Since the beginning of the satellite era, sea ice in the Northern Hemisphere (NH) has decreased in September extent, thinned substantially, and undergone earlier and longer melt seasons [1
]. Snow cover in the Arctic is also changing. The area of NH snow cover is on the decline with increased interannual variability and longer melt seasons [3
]. At the same time, Arctic surface and air temperatures have increased 2–3 times faster than the global average [5
]. In 2016, the average surface air temperature for 60–90° N was +2 °C above the 1981–2010 baseline, a deviation more than twice as large as the global average of +0.8 °C [6
]. These changes in the Arctic cryosphere and temperature are related through a series of feedbacks collectively known as Arctic amplification.
These feedback pathways manifest themselves through changes in the Arctic energy budget [7
]. At the top of the atmosphere (TOA), the energy balance is determined solely by the exchange of incoming shortwave (SW) and outgoing longwave (LW) radiative fluxes. The amount of SW radiation that is absorbed versus reflected is, in turn, determined by the planetary albedo. At the surface, latent and sensible heat fluxes also play a role, offsetting some of the excess radiative fluxes. While the surface and TOA are linked, only excess energy at the surface can melt sea ice and snow or heat the surface. Decreases in sea ice extent and snow cover have, in turn, been linked to decreases in surface and TOA albedos over Arctic waters in observations [8
]. Sea ice has a much higher albedo than open ocean, so when sea ice melts the ocean absorbs more SW radiation. More SW radiation at the surface warms the ocean and further melts the sea ice, creating the ice-albedo feedback central to Arctic amplification [11
Yet while sea ice and snow play a key role in the Arctic energy balance, their influence is strongly modulated by the atmosphere and, in particular, cloud cover [13
]. Clouds affect both LW and SW radiation throughout the atmospheric column. They can decrease the SW radiation that reaches the surface as well as increase the down-welling LW radiation. Thus, clouds modulate surface melting, warming the surface by trapping thermal radiation or cooling it by reflecting SW radiation [14
The presence of clouds also has implications for the ice-albedo feedback. We can readily see in satellite imagery that the contrast between ice-covered and open water is completely obscured by opaque cloud cover (Figure 1
). Studies have found that the TOA albedo changes less in response to reduced surface cover than the surface albedo. Using observations from the Earth Radiation Budget Experiment (ERBE), Gorodetskaya et al. [8
] found that sea ice and snow cover in high latitudes have reduced effectiveness, 0.22 and 0.16, respectively, on the TOA albedo. Kato et al. [16
] also concluded that clouds reduced the effects of sea ice loss on reflected SW radiation at the TOA in the Arctic Soden et al. [17
] found a roughly 30% reduction in surface albedo feedback kernels when using all-sky as opposed to clear-sky fluxes in global climate models. Hwang et al. [18
] found that clouds reduced the surface albedo feedback radiative kernel over the Northern High Latitudes from 1.13 ± 0.44 Wm
to 0.49 ± 0.30 Wm
However, while our knowledge of Arctic cloud properties has improved over the past decades with increased satellite observations [19
], the precise magnitude of the modulating influence of clouds is still unresolved. For example, the record low sea ice extent in 2007 was caused, in part, by the combination of reduced cloud cover and anomalously high solar absorption at the surface with an already weakened ice pack [20
]. In response to Kay et al. [20
], Graversen et al. [22
] found that the ice-albedo feedback exacerbated sea ice melt but claimed the true cause of the low sea ice extent increased downward LW emission and turbulent fluxes from anomalously high atmospheric energy transport. Schweiger et al. [23
] completely ruled out the negative cloud anomaly as contributing to the low sea ice minimum in 2007, and more recently Ding et al. [24
] declared that trends in summer atmospheric circulation are the dominant drivers of sea ice decline during the satellite era. However, these latter studies used data from reanalyses that have been shown to exhibit biases in surface radiation in polar regions [25
]. This study quantifies the modulation of TOA albedo and sea ice loss by clouds using satellite observations and compares the results with reanalyses.
