Unraveling driving forces explaining significant reduction in satellite-inferred Arctic surface albedo since the 1980s

Decreased Albedo.

The Advanced Very High Resolution Radiometer (AVHRR) and the Clouds & Earth’s Radiant Energy System (CERES) satellite data products have been widely used to investigate Arctic climate variability and trends (17, 39). We use versions of these satellite-based monthly mean surface and planetary (top of atmosphere) albedo data sets (Materials and Methods) to compare to and build confidence in the Community Atmosphere Model version 5 (CAM5) simulated changes over the Arctic, and then diagnose the model output to better understand the cause of simulated albedo changes. To compare the CAM5 results with satellite-based albedo, we use the mean albedo for the months from March to September. Mean surface albedo over the entire Arctic (60 to 90°N) is obtained using the surface downwelling shortwave flux as the weight (Materials and Methods). Fig. 1A shows a decreasing trend in both satellite-inferred and modeled multimonth mean Arctic surface albedo from 1982 to 2014. The trend from 2 AVHRR-inferred surface albedos is −1.25 ± 0.34 and −1.51 ± 0.41% decade⁻¹, respectively, somewhat steeper than that derived from the CAM5 simulation (−0.89 ± 0.26% decade⁻¹). The CAM5 surface albedo is mostly in between the 2 AVHRR-inferred surface albedos with a correlation coefficient (R) larger than 0.8. All of the trends and the correlation coefficients are statistically significant at the 99% confidence level. Relative mean bias (RMB) of modeled surface albedo with respect to the 2 AVHRR surface albedo estimates is −1% and 2%, respectively. There is a large spread between the 2 satellite products, indicating remaining uncertainties in the satellite-inferred surface albedo trends. Fewer assumptions are needed in retrievals of top of atmosphere (TOA) planetary albedo than in the satellite-inferred surface albedo. Here we also include a comparison of CAM5 results to CERES-based TOA albedo product over Arctic oceanic areas poleward of 60°N (40). The CAM5-simulated planetary albedo trend of −0.84 ± 0.19% decade⁻¹ for 1980 to 2014 agrees well with the CERES-based estimate (−0.88 ± 0.17% decade⁻¹) in Fig. 1B (trends and the correlation coefficient are statistically significant at the 99% confidence level). The RMB of CAM5 planetary albedo is −2%. Taking into account the uncertainties in the satellite-derived products, the CAM5 model with prescribed sea ice gives a reasonable estimate of the Arctic albedo and trend.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Albedo trend. Area- and energy-weighted average of multimonth (March to September of each year) mean (A) surface albedo over the entire Arctic and (B) planetary albedo over the Arctic oceanic area (north of 60°N). The solid lines are linear least-squares fit to the corresponding time series with the trend (in % decade⁻¹) marked in colors with the 95% confidence interval as uncertainty at the bottom. All trends are statistically significant at the 99% confidence level. In A the black, blue, and green dots are, respectively, CAM5 model results, APP-x, and CLARA-A2 satellite-inferred surface albedo from 1982 to 2014. The correlation coefficient (R) between CAM5 and APP-x results is 0.8 (P < 0.01) and the RMB of the simulation is −1%. The R between CAM5 and CLARA-A2 results is 0.88 (P < 0.01) and the RMB of the simulation is 2%. In B the red and black dots are CERES and CAM5 planetary albedo from 1980 to 2014, respectively. The R between CAM5 and CERES results is 0.85 (P < 0.01) and the RMB of the simulation is −2%.

The satellite-model comparison is based on March through September mean results. Before we use the CAM5 simulation to perform the attribution analysis, the annual-mean surface albedo over the Arctic during 1980 to 2014 is further evaluated against reanalysis products (SI Appendix, Fig. S1). The trend of −0.96% decade⁻¹ in the CAM5 simulation is within the range of (−0.7, −1)% decade⁻¹ given by 3 different reanalysis products. Thus, the model simulation, satellite datasets, and reanalysis products all indicate an ∼1% decade⁻¹ decreasing trend in annual-mean surface albedo over the Arctic since the 1980s, but the primary contributor to this albedo reduction is not obvious.

Surface Albedo Trend Attribution.

