The primary dataset for this study is the Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder Extended (APP-x) (Key et al., 2016). APP-x consists of twice-daily 25 km composites at two local solar times in the Arctic (04:00 and 14:00) and Antarctic (02:00 and 14:00) starting in 1982. Data from 1982 through 2015 at 14:00 local solar time (high sun) are employed. APP-x includes surface temperature, surface broadband albedo, sea ice thickness, cloud properties (coverage, optical depth, effective particle size, thermodynamic phase, and top pressure), and radiative fluxes at the surface and TOA. In APP-x, the retrieval of surface albedo involves four steps. First, the reflectances of the two shortwave channels are converted to a broadband reflectance. Then, the TOA broadband reflectance is corrected for anisotropy and atmospheric attenuation and converted from TOA broadband albedo to a surface broadband albedo. Finally, the surface clear-sky broadband albedo is adjusted for the effects of cloud cover in cloudy pixels over snow and ice (Key et al., 2001). The reflectance is also corrected for dependencies on sun-satellite-surface viewing geometry. Uncertainties in the retrieval of surface albedo are larger in cloudy-sky conditions than in clear-sky conditions. Downwelling fluxes at the surface are computed with a neural network, called FLUXNET, which is trained to simulate a radiative transfer model (Key and Schweiger, 1998). The neural network uses derived geophysical variables as input (Key and Wang, 2015). To determine the absorbed shortwave energy at the surface, the downwelling shortwave flux was multiplied by the surface absorption (1 minus albedo) for each pixel. More details of the algorithms are described in Key et al. (2016) and references therein.

The study areas are land and non-land between 60 and 90^∘ N latitude, where land is typically snow covered and ocean is ice covered during the winter, except for parts of the North Atlantic Ocean. Land includes Greenland. Non-land is almost exclusively ocean but does include some inland lakes. For simplicity, we use “ocean” to mean non-land throughout the rest of this paper. Over this domain, the land and ocean areas contain a similar number of equal-area pixels, with land areas consisting of 26 682 25 km pixels and ocean areas consisting of 27 674 pixels, a difference in area of 3.58 %.

Figure 1 Average annual surface shortwave absorption (W m⁻²) from 60 to 90^∘ N for the combined land and ocean area (purple), land only (orange), and ocean only (cyan). Dotted lines are linear trends.

Download

APP-x data show that annual mean absorbed solar radiation at the Arctic surface has increased over the 1982–2015 period (Fig. 1). The magnitude of absorption and the rate of increase, however, were different for land and ocean. Trends in surface albedo, surface temperature, cloud cover, and shortwave radiation are calculated using annual mean values with a linear least-square fit regression over the 34-year period, and confidence of the trends is calculated using two-tail Student's t test. Over land, the average increase in absorption was 0.21 W m⁻² year⁻¹, significant at the 90 % confidence level; over ocean, the average increase was 0.43 W m⁻² year⁻¹, significant at the 99.9 % confidence level. The shortwave absorption increase over ocean was, therefore, approximately 2 times as large as the increase over land. Absorption over the ocean increased by 0.3 % of the annual mean ocean absorption per year, resulting in an approximate 10 % increase over 34 years. Over land, the increase was 0.09 % of the annual mean per year, or about 2.7 % over the study period. The increased absorption over land can be attributed to the decreasing snow cover and hence decreasing albedo, especially in spring (Robinson and Frei, 2000; Déry and Brown, 2007). The increased solar absorption over ocean can be attributed to the shrinking sea ice cover (Pistone et al., 2014; Polyakov et al., 2012). Including or omitting Greenland in the calculations for land has a relatively small impact on the results. If Greenland is excluded, the average annual mean shortwave absorption over land increases by about 18 W m⁻² but the strength of the absorption trend is slightly weaker. Greenland's high albedo results in less shortwave absorption than other Arctic land areas, but the decrease in this albedo over time, especially over Greenland's coastal areas, contributed to a stronger absorption trend. Excluding Greenland decreases the absorption trend over land from 0.09 % of the annual mean to 0.06 %.