The TOA albedo is influenced by both the surface and the atmosphere, the latter being heavily influenced by clouds [26
]. Past studies have partitioned the TOA albedo into surface and atmospheric components both globally and at high latitudes [27
]. Observations reveal that the atmosphere dominates the TOA albedo across the Earth, even in the Arctic where the surface albedo is quite high. Although the surface does not make the largest contribution to the TOA albedo, it is responsible for the majority of variability
in the Arctic TOA albedo. From Qu and Hall [27
], more than 50% of TOA albedo interannual variability is due to the changes in the surface albedo. Since the atmospheric contribution is dominated by clouds, the albedo partitioning scheme from [28
], discussed in more detail later, can be further used to quantify the modulating effects of clouds on TOA albedo response to changing surface cover in the Arctic and evaluate their representation in reanalyses.
Reanalyses have been used extensively to study Arctic climate since they provide continuous temporal and spatial coverage as well as variables that can be difficult, if not impossible, to directly measure [29
]. These challenges also mean that there are few measurements to validate reanalyses in the Arctic because there are relatively few in situ observations in this remote and harsh region. However, some recent studies have found large biases in reanalyses for cloud properties, including cloud fraction, liquid water path, and ice water path, when compared with available ground and satellite observations [32
]. Such discrepancies are intertwined with surface radiative biases, where reanalyses still struggle to match observations in the Arctic [36
]. Cao et al. [39
] investigated the representation and trends of surface and TOA albedos in four reanalyses, finding that none accurately captured the decline of albedo in recent decades. Yet Cao et al. [39
] only compared surface albedos on annual and seasonal scales. Other studies have looked at planetary albedo, but the ability of reanalyses to correctly partition the TOA albedo into surface and atmospheric components has not been studied.
This work seeks to answer two fundamental questions: How do clouds modulate the impact of surface cover changes on the TOA albedo? And how well are these effects captured in modern reanalyses that are often used in polar climate studies? We partition the surface based on sea ice and snow cover to show the effects of surface cover on the energy balance over a time period with large surface cover variability. The surface partitioning is further applied to the TOA albedo and its contributions to show their sensitivities to changes in surface cover. We use new metrics to quantify the effects of clouds on the TOA albedo by comparing the all-sky and clear-sky TOA albedo contributions. These metrics are further used to evaluate five modern reanalyses in their representations of clouds and surface cover feedbacks in the Arctic.
3. How Does Planetary Albedo Respond to Surface Cover?
3.1. Effects of Surface Cover on TOA Albedo
The surface and planetary albedos exhibit distinct annual cycles that depend on surface type. Mean monthly albedos over the Arctic domain for 2002–2012 calculated from CERES all-sky fluxes are shown in Figure 5
for each surface condition. The Arctic-mean surface albedo reaches a peak of 0.68 in March when snow and sea ice cover are both at their maxima. Surface albedo decreases with snow/ice cover throughout the summer reaching a minimum of 0.23 in August, approximately one third of the spring maximum. Spatially, (Figure 6
) the exposed land and open ocean have much lower albedos once sea ice and snow melt. Evidence of surface melt is seen in the central Arctic ocean beginning in June where observed surface albedo decreases despite the high SIC. Land areas with high SCF through the summer, e.g., the Greenland ice sheet (GIS), maintain high surface albedos, >0.8, throughout the year due to their perpetually glaciated surfaces. They are the brightest surfaces from June to October.
The TOA albedo has a markedly different annual cycle than that of the surface. In Figure 5
, the TOA albedo peaks in February and is at a minimum in July, leading the surface albedo by a month. Furthermore, the amplitude of the annual cycle is dramatically reduced: the February (0.63) and December (0.61) maxima are less than the maximum surface albedo, and only 50% larger than the July minimum (0.41), which is almost twice as large as the minimum surface albedo. The TOA albedo also varies less spatially across the Arctic than the surface albedo, seen in Figure 6
. While the enhanced reflection from the GIS and the sea ice extent in the central Arctic ocean are visible at the TOA, these signals are less pronounced and the strong contrasts between open ocean and sea ice seen at the surface are muted at the TOA.
