Global temperature response and climate sensitivity
Annual mean high latitude warming (4 × CO2 − 1 × CO2) in simulations with different sea ice treatments is shown in Fig. 1a–c. Arctic warming is most strongly pronounced in the active ice simulations (1a). In comparison, in the absence of sea ice response, in zero and prescribed ice simulations, high northern latitude warming is up to 10 K smaller (Figs. 1a–c, 2a, b). In the absence of sea ice changes, the strongest warming is over the land and not over the ocean as is the case in the active ice simulations.
Also shown in Fig. 1, are 0.15 and 0.9 autumn (SON) ice fractions from the 4 × CO2 simulations (blue and white contour lines, respectively). In the active ice case, the Arctic is seasonally ice free (Fig. 1a). In the zero ice case, there is no sea ice present year round due to the experimental setup (Fig. 1b). In the prescribed ice simulation, the preindustrial sea ice cover is maintained (Fig. 1c).
Zonal temperature anomalies due to varying CO2 levels from 1 × CO2 to 2 ×, 4 ×, 6 × and 8 × CO2 (Fig. 1d–f) further highlight the striking difference in temperature responses due to the different sea ice treatments. Whereas the most pronounced difference between the active ice and simulations with disabled ice response is in the high latitudes, low latitude warming also appears to be affected by the sea ice decline. For example, in the case of CO2 quadrupling, ice response leads to an additional 1 to 2 K tropical temperature increase in active ice simulations relative to the zero and prescribed ice simulations (Figs. 1d–f, 2c, d).
Globally, mean temperature response to increasing CO2 concentrations is larger in the active ice than in the zero and prescribed ice simulations (Fig. 3a, b). For the case of CO2 quadrupling, global mean surface air temperature changes equal 6.59 ± 0.03, 4.86 ± 0.02 and 4.16 ± 0.02 K in active, zero and prescribed ice simulations, respectively. Compared to the active ice simulations, zero ice simulations show smaller global temperature increase at low CO2 concentrations (Fig. 3b). However, as the ice cover in active ice simulations disappears (with CO2 increase), the difference between temperature response in active and zero ice simulations steadily decreases. In 2 × CO2 simulations, temperature response with the zero ice treatment is 29 % smaller than with the active ice treatment, while in 8 × CO2 simulations temperature response is 21 % smaller with zero ice relative to active ice treatment (Fig. 3c, red circles).
In simulations with the sea ice cover prescribed to pre-industrial values, the overall warming at any CO2 level is ~37 % smaller than in the active ice simulations (Fig. 3c, blue circles). This near constancy of response with CO2 level in the prescribed ice simulations is not particular to the 1 × CO2 sea ice extent. Prescribing the 2 × CO2 sea ice cover in the 4 ×, 6 × and 8 × CO2 simulations yields 29 % less global mean temperature increase independently of CO2 concentration (not shown).
The main cause of the different global temperature responses in zero and prescribed ice simulations is the surface energy imbalance imposed in the prescribed ice simulations while maintaining a constant sea ice extent. We illustrate this by considering the global annual mean top-of-atmosphere net flux anomalies (ΔN
TOA) plotted against the global mean surface air temperature change (ΔT) for the case of CO2 quadrupling (Fig. 4). This type of plot provides information on the initial top-of-atmosphere net energy imbalance due to CO2 quadrupling (y-axis intercept), the ‘climate feedback parameter’ λ (based on the value of the linear regression coefficient) and the equilibrium surface air temperature change (Gregory 2004). In Fig. 4 we consider the transient period only—plotted values span from year 1 till year 30 of the model simulation. In order to facilitate more accurate parameter determination, two additional CO2 quadrupling simulations of a 10 year duration with slightly altered initial conditions are added for each set of simulations (active, zero and prescribed ice).