The larger trend over ocean than land results from the larger albedo difference between dry, snow-covered sea ice (greater than 0.8) and open water (0.1) (Rösel et al., 2012) than between snow-covered land (0.85) (Greenfell and Perovich, 2004) and land during the melting season (0.2–0.4) (Sturm et al., 2005). Though the change in shortwave absorption over ocean areas outpaces that of land, the greater magnitude of absorption over land, i.e., the actual amount of energy absorbed, is due to greater insolation at lower latitudes. The radiative feedbacks associated with these changes in absorption over both land and ocean are discussed later.

Figure 2Trends in absorbed radiation for selected months over ocean (a) and land (b).

Download

Figure 2 shows the spatial pattern of shortwave absorption trends over the Arctic for April, May, June, and September. These months were chosen because they illustrate the changes during the annual transition from high to low snow cover over land (April and May), high to low sea ice cover over ocean (June), and the annual sea ice minimum (September). Over Arctic land, the strong increase in absorption due to decreasing springtime snow cover (Robinson and Frei, 2000; Stone et al., 2002) is seen in May. Absorption trends in northern Europe, central Siberia, and the Alaskan interior are particularly affected by this loss in snow, and this spatial pattern of radiative forcing was also seen by Flanner et al. (2011). Land areas show the greatest absorption increase from March through May, with average May absorption increasing by 1 W m⁻² year⁻¹. Some of the increasing absorption trends in the early spring may be caused by changes in vegetation. Land with more exposed shrub experiences albedo decreases earlier in the year than where there is less shrub or no vegetation at all (Sturm et al., 2005). Once temperatures are above freezing, sensible heat flux overtakes solar heating and the impact from vegetation causing lower albedo values is reduced (Loranty et al., 2011; Sturm et al., 2005). This means that changes in absorption over snow-covered vegetative land during the summer months are primarily driven by changes in snow cover, not vegetation. Chapin et al. (2005) determine that on cloud-free summer days, broadband albedo over the Alaska North Slope has been reduced by 0.0002 year⁻¹ due to changes in vegetation from 1982 to 1999.

In contrast, most of the sea ice lasts through early summer, but changes in sea ice thickness and the formation of melt ponds still allow for changes in absorption (Perovich and Polashenski, 2012). Sea ice albedo typically decreases with thickness (Lindsay, 2001), and an increase in melt pond fraction (open water) further reduces surface albedo. As higher temperatures cause the surface of the sea ice (0.8 albedo) to begin melting, the thin layer of water atop the ice (0.6 albedo) can reduce the absolute albedo by 20 %. Liquid water more readily absorbs radiation than the surrounding ice and causes more water to pool and create melt ponds, further reducing the ice concentration and albedo of an ice-covered surface (Rösel et al., 2012). Melt ponds that appear early in the melting season allow for greatly increased absorption over sea ice, and may even drive regional-scale sea ice changes in extreme cases (Rösel and Kaleschke, 2012). By late February or early March, sea ice concentration and extent reach their annual maximum under weak sunlight, so absorption trends over the Arctic Ocean are very small. From June to October, however, the multi-decadal changes to the extent, thickness, and the surface albedo of summer sea ice caused the absorption rate to increase faster than absorption over land, particularly in the Beaufort and Chukchi seas. Flanner et al. (2011) also noted that increases in radiative forcing from 1978 to 2008 over lower-latitude Arctic seas were greater than those over land during June–October. Sea ice extent and concentration have decreased over the last few decades, and thick multiyear sea ice that was prevalent in the 1980s and 1990s has lost as much as 50 % of its thickness (Kwok and Rothrock, 2009), if not vanished altogether (Serreze et al., 2007). First-year ice is more susceptible to the formation of melt ponds, which can cause precipitous decreases in albedo (Rösel et al., 2012). The increase in surface absorption over the Arctic Ocean, then, is due to a combination of the replacement of multiyear sea ice with first-year ice and open water over the study period.

