Arctic amplification is caused by sea-ice loss under increasing CO2

Abstract

Warming in the Arctic has been much faster than the rest of the world in both observations and model simulations, a phenomenon known as the Arctic amplification (AA) whose cause is still under debate. By analyzing data and model simulations, here we show that large AA occurs only from October to April and only over areas with significant sea-ice loss. AA largely disappears when Arctic sea ice is fixed or melts away. Periods with larger AA are associated with larger sea-ice loss, and models with bigger sea-ice loss produce larger AA. Increased outgoing longwave radiation and heat fluxes from the newly opened waters cause AA, whereas all other processes can only indirectly contribute to AA by melting sea-ice. We conclude that sea-ice loss is necessary for the existence of large AA and that models need to simulate Arctic sea ice realistically in order to correctly simulate Arctic warming under increasing CO₂.

Fig. 1

Seasonality of the historical (1979–2016) trends in Arctic (67°–90°N) sea-ice cover (SIC), Arctic amplification (AA), and Arctic energy fluxes. a from ERA-Interim reanalysis data and b from the ensemble mean of historical (for 1979–2005) and RCP8.5 (for 2006–2016) simulations averaged over 38 CMIP5 models. The SIC trend (gray bars) is in 10⁵ km²/decade; the AA (black line) is defined as the ratio of the surface air temperature trends between the Arctic and the globe; the surface net shortwave (red, positive downward), sensible plus latent heat (blue, positive upward) and upward longwave (magenta) flux trends are in W/m²/decade

Fig. 2

Spatial distributions of the linear trends during 1979–2016. For November–December mean surface air temperature (red contours, K/decade), sea-ice concentration (SIC, color shading, %/decade), and surface turbulent (sensible + latent) heat fluxes (yellow contours, W/m²/decade, positive upward) based on a the ERA-Interim reanalysis data and b the ensemble mean of historical (for 1979–2005) and RCP8.5 (for 2006–2016) simulations averaged over 38 CMIP5 models. The SIC trends are similar to those based on NOAA satellite data from https://sidads.colorado.edu/DATASETS/NOAA/G02202_V3. Spatial pattern correlations: Trend pattern correlations: r (SIC,Tas) = −0.61, r (SIC,LH + SH) = −0.68, r (Tas,LH + SH) = 0.56 in a; and r (SIC,Tas) = −0.40, r (SIC,LH + SH) = −0.70, r (Tas,LH + SH) = 0.65 in b. The upward longwave radiation trend (not shown) is highly correlated with the air temperature trend (r ≥ 0.96). These correlations have a p-value well below 0.01

Fig. 3

Centennial changes as a function of month from CMIP5 models. For Arctic (67°−90°N) sea-ice concerntraion (SIC, in % of Arctic area, shading, multiplied by −1), Arctic-to-global ratio of the Tas change (AA, black line, multiplied by 10 in order to use the left y-axis), and Arctic surface energy fluxes (in W m⁻²). a 2070–2099 minus 1970–1999, b 2170–2199 minus 2070–2099, and c 2270–2299 minus 2170–2199 under the historical and RCP85 scenarios from the ensemble mean of nine model runs from the nine CMIP5 models. Net SW = net shortwave radiation (positive downard), upward LW = upward longwave radiation, SH = sensible heat, LH = latent heat. d Time-dependent warming and sea-ice loss from CMIP5 models. Time series of the difference between the 20 year periods separated by the plotted year in annual Arctic (67°−90°N, red solid) and global-mean (red dashed) surface air temperature (Tas), annual Arctic SIC (blue), and the difference between the red solid and red dashed lines (black) based on the ensemble mean of nine simulations from nine CMIP5 models. The correlation coefficient between the lines are: r (SIC, Tas_Arctic) = −0.97, r (sic,Tas_global) = −0.90, r (sic,Tas_diff) = −0.95, and r (sic, AA) = −0.80, where AA = the ratio of the Arctic to global Tas change (data before 2000 were not used for AA). These correlations have a p-value well below 0.01