Given the important role of both snow cover and sea ice in surface albedo, detailed and internally consistent CAM5 results can be used to attribute the contributions from snow cover fraction (SCF) and sea ice fraction over the Arctic (Fig. 2). The surface of each model grid may include bare land, snow, sea ice, and/or open ocean, for which the albedo varies and cannot be directly isolated from the grid mean radiative fluxes. To perform the attribution analysis of trends in total albedo we express the monthly grid mean albedo (α) in the formα=SCF⋅fland⋅αlandsnow+(1−SCF)⋅fland⋅αland+(1−fland−fice)⋅αocean+SCF⋅fice⋅αicesnow+(1−SCF)⋅fice⋅αice,[1]where SCF,fland, and fice are monthly mean snow cover fraction, land fraction, sea ice fraction for a given grid cell, respectively; and αlandsnow,αland,αocean,αicesnow, and αice denote albedo of snow-covered land, snow-free land, open ocean, snow-covered sea ice, and snow-free sea ice, respectively. Mean values for αocean,αicesnow, and αice are determined to be 7.88%, 84.08%, and 30.91%. Annual-mean climatology of αland and monthly mean climatology of αlandsnow are quantified at each grid cell where fland>0 (Materials and Methods and SI Appendix, Fig. S2). The physically based surrogate model in Eq. 1 can reproduce the annual-mean surface albedo over the Arctic from CAM5 (Fig. 2A). Thus, the decreasing trend in annual-mean surface albedo can be attributed to reductions in snow cover fraction and sea ice fraction (Materials and Methods). The surrogate model gives a surface albedo trend of −0.94% decade⁻¹ (Fig. 2A), among which 70% (i.e., −0.66% decade⁻¹) is attributed to snow cover fraction change, with 38% over land (−0.36% decade⁻¹ in Fig. 2B) and 32% over sea ice (−0.3% decade⁻¹ in Fig. 2C), while the sea ice fraction retreat explains the remaining 30% (−0.26% decade⁻¹ in Fig. 2D). Our results appear inconsistent with a recent observational study (19) that suggests a more important impact of sea ice than snow cover on Arctic albedo reduction. This is because that study attributes all changes over ocean to the sea ice. Here we identify a nearly equally important role of snow cover fraction changes over sea ice compared to sea ice cover changes themselves and compared to terrestrial snow cover changes. Given the importance of surface air temperature (SAT) and snowfall to snow cover fraction, we turn next to an assessment of trends in SAT and precipitation/snowfall over the Arctic.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Surface albedo trend and attribution. (A) Time evolution of area- and energy-weighted average of annual-mean surface albedo over the Arctic during 1980 to 2014 from CAM5 (red) as well as using a surrogate model (black). The Arctic albedos derived from these 2 methods are highly correlated (R = 0.99, P < 0.01). (B) Contributions to surface albedo trend from SCF change over Arctic land. (C) Same as B but over Arctic ocean. (D) Same as C but for contributions from sea ice fraction change. In all panels, the solid line is a linear least-squares fit, of which the slope represents the trend in units of % decade⁻¹ (marked in each panel). All of the regression coefficients are statistically significant at the 99% confidence level.

Attribution of Snow Cover Fraction Changes.

SI Appendix, Figs. S3 and S4 illustrate the trends of anomalous area-averaged annual-mean Arctic SAT and total precipitation relative to the 1980 to 2014 mean. The CAM5 precipitation rate shows an increasing trend (6.93 mm decade⁻¹) as the Arctic warms (0.43 K decade⁻¹ for SAT). However, the area-averaged annual-mean snowfall exhibits a decreasing trend from 1980 to 2014 (SI Appendix, Fig. S5A). The results of 3 different reanalysis products show similar increasing trends in SAT and total precipitation (SI Appendix, Figs. S3 and S4) as well as decreasing trends in snowfall (SI Appendix, Fig. S5). The CAM5 trends are all well within the ranges defined by the reanalysis results. The trend of 6.93 mm decade⁻¹ in the Arctic total precipitation from CAM5 (SI Appendix, Fig. S4A) also agrees well with the trend of 7.13 mm decade⁻¹ from the GPCP (Global Precipitation Climatology Project) observations (SI Appendix, Fig. S4E).