Figure 3Trends in absorbed shortwave radiation over land (a) and cloud cover trends over land (b) during September.

Download

While the increase in the absorption of shortwave radiation is largely due to reductions in sea ice and snow cover extents, the linear correlations between snow cover anomalies or sea ice extent anomalies and shortwave absorption anomalies are both approximately −0.6 (not shown). Regional and seasonal changes in cloud cover explain some of the variance in these relationships. The 34-year trends in cloud cover were explored using APP-x data from 1982–2015. Over land, an increase (decrease) in highly reflective cloud cover is associated with decreases (increases) in surface absorption. For example, Arctic land areas that have experienced an increase in cloud cover (Alaska, western Russia, and north central Siberia) show decreasing trends in shortwave absorption. The spatial variability in the surface shortwave absorption over land in Fig. 2 can be explained, in part, by trends in cloud cover. Figure 3 provides an example for September, for which both positive and negative trends in cloud cover over eastern Siberia show a strong relationship with trends in absorbed solar radiation. While portions of the Arctic Ocean have also experienced changes in cloud cover, their effect on trends in shortwave absorption are much less, primarily because most of the ocean is still ice covered and the reflectivities of ice and cloud are similar. We found that the trends in absorbed shortwave radiation over land are more affected by changes in cloud cover than over the ocean and that trends in cloud cover can result in radiative absorption increases or decreases over land during the period of study, as shown in Fig. 3.

While it can be seen qualitatively that the regional effect of clouds can be large, quantitatively determining their overall influence on the trend in absorbed shortwave radiation, i.e., to separate the influence of changes in cloud cover from changes in sea ice and snow cover, is not possible with the data available. Instead, we quantify the contribution of clouds by determining their maximum possible effect on downwelling shortwave radiation at the surface over the study period. This is performed by using the 34-year average downwelling shortwave surface flux for each of the sunlit months (March–September) and the 34-year average cloud cover trend (fractional cloud cover) to determine the changes in instantaneous surface shortwave flux. At each grid point in each month, the 34-year average downwelling flux is multiplied by the cloud cover trend. A positive cloud cover trend will result in a decrease in the downwelling and vice versa. For this calculation it is assumed that all clouds are optically thick (“black”) and reflect almost all incident sunlight, as optically thick clouds would have the maximum effect on downwelling shortwave radiation. This assumption is valid based on Wang and Key (1995), who found that visible optical depths for Arctic clouds are in the range of 5–6, corresponding to a transmittance of near zero (0.2 %–0.6 %). The average cloud cover trends over land and ocean are −0.265 % and −0.392 %, respectively, or 10 % and 4 % of the change in shortwave absorption. For March–September, changes in cloud cover from 1982 to 2015 resulted in an increase in surface absorption by 1.94 W m⁻² over land and 2.19 W m⁻² over ocean. These cloud-based changes in surface absorption account for only a 0.5 % increase in surface insolation over the ocean and a 0.4 % increase over land.

Even though September experienced the greatest decrease in sea ice extent, the smaller incoming solar flux at this time of year results in smaller absorption increases than those of early summer. The early spring, late fall, and winter months exhibit far weaker trends in shortwave absorption over ocean than land due to lower variability in the sea ice cover and smaller solar fluxes – decreasing to zero in the winter – at the high latitudes.

Surface radiation and cloud cover data from the National Aeronautics and Space Administration (NASA) Modern-Era Retrospective Analysis for Research and Applications 2 (MERRA-2) reanalysis (Rienecker et al., 2011) are employed to provide verification of the results from APP-x. This study used MERRA-2 version 1.3 and determined the absorbed shortwave radiation trends at the surface from the surface incoming shortwave flux (SWGDN) and surface albedo (ALBEDO) variables.