The atmosphere is in equilibrium when the top-of-the-atmosphere net energy budget equals the surface net energy budget. As in the active and zero ice simulations there is no imposed surface energy imbalance, global steady state temperature responses can be estimated from an intercept of the regression line and the ΔN
TOA = 0 W/m2 line. In the prescribed ice 4 × CO2 simulation, the imposed surface energy imbalance (relative to 1 × CO2 prescribed ice simulation), ΔF
SURF, equals 1.24 ± 0.06 W/m2. This is the energy removed in order to maintain the 1 × CO2 sea ice extent in the 4 × CO2 climate. The global temperature response is thus defined as the intercept of the regression line and the ΔN
TOA = 1.24 W/m2 line (that equals this surface net imbalance) and not the ΔN
TOA = 0 W/m2 line. The intercept of the regression line and the ΔN
TOA = 0 W/m2 line would lead to an incorrect estimate of the global temperature response in the 4 × CO2 prescribed ice simulation (about 1 K higher).
The slopes of the regression lines in simulations with no sea ice response are similar and equal −1.33 ± 0.04 W/(m2 K) and −1.38 ± 0.08 W/(m2 K), for zero and prescribed ice simulations, respectively. Thus, the climate feedback parameters, λz and λp, are statistically indistinguishable between the prescribed and zero ice treatments. This further implies that the climate sensitivity (estimated as 1/λ) is ~25 % smaller in simulations with disabled sea ice feedbacks relative to the simulation with active ice treatment [λa = 1.00 ± 0.05 W/(m2 K)]. However, the temperature responses between zero and prescribed ice simulations are different, due to the imposed surface forcing in the prescribed ice simulations. The global mean temperature increase due to CO2 quadrupling in prescribed ice simulations is ~37 % smaller than the global mean temperature increase in the active ice simulation, while the same global mean temperature increase in the zero ice simulation is ~25 % of the corresponding active ice temperature increase (see Fig. 3b). Impacts of sea ice decline on global annual mean climate sensitivity and radiative forcing are discussed in greater detail in Caldeira and Cvijanovic (2014). However, due to a different sea ice prescription applied and different analytic approach used, values reported in Caldeira and Cvijanovic (2014) differ slightly from those presented here.
Finally, it is important to notice that different sea ice treatments have different control (1 × CO2) climates (Fig. 3a). At low CO2 levels, global mean surface air temperature in prescribed ice simulations is similar to the that in active ice simulation and both of these are cooler compared to the global mean temperature in the zero ice simulations. At higher CO2 concentrations, active ice and zero ice simulations have similar global mean temperatures that are warmer than the corresponding global mean temperatures in the prescribed ice simulations. We take into account these different background climates when discussing the impacts of sea ice decline on global warming.
Equator-to-pole temperature gradient and atmospheric heat transport changes
Larger warming in high latitudes compared to the low latitudes (Fig. 1d–f) results in a decreased equator-to-pole temperature gradient. The largest weakening of the equator-to-pole gradient with CO2 increase occurs in the active ice simulations. At low CO2 concentrations, equator-to-pole temperature gradient in active ice simulations resembles the one in prescribed ice simulations. As the CO2 concentrations increase, equator-to-pole temperature gradient in active ice simulations becomes similar to the gradient value from the zero ice simulations (not shown).
If we consider the ratio of surface warming over the area 60°–90°N and 0°–30°N, we observe that for any sea ice treatment this ratio is largely insensitive to the amount of global warming (Fig. 3d). In active ice simulations, Arctic warming is almost 3 times greater than tropical warming with the ratio changing from 2.91 ± 0.01 in the 2 × CO2 simulation to 2.98 ± 0.04 in the 8 × CO2 simulation. In the absence of sea ice changes, this ratio lies between 2.18 ± 0.08 and 2.24 ± 0.03 in zero ice and 1.95 ± 0.08 and 1.83 ± 0.03 in prescribed ice simulations, respectively. Thus, in the absence of sea ice response, both high and low latitude warming is weaker, with tropical changes being about half of the Arctic changes (in comparison to about one-third in the presence of a sea ice response).
Smaller changes in the high-to-low latitude temperature gradient in the absence of sea ice feedbacks may have an affect on the zonal flow and northward atmospheric heat transport (Jain et al. 1999; Cvijanovic et al. 2011; Karamperidou et al. 2012). Thus, we investigate if there is evidence that the sea ice loss is also leading to large-scale atmospheric circulation changes in our model simulations.