These findings are consistent with previous work that has shown that the TOA albedo is dominated by the atmospheric contribution (i.e., clouds) that masks the surface albedo [27
]. Figure 5
demonstrates that, without exception, the atmospheric contribution to the TOA albedo is consistently larger than that of the surface, even over the brightest surfaces (e.g., GIS) throughout the year. The atmospheric contribution therefore dominates the seasonal cycle of TOA albedo, rising to 0.58 in winter and falling to 0.35 in July, accounting for an average 84% of the TOA albedo. As with the TOA albedo, the atmospheric contribution is also much less varied spatially across the Arctic than the surface albedo (Figure 6
), although there is still contrast between land and ocean in early spring.
Although the surface contribution to the TOA albedo is proportional to the surface albedo, its average annual behavior is quite different. The spring peak has a maximum of 0.21 in April when surface ice and snow cover are high and cloud cover is relatively low [65
], contributing 35% of the TOA albedo. By June, the surface contribution decreases to 0.1, about 10% of the TOA albedo, when snow on land has largely receded, and remains low for the remainder of the year. These surface contributions are 2–3 times smaller than the actual surface albedo. As the surface albedo decreases, clouds play an increasingly dominant role in defining the TOA albedo since the surface reflects less SW radiation. The one exception to this trend is the GIS where the surface contribution increases to 0.28 in summer relative to spring and fall. This is likely a result of the fact that as wetter snow at mid-latitudes recedes, snow cover in the summer is dominated by the high-altitude, brighter snow on the GIS.
and Figure 6
both clearly indicate that the large differences in surface albedo between the different surface partitions are muted at the TOA. The surface albedos for fully ice- and snow-covered surfaces are roughly 3–5 times greater than ice- and snow-less surfaces in all months when these surfaces are present (as seen in Figure 3
b, on average, land without snow is only present April through September). Conversely, the TOA albedo over snow- and ice-covered surfaces are at most 1.7 times larger than their bare counterparts. The differences in surface albedo are being masked by clouds.
The contributions to TOA albedo also have notably different behaviors across the different surfaces. The surface contribution to the TOA albedo shows similar patterns to the surface albedo across surface types: uncovered ocean and land are low throughout the year; partially covered ocean and land follow the Arctic-wide average; and fully snow-covered land is larger than all other surfaces through the summer. Although the surface contribution has small absolute differences relative to other albedos, the maximum difference between surface contributions is a factor of 13 greater between snow-covered land (0.27) and open ocean (0.02) in the summer. In contrast to the surface contribution, the atmospheric contribution behaves similarly for all surface types throughout the year. The maximum difference between atmospheric contributions is only a factor 1.33 between 0.3 (land with no snow) and 0.4 (ice-covered ocean) during the summer, a fractional difference ten times less than the surface contribution. This shows that the atmospheric contribution has reduced dependence on the underlying surface.
Comparing Figure 5
and Figure 6
, the relative roles of the atmosphere and surface contributions to the TOA albedo variability can be summarized as follows: the annual cycle of Arctic-wide average TOA albedo is dominated by the annual cycle of the atmospheric contribution while the surface contribution is responsible for variations in the spatial pattern of TOA albedo between surface types. This is clear since the TOA albedo over most surfaces, except for snow-covered land surfaces, exhibit similar annual cycles as the Arctic-mean (and the atmospheric contribution). Snow-covered land has a larger TOA albedo during summer when the Arctic overall sees a decrease; however, the peak at the TOA (0.61) is still damped compared to at the surface (0.78).
3.2. Cloud Modulation of Ice-Albedo Relationships
To cast these atmospheric effects into energetic units, the difference between all-sky and clear-sky contributions is multiplied by TOA solar insolation, giving the cloud impacts on surface and atmospheric contributions to TOA albedo. Recall that the difference between all- and clear-sky atmospheric contributions represents the amount of reflected SW radiation owing to clouds, and the difference for surface contributions shows the amount of SW radiation that would have been reflected at the surface if clouds were not present. When summed together these cloud effect contributions have the same magnitude as the TOA SW CRE, which is opposite in sign to TOA LW CRE and larger April through September.