Figure 4Trends in absorbed radiation from APP-x over land (a) and ocean (b) compared to trends from MERRA-2 over land (c) and ocean (d) during June.

Download

Figure 5Day-of-year range between 0.4 and 0.25 of albedo over ocean and land (blue) from 1982 to 2015. The dotted trend line (red) shows the regression of the DOY midpoint (pink) over the time period.

Download

Performing the same analysis as before on MERRA-2 data produced similar results. The trends in absorbed radiation for the month of June from APP-x and MERRA-2 show similar patterns, though with larger magnitudes in APP-x (Fig. 4). The reanalysis data show an increase in absorption over ocean during June and mixed trends over land, which correspond spatially to APP-x trends. The results were consistent with APP-x, with increasing, uniform ocean heating during high summer and changes over land influenced by factors other than surface albedo. The most obvious differences between the reanalysis data and APP-x occur over the central Arctic Ocean, where MERRA-2 absorption trends are weaker than those in APP-x. The cause of this difference is due to the fixed albedo value that MERRA-2 assigns to sea ice, which does not take sea ice thickness or melt ponds into account. As seen in the APP-x results, thinner ice and irregularities in the ice surfaces increase the absorbed surface radiation.

The trends in solar energy absorption at the surface are both a result of, and a forcing for, changes in surface albedo. As increasing solar absorption over the Arctic continues to affect land and ocean differently, we now explore how the timing of the low-albedo portion of the year has changed over time and how the timing relates to the available solar energy. Markus et al. (2009) determined that between 1979 and 2007, nearly all regions of the Arctic showed a trend towards earlier annual melting and later refreezing, which self-enhances as sea ice thickness decreases. Results presented here are consistent with their analysis and expand upon the surface energy implications.

Using APP-x data, we are able to track the changes of land and ocean albedo throughout the study period. The impacts on the surface energy budget are apparent in Fig. 2. However, the absolute timing of the low-albedo period as well as the shift in timing of this period over the last few decades require further examination. One approach to analyzing these changes in the land and ocean albedo is to determine the day of year (DOY) on which the average albedo over land and over ocean reached their minima for each year. However, due to late freezing and thawing events and dynamically driven changes in the sea ice edge, changes in the albedo minimum DOY do not accurately explain trends in absorbed solar energy over the last 34 years. We find that using the DOY range from when the Arctic transitioned from a relatively low albedo (the day that albedo first went below 0.4) to a very low albedo (the day that albedo went below 0.25) provides a better metric for comparing the changes in albedo over land and ocean (Fig. 5). Figure 5 shows that the majority of the snow cover over land melts earlier in the year than sea ice, which is due to higher sun and temperatures at a lower latitude. Terrestrial snow cover also melts earlier because the snow-free land adjacent to snow-covered land warms faster than the unfrozen ocean around the sea ice.

An examination of Fig. 5 shows that the Arctic has reached a lower-albedo state increasingly early in the calendar year over both land and ocean since 1982. A linear fit of the midpoint between the days of year at which the 0.4 and 0.25 albedo levels were reached shows a decrease of 0.64 days year⁻¹ over ocean and 0.62 days year⁻¹ over land over the last 34 years. Both of these trends are significant at the 99.9 % confidence level. Furthermore, the rate at which the albedo is decreasing from 0.4 to 0.25 has accelerated. Over ocean, a linear fit of the length of the interval showed that it took over 16 days for the average ocean albedo to decay to 0.25 from 0.4 in the initial years of the study. By the end of the record, this albedo decrease took just 8 days (significant at the 99.9 % confidence level). Over land, the change in rate of albedo decrease showed a prominent decrease from 17 to 9 days over 34 years, although the statistical confidence level is less than 90 %.