Fig. 4

Dependence of Arctic warming and amplification on sea-ice loss among 38 CMIP5 models. Scatter plot of 2070–2099 minus 1970–1999 difference under the RCP85 scenario between annual Arctic SIC loss and Arctic surface warming (red), Arctic-minus-global warming difference (blue), or the Arctic-to-global warming ratio (i.e., the Arctic amplification, or AA, black). Each dot is for one CMIP5 model. Correlation coefficients: r(SIC, dTas_Arctic) = 0.87 (p = 0.00), r(SIC, Tas_diff) = 0.84 (=0.00), and r (SIC, AA) = 0.54 (p = 0.01)

Fig. 5

Centennial changes from CMIP5 models. a from 1970–1999 to 2070–2099 and b from 2070–2099 to 2170–2199. Shown are changes in December sea-ice concentration (SIC, %, color shading), air temperature (Tas, °C, red contours, interval = 2), and latent and sensible heat fluxes (LH + SH, W m⁻², yellow contours, interval = 10). Dashed contours are for negative values. Based on the ensemble mean of nine simulations from nine CMIP5 models under the historical and RCP85 scenarios. The spatial pattern correlations are: r (SIC, Tas) = −0.79, r (SIC, LH + SH) = −0.68, r (Tas, LH + SH) = 0.62 in a, and r(SIC, Tas) = −0.85, r (SIC,LH + SH) = −0.65, r (Tas, LH + SH) = 0.60 in b. Surface net energy flux change (not shown) is similarly correlated with the SIC and Tas changes, while the upward LW flux change (not shown) is highly correlated (r≈0.90) with the Tas change. These correlations have a p-value well below 0.01

Fig. 6

Time series of CESM1-simulated changes in surface air temperature (Tas) and sea-ice over the Arctic (67°-90°N) and globe. a Standard 1% CO₂ run, b FixedIce run, and c their difference (panel a minus panel b). Shown are the annual (solid red), November–December (magenta), and June–July (green) mean Tas, and annual Arctic sea-ice concentration (blue, right y-axis, increase downward), together with the Arctic-minus-global annual Tas difference (black). The change is relative to the control-run climatology and five-year averaging is applied. Note that global Tas changes for November–December and June–July (not shown) are very similar to the annual change

Fig. 7

Centennial changes as a function of month from CESM1 standard simulation with a 1%-per-year CO₂ increase. Same as Fig. 3, except for the change during the a first (year 61–80 mean minus control climatology), b second (year 131–150 mean minus year 61–80 mean), and c third (year 201–220 mean minus year 131–150 mean) CO₂ doubling from the standard simulation with a 1% CO₂ increase per year using the CESM1. d Time-dependent warming and sea-ice loss from CESM1 simulation with a 1%-per-year CO₂ increase. Same as Fig. 3d except for the CESM1 1% CO₂ run

Fig. 8

Centennial changes as a function of month from CESM1 special simulation with fixed sea-ice cover. Same as Fig. 7 except for the 1%-per-year CO₂ run with fixed sea-ice in calculating the ice–atmosphere and ice–ocean fluxes (FixedIce run). Note the upward longwave radiation change (LW_up) is plotted on the right y-axis in a–c here. d Time-dependent warming and sea-ice loss from CESM1 simulation with fixed sea-ice cover. Same as Fig. 3d except for the CESM1 FixedIce run

Fig. 9

Height–latitude distributions of zonal-mean temperature change from the CESM simulations. a-c The standard 1% CO₂ run. d-f FixedIce run. The temperature change is relative to the control-run climatology and is around the time of the second doubling (i.e., years 131–150) of the pre-industrial CO₂ level. Top row: for December–January–February (DJF). Middle row: for June–July–August (JJA). Bottom row: for annual-mean. The change patterns are similar around the 1^st and 3^rd doubling of the pre-industrial CO₂

Fig. 10

Mean changes (relative to the control-run climatology) for year 131–150. a from the 1% CO₂ run and b from the FixedIce run. Shown are changes in December sea-ice concentration (SIC, %, color shading), surface air temperature (Tas. °C, red contours, interval = 2), and surface latent plus sensible heat flux ((LH + SH, W m⁻², yellow contours, interval = 10, positive upward). Dashed contours are for negative values. The spatial pattern correlations are: r (SIC, Tas) = −0.20, r (SIC, LH + SH) = −0.75, and r (Tas, LH + SH) = 0.55 in a, and r (SIC, Tas) = 0.12, r (SIC,LH + SH) = −0.45, and r (Tas, LH + SH) = −0.28 in b. These correlations have a P-value below 0.01