The atmospheric contribution cloud impact (Figure 7
a) follows the cycle of solar insolation with near zero reflection in winter and a June maximum of 100 Wm
for the Arctic-wide average. The different surface types follow the same pattern as the Arctic average, but surface cover clearly exerts a strong influence on the magnitude of reflected SW due to clouds. During months with significant solar insolation, clouds exert a much stronger influence over less reflective surfaces. For example, clouds reflect nearly twice as much SW radiation over open ocean in June (115 Wm
) than over land with all snow (62 Wm
). It is this compensating cloud effect that explains how the atmospheric contribution dominates the TOA albedo while itself having negligible dependence on surface cover.
This result is illustrated more directly by the cloud impact on surface contribution in Figure 7
b. In general, clouds reduce the surface contribution to TOA albedo because they block radiation reflected at the surface from reaching the TOA, but this effect is much more pronounced over brighter surfaces that reflect more SW radiation in the absence of clouds. The overall magnitude of this cloud masking
effect is 5–6 times larger over ice- and snow-covered surfaces than open water or bare land. Collectively these analyses quantify two important effects: the extent to which surface conditions modulate clouds effects on the planet’s albedo and the masking influence of clouds on the effect of surface albedo changes on SW absorption.
While different surfaces have a strong influence on surface albedo, clouds significantly reduce the magnitude of these variations across the seasonal cycle. From a climate perspective, this leads us to ask: to what extent do clouds further modulate the ice-albedo feedback on longer timescales? While the length of the data record examined here is too short to examine trends, the large variations in SIC (Figure 2
) over the period examined allow the impacts of clouds on year-to-year surface cover variations to be quantified.
In pursuit of answering this question, we examine the response of monthly mean Arctic albedos and the TOA albedo contributions to average SIC and SCF for March through September in Figure 8
and Figure 9
. In each month, the surface albedo and the surface contribution both exhibit strong positive relationships with SIC as expected (Figure 8
). Both surface albedo and TOA albedo decrease from March through July. From July to August the TOA albedo increases due to the increased atmospheric contribution during these months, as opposed to the surface albedo which is reduced further as sea ice approaches its September minimum. The atmospheric contribution has a parabolic shape that depends less on surface cover and more on clouds.
In addition to these larger patterns between albedos and SIC, there are also trends within individual months owing to interannual variations in SIC. Linear fits for each month reveal that the sensitivity of surface albedo to SIC is quite constant (∼0.3) July through September (Table 2
). TOA albedos and average SIC have less spread and no consistent trends in early spring (March–May). Beginning in June, surface albedo increases linearly with SIC. The surface contribution also increases with SIC but exhibits smaller trends due to the cloud masking effects described above. Further evidence of these effects is observed in the much weaker variation of monthly mean TOA albedo with SIC. The atmospheric contribution is fairly constant after July. This indicates that clouds are not particularly sensitive to changes in surface cover, consistent with the findings of Kay et al. [19
] who note that cloud feedbacks may be limited to fall in the Arctic.
The relationships between albedos and SCF, Figure 9
, are largely determined by the cycle of snow cover. The largest interannual variations in SCF occur in the melt season (May and June) and accumulation season (September). In May and June, the surface albedo decreases as the snow melts, with sensitivities of 0.26 and 0.20, respectively (Table 3
). To a lesser extent, variations in the timing of snow cover accumulation in September lead to a similar relationship between surface albedo and SCF (0.37) but over a smaller range of SCF. Surface albedos in early spring and late summer have no significant relationship with SCF because there is either widespread snow cover or widespread bare land, as seen in Figure 6
. The TOA albedo and its contributions follow this same pattern to varying degrees. TOA albedo has an overall linear relationship with SCF for March–September, but when focusing on individual months, May and June are again the months with statistically significant sensitivities, approximately half (0.14 and 0.10) those of the surface albedo. As with SIC, the surface contribution and surface albedo have similar patterns, but the surface contribution is roughly five to ten times smaller. The sensitivity of the atmospheric contribution is relatively large during May and June (0.12 and 0.07) but there is notable spread. While the TOA albedo is slightly more sensitive to SCF (0.16) than SIC (0.14) in May, by June SIC (0.25) has a stronger influence than SCF (0.10) at the TOA.