The regression in time of the low-albedo period towards earlier in the year over both land and ocean may have important radiative implications in the future. Over ocean, the low-albedo period was reached 2 weeks closer to the summer solstice (DOY 172) in 2015 than in 1982–1985, with the low-albedo range midpoint going from DOY 188 to DOY 167. Over land, the low-albedo range midpoint regressed nearly 20 days away from the summer solstice, closer to DOY 152. Though both land and sea experienced lower albedos migrating closer to the beginning of the year, the low-albedo period over land now occurs before the summer solstice, while the low-albedo period over the ocean occurs closer to the solstice and therefore at a time with much greater solar insolation. Even though the insolation during the low-albedo period is greater today than it was in the early portion of the study, the midpoint of the low-albedo interval has regressed past the summer solstice in the last few years (Fig. 5). This implies that current trends in sea ice changes may cause the albedo transition to occur even further towards the beginning of the year, thereby experiencing weaker insolation, similar to the regression of the low-albedo period over land. As such, the differences between land and ocean absorbed shortwave radiation trends may grow smaller as their albedo transition occurs earlier in the year.

The magnitude of insolation on any given day at the peak solar time is greater at lower latitudes. Therefore, even small changes in albedo in the lower Arctic can have large effects on the amount of energy absorbed at the surface. Conversely, large changes in albedo at higher latitudes are required to significantly affect shortwave absorption due to the weaker instantaneous insolation at higher latitudes. For instance, in 1982, the average albedo of all ocean pixels at 75^∘ N was 0.345 on 1 July. By 2015, the average ocean albedo on that date had decreased to 0.234, a change of over 11 % (absolute). The corresponding change in average absorbed shortwave energy at 75^∘ N on 1 July between 1982 and 2015 was 14.3 W m⁻². In contrast, the average land albedo at 65^∘ N on 1 July decreased only 1.6 % (absolute) between 1982 and 2015, yet the change in absorbed energy over land (4.8 W m⁻²) was 34 % of the change that occurred over ocean. At 75^∘ N, albedo must decrease 3 times as much as it does at 65^∘ N for the same increase in absorption in July, based on differences in the magnitude of insolation.

However, the magnitude of the flux accumulated over the entire day around the summer solstice is larger at higher latitudes. Figure 6 provides a simple illustration of the changes in the accumulated TOA incoming shortwave flux at 65 and 75^∘ N. For an equivalent change in albedo, the accumulated absorbed TOA shortwave flux is larger at higher latitudes because, even though the sun is lower, there are more hours of sunlight. The change in TOA absorbed accumulated flux at 65^∘ N is 96 % of the change in accumulated flux at 75^∘ N. This relationship is also true at the surface. The accumulated flux on 1 July was calculated for the average ocean and land surface at 65 and 75^∘ N using albedos from the years 1982 and 2015. Results showed that at both latitudes, the accumulated flux on 1 July increased more over ocean between 1982 and 2015 than over land. Accumulated flux increases over ocean at 75^∘ N (4.73 MJ) were more than twice as high as the changes at 65^∘ N (2.05 MJ). Accumulated flux changes over land at 75^∘ N (3.11 MJ) were also much higher than at 65^∘ N (0.16 MJ). The greater changes in accumulated flux are related to larger albedo decreases at higher latitudes, where snow cover and sea ice may have changed more drastically than at 65^∘ N. Therefore, at both latitudes over the last 34 years, the average ocean pixel has experienced a greater change than the average land pixel.

Figure 6Accumulated top-of-atmosphere incoming shortwave flux for each day and for the 65^∘ N (orange) and 80^∘ N (blue) latitudinal bands, roughly representing the Arctic Ocean and Arctic land, respectively. Darker symbols represent the day of year that the midpoint trend in the low-albedo period (Fig. 5) was reached over land (circle) and ocean (star) in 1982, while lighter symbols show the day of year of the 2015 low-albedo period midpoint trend. Arrows clarify the direction in time for the change in the low-albedo period midpoints.