Atmospheric heat transport (AHT) anomalies (4 × CO2 − 1 × CO2) and its dry static energy and LH transport components are shown in Fig. 5a. These anomalies are calculated from the atmospheric energy and fresh water budgets, assuming a steady state with constant (long term) energy and moisture content in the atmospheric column, as described by Kay et al. (2012). This is a correct assumption for the multi-year averages used in our study.
In 4 × CO2 relative to the 1 × CO2 simulation, overall AHT increases for all sea ice treatments (Fig. 5a). This is a consequence of a large decrease in dry static energy (DSE) transport and smaller increase in LH transport (Fig. 5b). In the absence of sea ice response, midlatitude AHT changes are weaker than in the presence of sea ice changes. AHT decomposition, shown in Fig. 5b, indicates that this is mainly due to a weaker DSE transport increase, which ranges from largest in the active ice to smallest in the prescribed ice simulations. The relative magnitudes of DSE transport changes in active, zero and prescribed simulations are in accordance with the corresponding magnitudes of the equator-to-pole temperature gradient changes shown in Fig. 3d. In contrast, northward midlatitude LH transport response appears to be relatively insensitive of the sea ice changes.
In the tropics, AHT changes are largest in the active ice case. In accordance with the magnitude of their individual tropical AHT anomalies, tropical precipitation shifts are also most pronounced in the active ice simulations and weakest in the prescribed ice simulations (not shown). This is in line with the previous studies showing that the cross-equatorial AHT anomalies are associated with tropical precipitation shifts (Chiang and Bitz 2005; Kang et al. 2009; Cvijanovic and Chiang 2013).
Zonal wind responses and changes in extreme precipitation and temperature events
Annual mean 500 hPa zonal wind strength anomalies (4 × CO2 − 1 × CO2) obtained with different sea ice treatments are shown in Fig. 6a–c. In the active ice simulation, zonal westerly wind flow weakens over most of the northern midlatitudes (Fig. 6a). This is in agreement with other studies finding a weakening of the zonal westerly flow in response to enhanced Arctic warming and decreased mid- to high latitude geopotential thickness gradient (Francis et al. 2009; Overland and Wang 2010; Francis and Vavrus 2012). In contrast, in the prescribed and zero ice simulations, westerly flow increases in strength in the upper midlatitudes, especially over the North Atlantic and northern Europe (Fig. 6b, c). Similar responses are also detected in the upper level flow (300 hPa) (not shown). Analysis of seasonal means shows that different 500 hPa zonal wind responses observed in annual means are most pronounced in the Northern Hemisphere winter (DJF) and fall (SON). Figure 6, panels d–f, shows the 500 hPa zonal wind changes for the winter DJF season. Strengthening of the zonal wind flow is larger in the prescribed ice than in the zero ice simulations. The difference between the DJF zonal wind strength changes in simulations with and without sea ice response is illustrated in Fig. 7. Zonal wind strength anomalies in zero ice simulations relative to the active ice simulations (Fig. 7a) are smaller than the corresponding anomalies between the prescribed ice and active ice simulations (Fig. 7b). This is in accordance with the smaller difference in the amount of high latitude warming between the zero ice and active ice simulations (Fig. 2a) and the prescribed ice and active ice simulations (Fig. 2b), consistent with the thermal wind balance.
Different westerly wind responses to CO2 induced warming in the presence or absence of sea ice changes (Fig. 6) may have an impact on extreme weather development in the northern midlatitudes. Previous study by Francis and Vavrus (2012) suggested that the weakened zonal winds and slower progression of upper-level waves could lead to an increase in extreme weather events as a result of prolonged conditions. Similarly, Peings and Magnusdottir (2014) found that Arctic amplification and reduced midlatitude westerlies favor the increased intensity of cold extremes over certain midlatitude regions. Moreover, since the midlatitude dry static energy transport is maintained by eddies, different DSE transport responses in simulations with active and disabled sea ice response (see Sect. 3.2) may be another indicator of sea ice response affecting the atmospheric circulation patterns in the midlatitudes. We thus continue our analysis by comparing the strengths and frequencies of various extreme weather indices under different sea ice treatments. The indices considered are the maximum precipitation over a given period, the number of heavy precipitation days, the minimum temperature over a given period and the number of frost days (as described in Sillmann et al. 2013). These indices are evaluated from a total of 30 years of daily data. The analysis is limited to the winter (DJF) season as this is the season with the largest impact of sea ice treatment on zonal wind strengths. Values of all the indices are first estimated at a given grid-cell and then averaged over the “lower” (35°–45°N) and “upper” (45°–55°N) midlatitude bands.