3.3. Representation in Reanalyses
Clearly clouds exert a significant influence on how strongly the effects of surface cover change influence the Arctic radiation balance. Reanalyses are also frequently used to study climate variations in this region of few observations. Given their pervasiveness in studying the Arctic regional processes and driving models [67
], it is imperative to assess how well reanalyses capture the observed modulating effects of clouds.
It is, therefore, important to ask how well reanalyses represent these cloud effects in the Arctic. The five reanalyses considered generally capture the shape of the annual albedo cycles but not necessarily their amplitudes. Given that a 0.05 difference in June albedo corresponds to 25 Wm
difference in summer, biases of this magnitude are sufficient to exert a significant influence on surface processes, as noted by Cao et al. [39
]. Winter months account for less than 5% of incoming SW radiation and have large errors in observations (Figure 10
), so large biases during these months are less concerning than the rest of the year. In line with that fact, the largest differences in surface albedo occur in the winter, with over-estimations on the order of 0.10–0.20 but within the uncertainty of observations. The reanalyses perform better during the rest of the year, with surface albedo biases ranging from 0.01 (MERRA-2) to 0.1 (NCEP R2).
The spread in TOA albedo in the reanalyses is large in summer months owing primarily to the underestimation by ASR. ASRv2 moderately improves relative to ASRv1 but still underestimates TOA albedo by up to 0.08 (40 Wm) in the summer peak. The performance of the other reanalyses vary more by season. For example, ERA-Interim predicts a smaller annual cycle in TOA albedo than CERES, as well as surface albedo, while the minimum in NCEP R2 lags observations by a month.
c demonstrates that the discrepancies in TOA albedo largely result from differences in the atmospheric contribution. The annual cycles of biases between reanalyses and observations map closely onto those in TOA albedo. ASRv1, ASRv2, and NCEP R2 underestimate the atmospheric contribution to TOA albedo for all months, while MERRA-2 and ERA-Interim overestimate it for all but the winter months. This is not necessarily surprising since the atmospheric contribution is dominated by clouds, which is a common problem for reanalyses [32
Biases in the surface contribution to the TOA albedo are significant in spring when the spread in reanalyses approaches a factor of four. Both ASR versions and NCEP R2 represent the spring maximum in surface contribution relatively well, overestimating it by 0.05 or less. ERA-Interim and MERRA-2, on the other hand, underestimate the surface contribution to TOA albedo throughout the year with biases on the order of 0.1 (50%) during the spring maximum. This difference is enough to completely offset a positive bias in the atmospheric contribution in MERRA-2, artificially leading to a good agreement with observed TOA albedo that masks substantial biases in the partitioning of energy between the atmosphere and surface. In ERA-Interim, the large bias in springtime surface reflection leads to a substantial over-estimation of SW absorption in the spring (March–May).
presents albedos and contributions averaged over March-September for each surface condition. Except for NCEP R2, which overestimates the albedo of all ocean surfaces, the reanalyses capture surface albedos for the surface partitions. Potentially more serious, however, are the biases in ERA-Interim over land with and without snow. ERA-Interim predicts an average surface albedo over land without snow (0.31) that is over twice that seen in observations (0.14). ERA-Interim also underestimates the surface albedo of snow-covered land (0.62 versus 0.71 in observations), which leads to a much lower contrast between land with and without snow (0.31) compared with observations (0.57). This suggests that the effects of snow cover changes may be significantly underestimated in ERA-Interim. In general, the other reanalyses represent variations in TOA albedo with surface conditions fairly well. The contributions to TOA albedo, however, are more widely spread relative to observations and each other. Both versions of ASR and NCEP R2 have positively biased surface contributions and negatively biased atmospheric contributions (≤0.1), compared to observations, that tend to cancel in the TOA albedo. Inversely, ERA-Interim and MERRA-2 have positive biases in atmospheric contributions that are compensated by negative biases in their surface contributions across most surface types. Thus, the apparent agreement in Figure 12
a derives from a fortuitous cancellation of atmospheric and surface biases that have significant implications for how energy is partitioned between the atmosphere and surface in the reanalyses.