Download

Figure 6 also shows the changes that occur due to a low-albedo regression towards earlier times of the year. Over ocean, the shift in the timing of lower albedos to earlier in the year means that more sunlight was absorbed over the ocean in 2015 than in 1982, all else being equal (e.g., cloud cover). Over land, the regression of low albedo towards earlier in the year still results in an increase in absorbed energy, but it can only increase modestly due to decreasing sunlight further from the summer solstice. This relationship is valid for both the peak solar time and the accumulated absorbed fluxes.

The increased solar absorption due to the temporal regression of the low-albedo period results in a positive surface albedo feedback. One way to define the strength of the albedo feedback is the change in net incoming shortwave radiation with respect to surface temperature due to changes in surface albedo (Cess and Potter, 1988; Qu and Hall, 2007; Fernandes et al., 2009):

where Q is the net (absorbed) shortwave radiation at the TOA (W m⁻²), I is incoming solar radiation at the TOA surface (W m⁻²), T is temperature (K or C), α_p is the planetary albedo at the TOA, and α_s is the surface albedo. The term I is calculated as the monthly mean incoming solar radiation at the APP-x grid level. The term $\partial {\mathit{\alpha }}_{\text{p}}/\partial {\mathit{\alpha }}_{\text{s}}$ over land and over ocean is calculated using an analytical model developed by Qu and Hall (2007) and surface albedo for all sky and clear sky only, albedo at the TOA for all sky and clear sky only, cloud amount, and cloud optical thickness monthly means from APP-x 1982 to 2015 at the APP-x grid level. Coefficients required in this analytical model, ε₁ and ε₂ in each month, are derived following Eq. (10) in Qu and Hall (2007) with collocated monthly means of planetary albedo at the TOA for all-sky and clear sky, cloud amount, cloud optical depth, and surface albedo at the APP-x grid level as a regression sample; another parameter (coefficient) required in this model is $T_{\text{a}}^{\text{cr}}$, effective clear-sky atmospheric transmissivity, which is derived monthly following Eq. (5) in Qu and Hall (2007), using each collocated planetary albedo at TOA for clear-sky and surface albedo as a regression sample. The term $I\partial {\mathit{\alpha }}_{\text{p}}/\partial {\mathit{\alpha }}_{\text{s}}$ over land and over ocean is calculated following Eq. (12) in Qu and Hall (2007) at the APP-x grid level and then averaged. The term dα_s∕dT over land and over ocean is calculated as the averaged ratio of the monthly surface albedo trend to the monthly surface temperature trend at the APP-x grid level following Fernandes et al. (2009). All procedures follow Qu and Hall (2007) and Fernandes et al. (2009).

For a unit temperature change, the net solar radiation absorbed by the Earth system over ocean is less than that over land in April but about 4 times as large as that over land in June and July (Fig. 7). The feedback strengths in June are 16.3 W m⁻² K⁻¹ over ocean and 3.8 W m⁻² K⁻¹ over land. The stronger surface albedo feedback over the ocean at the high-sun time of the year will amplify the warming effect, allowing for even more solar radiation to be absorbed by the Earth system in the future, pushing the low-albedo threshold back even earlier in the year and leading to a further decline in the Arctic sea ice cover.

Figure 7The snow–albedo and ice–albedo feedbacks (Eq. 1) for Arctic land (orange) and ocean (cyan) for the period 1982–2015.

Download

Data used in this paper are stored publicly and are readily accessible. APP data (https://doi.org/10.7289/V5BC3WHM, Key et al., 2015) and APP-x data (https://doi.org/10.7289/V5MK69W6, Key et al., 2014) are accessible though the National Centers for Environmental Information (NCEI) website. MERRA-2 data (https://doi.org/10.1175/JCLI-D-16-0758.1, Gelaro et al., 2017) are accessible through NASA's Global Modeling and Assimilation Office (GMAO) website portal.

All authors contributed to the writing of this paper. AL performed much of the data analysis and drafted the paper. YL led the snow–albedo and ice–albedo feedback section. JK formulated the research idea and goals and performed some calculations.

The authors declare that they have no conflict of interest.