We have identified the reduction in snow cover fraction as the primary contributor to the decreasing trend in annual-mean surface albedo (Fig. 2). Here, we further show what determines the reduction in snow cover fraction. A recent study (41) has suggested that the midlatitude land SAT can explain ∼50% of the variance in Northern Hemisphere land snow cover during the spring of 1922 to 2010. Physically the variation in snow cover is closely related to changes in air temperature and snowfall at the surface. The former affects snow melting while the latter is the main source of snow cover. However, there is no simple physically based equation to directly relate the 3 quantities, so we derive their relationship based on a statistical analysis. To maintain the consistency with Fig. 2, the snow cover area (SCA) fraction is defined as the ratio of SCA to the entire Arctic surface area with a fixed sea ice fraction. Thus, the SCA is calculated as the following summation over grid cells within the Arctic: SCA=∑SCF⋅(fland+fice¯)⋅A, where A represents grid cell area and fice¯ is monthly mean climatology of fice. The anomalies of annual-mean SCA fraction are shown in Fig. 3A (red circles) with a trend of −0.95% decade⁻¹ (red line). Given the strong and distinct seasonal cycle of SCA fraction (SI Appendix, Fig. S6), we define September to March as the snow-accumulation season and April to August as the snow-melting season. To attribute the trend of annual-mean SCA fraction, a multiple linear regression analysis is performed using the anomaly of annual-mean SCA fraction (denoted as F) as the dependent variable and anomalies of seasonal-mean snowfall and SAT as independents. The regression equation is obtained as F=0.019⋅Sa+0.024⋅Sm−0.258⋅Ta−1.089⋅Tm−5⋅10−6, where Sa,Sm,Ta, and Tm are the anomalies of accumulation-season mean and melting-season mean of snowfall and SAT over the Arctic, respectively. The anomalies of Sa,Sm,Ta, and Tm and their trends are shown in Fig. 3 B, C, D, and E, respectively. Based on the regression equation, we obtain the dependent F and its trend (−0.94% decade⁻¹), as shown in Fig. 3A (black circles and line). F has a strong correlation (R = 0.94) with the model-simulated SCA fraction, and their trends are almost identical, suggesting that the variation in SCA fraction can be fully explained by SAT and snowfall. Furthermore, the contributions of 7, 24, 10, and 59% to the trend in F by Sa,Sm,Ta, and Tm, respectively, are quantified based on the corresponding regression coefficients and their trends. Therefore, the decreasing trend of SCA fraction is largely determined by the increase of SAT (69%) and decrease in snowfall (31%) over the Arctic, especially during the snow-melting season.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Trend and attribution of snow cover area fraction. (A) Anomalies of annual-mean Arctic SCA fraction relative to 1980 to 2014 climatology with fixed sea ice fraction from CAM5 (red) and multiregression method (black). The 2 SCA fractions are highly correlated (R = 0.94, P < 0.01). (B) Anomalies of area-weighted average of Arctic mean snowfall during the accumulation season (September to March). (C) Same as B, but for the melting season (April to August). (D) Anomalies of area-weighted average of Arctic mean SAT during the accumulation season. (E) Same as D, but for the melting season. In all panels, the solid line represents a linear least-squares fit. The linear trend with units is marked for each panel. Except for the regression coefficient in panel B where P value is 0.17, all of the regression coefficients are statistically significant at the 99% confidence level.

Role of Absorbing Soot Deposition.

Impurities such as soot in snow can significantly reduce snow albedo due to the high albedo of pristine snow at visible wavelengths (27). Some modeling studies have suggested that one of the reasons for the Arctic warming is that soot strongly absorbs solar radiation in the atmosphere and snowpack (42). However, atmospheric soot concentrations and deposition fluxes have been declining in the Arctic since the 1980s, as shown at remote surface stations (43⇓–45) and in ice core data (46). Based on the CAM5 simulation, we are able to calculate the trend in deposition flux and soot-in-snow concentration/forcing. Note that soot-in-snow results are based on the SNICAR (SNow, ICe, and Aerosol Radiative) model (47), which is coupled to CAM5 and embedded in the land component of the Community Earth System Model. Compared to the observed soot concentration in surface snow (30), the modeled soot RMB is about 27.6% (SI Appendix, Fig. S7B). This indicates that the model likely overestimates the effect of soot in the Arctic snow/ice. The evaluation of the trend in concentration in surface air is described in SI Appendix, Supplementary Text. There was a decreasing trend in area-averaged annual-mean concentration over Arctic land in the CAM5 simulation over the period 1980 to 2014 (SI Appendix, Fig. S7C).

How does the change in soot-in-snow forcing compare to the change in radiative flux due to surface albedo reduction? As expected from the decreased snow cover and surface albedo, the area-averaged annual-mean solar flux absorbed over the Arctic land surface increased from 1980 to 2014 (SI Appendix, Fig. S9B). In contrast, the snow forcing over Arctic land from soot decreased during this period (SI Appendix, Fig. S9A). In addition, the ratio of the soot-in-snow forcing to net absorbed solar flux at surface is less than 1% (SI Appendix, Fig. S9C). The total decrease in the positive forcing from 1980 to 2014, estimated using the trend multiplied by the time interval, is 0.17 W m⁻², which is overwhelmed by the overall increase of 3.5 W m⁻² in net surface shortwave flux over Arctic land. While the spatial distribution of annual-mean soot deposition trends (Fig. 4A) shows a weak increase in northern Canada, Alaska, and northeastern Russia, strong and significant decreasing trends appear in the regions from the east of Greenland through northern Europe to Siberia, especially over the land regions. Overall, the trend of area-averaged annual-mean soot deposition over the Arctic (Fig. 4B) is negative, with an abrupt decline during 1991 to 1993, which is related to the dissolution of the former Soviet Union (45). Our analysis thus does not support the argument that deposition of soot on snow caused the decline of Arctic snow and ice cover, and consequently the Arctic amplification since the 1980s.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Soot deposition trend. (A) Spatial distribution of linear trends (mg m⁻²⋅decade⁻¹) in annual-mean soot deposition during 1980 to 2014. Stippling indicates statistical significance at the 95% confidence level using Student’s t test. (B) Area-averaged annual-mean soot deposition (mg m⁻²) over the Arctic from 1980 to 2014. The black line is a linear least-square regression and the trend is statistically significant at the 99% confidence level.