The observations suggest that clouds play an important role in moderating sea ice/snow cover albedo effects. To accurately predict surface warming and melt rates associated with climate responses, reanalyses must also capture these relationships between albedos and clouds. The biases in Figure 11
c,d can be traced to errors in these relationships. While the three reanalyses with available clear-sky fluxes capture the annual cycle of cloud TOA albedo contributions in Figure 12
a,b, they all vary in magnitude for the summer maximum, with a range of ±20 Wm
for the atmospheric contribution and ±15 Wm
for the surface contribution. The maxima and minima often lag observations by a month as well. This bias is also reflected at the TOA but is partially compensated by an enhanced atmospheric contribution.
When partitioned by surface type (Figure 12
c,d), the three reanalyses exhibit mixed behaviors. MERRA-2 and both versions of ASR capture the cloud effects on atmospheric and surface TOA albedo contributions. ERA-Interim shows the opposite trend—the cloud impact on the atmospheric contribution is larger over sea ice than open water and there is no clear dependency of it on snow cover. All three reanalyses show that the surface contribution cloud effect increases with surface brightness, although ERA-Interim overestimates the change between high and low albedo surfaces by nearly a factor of two over oceans (45 Wm
versus 23 Wm
Some of the biases in reanalyses albedos can be attributed to their representations of clouds. Average cloud fraction over the Arctic up to 82° N is compared between CloudSat/CALIPSO and reanalyses, Figure 13
. ERA-Interim matches observations fairly well May through October and overestimates the amount of clouds November through April. MERRA-2 is also near observations in summer and fall, although it underestimates cloud fraction during March-October and again overestimates cloud amount in winter. NCEP R2 underestimates cloud fraction for the entire year by 0.2–0.3. While none of these reanalyses directly assimilate cloud observations, the prognostic schemes used by ERA-Interim and MERRA-2 seem to perform better than the diagnostic cloud scheme from NCEP R2. The reader is directed to [34
], for more detailed investigation of cloud representation in reanalyses.
The reanalyses capture broad relationships between surface cover and albedos but struggle to accurately represent the relative magnitudes between different surfaces and TOA albedo contributions. To test the implications of these differences on the representation of interannual variability in the Arctic, the sensitivities of albedos and contributions to SIC in reanalyses and observations for June and September are presented in Figure 14
and Table 4
. Surface albedo and surface contribution sensitivities to SIC are well represented by all five reanalyses in September and marginally less so in June. All reanalyses show that surface albedo is sensitive to SIC in June but have a range of 0.13 (ASR) to 0.85 (NCEP R2) with observations falling in the middle (0.57). Larger biases are present in the TOA albedo sensitivity to SIC. ASRv1 and ASRv2 show essentially no influence of SIC on the TOA albedo in June, contrary to observations. NCEP R2 tends to overestimate the influence of SIC on the TOA albedo. Reanalyses predict sensitivities closer to observations in September, although the MERRA-2 predicts no significant influence of SIC and NCEP R2 again overestimates the influence.
The differences between observations and reanalyses at the TOA can be found in the surface and atmospheric contributions. In June, all the reanalyses overestimate the influence of SIC on the surface contribution. They either underestimate its influence on the atmospheric contribution (MERRA-2, NCEP R2) or predict the opposite relationship (ASRv1, ASRv2, ERA-Interim). These competing biases result in greatly reduced TOA albedo sensitivity to sea ice, except for NCEP R2 that has a much larger surface contribution compared to observations. Most reanalyses are better aligned with observations in September when their biases in SIC influence on TOA albedo contributions are smaller.
