markers and connected lines show summer average values and interannual range for the CMIP5 models. The gray-shaded area indicates the observed interannual range of sea-ice cover amplitude (NIMBUS) and summer sea-ice albedo (CLARA-A1 and APP-x). The contours show the amount of increased Arctic basin summer absorbed solar radiation (W m−2) expected if summer sea-ice cover completely melts away, assuming an open-ocean albedo of 0.1, a summer-averaged incoming surface solar radiation of 220 W m−2 and a present-day winter sea-ice concentration maxima of 80%.

[22] Somewhat surprisingly, no statistically significant across-model correlation between the summer sea-ice albedo and the linear trend of either minima, maxima, September, or summer-averaged Arctic sea-ice concentrations are found (for the period 1980–2004). Since the natural variability in sea-ice cover is considerable [e.g., Stroeve et al., 2012] and our analysis only utilizes one ensemble member of each model, results of the trend analysis should be treated with caution.

[23] Note that the year-to-year variability in sea-ice albedo is relatively small and no large trend over the period is evident (Table S2). The latter is unexpected, considering the period analyzed is associated with a general shift from multiyear to seasonal ice and that this transition would result in a sea-ice albedo decrease since seasonal ice is thinner and has a higher melt pond fraction [Perovich and Polashenski, 2012; Agarwal et al., 2011].

7 Conclusions and Discussion

[24] The across-model spread in Arctic cloud cover and cloud condensates is substantial, and no improvement is seen from previous model intercomparisons [Karlsson and Svensson, 2011]. This diversity of simulated Arctic clouds in the CMIP5 ensemble contributes to a spread in the models' cloud influence on the surface energy budget.

[25] During the sunlit season, model differences in sea-ice albedo explain more of the across-model spread in the SCRE than cloud fraction or the amount of cloud condensates do. The potential of the cloud-albedo effect increases with decreasing sea-ice albedo. Consequently, for models associated with higher sea-ice albedo, the warming cloud greenhouse effect exceeds the cooling cloud albedo effect, resulting in a net positive cloud influence on the surface energy budget. For models with less reflective sea-ice, the cooling cloud albedo effect is stronger and these models are associated with a negative SCRE. The implication is that the radiative signature to a specific cloud change would not only differ in size, but also in sign.

[26] There is also a significant across-model correlation between the sea-ice albedo and the annual amplitude of Arctic sea-ice concentration, such that models with smaller sea-ice albedo are associated with larger annual sea-ice amplitudes (originating from larger summer melt off).

[27] Recent studies predict near ice-free summer conditions in the Arctic ocean during the 21st century [e.g., Stroeve et al., 2012; Wang and Overland, 2012]. Thus, the region will experience a large change in the amount of surface solar radiation absorbed, a change that is influenced by clouds, and how clouds change but ultimately is conditioned by the present-day sea-ice characteristics. For example, a model with larger present-day sea-ice extent is potentially prone to larger future changes in absorbed solar energy than a model with more modest present-day sea-ice extent because larger areas of highly reflective sea-ice will be exchanged for low albedo ocean surface. The magnitude of the albedo drop depends, apart from the present-day sea-ice extent, also on the sea-ice albedo. This dependence can be illustrated by applying a simple theoretical model of the resulting increase in surface solar radiation absorption (ΔSW) to a transition from a partially sea-ice covered to an ice-free ocean. The isolines in Figure 3 give the average summer change in ΔSW as a function of present-day sea-ice annual amplitude and albedo, assuming an open-ocean albedo of 0.1, a surface incoming solar radiation of 220 W m−2 (CMIP5 ensemble mean), and a winter maximum sea ice concentration of 80%. The simple model translates the CMIP5 ensemble spread in sea-ice amplitude and sea-ice albedo to a spread in the ΔSW of about 75 W m−2 (Figure 3), which is attributable to both differences in the simulated sea-ice albedo and sea-ice amplitude and to the nonlinear interplay between them. Thus, the transition to ice-free conditions will impact the Arctic surface energy budget very differently in the models and could, for example, result in intermodel discrepancies in the change in sea surface temperature, ocean circulation, and the meridional circulation.

[28] The sea-ice extent is the sea-ice parameter best constrained by observations [Kattsov et al., 2010] and, naturally, also the parameter most commonly evaluated in models. Compared to CMIP3, the representation of sea-ice extent is improved in the CMIP5 model ensemble [Stroeve et al., 2012]. Although being a key component of summer surface energy budget, the albedo of sea-ice (to the authors knowledge) has previously not been evaluated for a large model ensemble. Less reliable large-scale observations may be one explanation. The observational records of sea-ice albedo do show discrepancies, but the spread among models is substantially larger. The fact that present-day sea-ice albedo is so badly constrained in GCMs impacts the fidelity of future scenario assessments of the Arctic region and should therefore be a concern for the modeling community.


[29] We thank Thorsten Mauritsen for suggestions and fruitful discussion. We also acknowledge the World Climate Research Program's Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table S1 of this paper) for producing and making available their model output. For CMIP, the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. The research leading to these results has received funding from the European Union, Seventh Framework Program (FP7/2007-2013) under grant agreement n°244067.

[30] The Editor thanks James Screen and two anonymous reviewers for their assistance evaluating this manuscript.