4. Discussion and Conclusions
While Arctic amplification is an important component of climate change, the mechanics are not fully understood, particularly the effects of clouds. In this paper, we have quantified how clouds modulate albedos in the Arctic. The surface albedo changes with snow and sea ice cover through the seasons, but the TOA albedo varies only half as much. Reanalyses do not capture this relationship, generally predicting that SIC has less influence on the TOA albedo than seen in observations. In satellite observations, the atmosphere contributes 2–3 times more to the TOA albedo than the surface throughout the year. Clouds are the main contributor to the atmospheric contribution, increasing the atmospheric reflectivity and blocking SW reflected from the surface from leaving the Earth system. The atmospheric contribution to TOA albedo also depends less on the underlying surface, meaning that changes to the surface cover may have less impact on it. Despite differences in methods, our sensitivities are similar to the radiative effectiveness reported by [8
The reduced sensitivity of TOA albedo to surface cover is important for the ice-albedo feedback. Our work supports previous studies that have found reduced ice-albedo feedback parameters due to clouds [17
]. We have found that clouds mask the surface albedo and damp changes in surface cover at the TOA. When the surface albedo is sensitive to SIC changes in the summer and fall, the surface contribution to the TOA albedo is low, leading to reduced changes at the TOA. There is nuance, though. Clouds do not simply replace underlying snow and ice. While clouds have higher albedos than open ocean, there is still a measurable difference (∼0.15) in TOA albedo between land with and without snow cover and ocean with and without sea ice cover. Clouds may reduce the ice-albedo feedback, but the radiative effects of clouds at the TOA are unlikely to be large enough to prevent the ice-albedo feedback from continuing and contributing to Arctic amplification.
Since the interaction between clouds and albedos is essential to understanding and modeling the evolution of the Arctic, it is important that datasets used to study the Arctic, i.e., reanalyses, accurately portray these relationships. These results demonstrate that reanalyses cannot always be taken as proxies for observational datasets, especially in regions where limited observations are assimilated and for variables that cannot be directly measured. In such instances, models rely heavily upon parameterizations that are frequently based on lower latitudes where more abundant observations are available. To that end, the observational benchmarks provided here are also used to evaluate modern reanalyses. While no single reanalysis perfectly represents these variables for all months, most trends seen in observations are captured to some extent. The timing of annual maximum and minimum albedos and TOA albedo contributions are fairly accurate in reanalyses, but their magnitudes vary considerably. Reanalyses show that surface and TOA albedos and the surface contribution to TOA albedo all increase with higher snow and ice surface cover, to varying degrees. The reanalyses also show that the atmospheric contribution to TOA albedo varies with SIC and SCF, but this is not seen in the observations. While averaged surface and TOA albedos have relatively low biases compared to observations, the partitioning of TOA albedo contributions and surface cover vary far more. The consequences of this are highlighted in the differing sensitivities of reanalyses to SIC where reanalyses show a spread of changes in response to SIC. These discrepancies are important as reanalyses are frequently used to study the rapidly evolving Arctic and drive other models, and differences between reanalyses will affect the outcome depending on the chosen reanalysis (e.g., [36
Many of the biases found in partitioning reanalyses are related to their representations of clouds. MERRA-2 and ERA-Interim overestimate the atmospheric contribution during summer and underestimate the surface contribution for the entire year. This is likely due to clouds that are too bright given the reasonable cloud fractions during summer. Overestimating the cloud effect on the TOA albedo surface contribution leads to higher TOA albedos which means too little SW absorption. This in turn reduces melt rates at the surface. NCEP R2 has too few clouds from their diagnostic scheme leading to overestimated surface contributions and underestimated atmospheric contributions for much of the year. ASRv1, and ASRv2 likely suffer from too few clouds as well. It has been noted that the microphysics used in ASRv1 is known to under-represent liquid water in polar clouds [70
]. Although the cloud microphysics was updated in ASRv2, it does not seem to have a large impact. ASRv1 also has large radiative biases at the surface, particularly SW fluxes in excess of 40 Wm
at high latitudes. While ASRv2 improved these biases, they still deviate from observations by approximately 20–30 Wm
. The trade-off with a more complex microphysics scheme is that there are more inputs that must be fine-tuned, meaning there are more chances to be wrong. Reanalyses with simpler cloud schemes, e.g., ERA-Interim, have more realistic outputs, but their results may be due to better tuning and assimilation rather than accurate representation of the physics.
This work has shown the behavior of surface and TOA albedos with changing surface cover in the Arctic. Because the atmosphere contributes more to the TOA albedo than the surface, we have found that clouds have a damping effect on the TOA albedo. While we have touched on the sensitivity of albedos and TOA albedo contributions to SIC and SCF, more work should be done to quantify these relationships.