We first consider the change in the intensity of the winter precipitation maxima (Pmax). For each DJF season in the 30 year period, we select the day with the largest amount of precipitation (Fig. 8). In all simulations, independent of sea ice treatment, we see an increase in the precipitation maxima with an increase in CO2 concentration (Fig. 8 panels a and c). At a given CO2 level, the differences between simulations with different sea ice treatment are not statistically significant. In order to account for different background states in simulations with different sea ice treatments, precipitation maxima are also plotted relative to the DJF northern hemispheric surface air temperature (Fig. 8 panels b and d). At a given temperature, prescribed ice simulations show higher winter precipitation maxima relative to the zero ice simulations (both in the lower and upper midlatitudes). In the active ice simulations, precipitation maxima tend to resemble the prescribed ice simulations at lower temperatures and zero ice simulations at higher temperatures. Overall, Pmax appears to be less sensitive to NH temperature increase in simulations featuring larger equator-to-pole gradient changes (active ice) than in the simulations featuring smaller equator-to-pole gradient changes (zero and prescribed ice).
Next we consider the impacts of sea ice response on the frequency of heavy DJF precipitation events (defined as the number of days within a season with total precipitation larger than 10 mm). This is illustrated in Fig. 9, showing the number of DJF heavy precipitation events relative to: the CO2 level (panels a and c) and the hemispheric mean DJF temperature (panels b and d). In the lower midlatitudes, at high CO2 concentrations, prescribed ice simulations show a lower number of heavy precipitation events compared to zero or active ice simulations (Fig. 9a). However, sea ice treatment appears not to affect the number of heavy precipitation events after taking into account different NH temperature responses in the active, zero and prescribed ice simulations (Fig. 9b). Thus, in the lower midlatitudes, the hemispheric temperature change appears to be a very good indicator of the number of heavy precipitation events for any sea ice treatment. As shown earlier, global (and hemispheric) temperatures are influenced by the sea ice changes; the relative impact of sea ice changes due to increased CO2 concentrations accounts for 21–37 % of the overall CO2 induced warming in our model simulations (Fig. 3c). A notable difference in warming between the simulations with and without sea ice responses is required in order to achieve a substantial (and statistically significant) difference in the number of heavy precipitation events. This is satisfied for large CO2 forcing (e.g., in Fig. 9a this is achieved at 8 × CO2 concentration).
In the upper midlatitudes, simulations with different sea ice treatments also show a statistically different number of heavy precipitation events only at high CO2 concentrations (Fig. 9c). At a given northern hemispheric DJF temperature, prescribed ice simulations have more heavy precipitation days than the zero ice simulations, while the number of heavy precipitation events in the active ice simulations resembles that in the prescribed ice simulations at lower temperatures and that in the zero ice simulations at higher temperatures (Fig. 9d). Similarly as before, we find the simulations featuring smaller equator-to-pole gradient change (prescribed and zero ice simulations) to be more sensitive to NH temperature increase than the simulations featuring larger equator-to-pole gradient change (active ice simulations).
Changes in winter temperature minima and the number of ice days are shown in Figs. 10 and 11, respectively. Winter temperature minima increases with CO2 concentration in all simulations, both in lower and upper midlatitudes, while different sea ice treatments result in different responses in Tmin. Winter temperature minima are larger in zero ice than in prescribed ice simulations, while the active ice simulations minima are located between these two (Fig. 10a, c). When considering Tmin relative to the corresponding northern hemispheric DJF temperature (Fig. 10b, d) differences between the sea ice treatments are mostly not present. Northern hemispheric DJF temperature thus appears to be a very good indicator of the grid-cell scale change in winter temperature minima. However, it is important to recall that hemispheric temperatures are affected by the sea ice treatment. Different responses in Tmin originating from the different sea ice treatments are eliminated when taking into account different NH temperature responses in active, zero and prescribed ice treatment.
The number of ice days in DJF season decreases with increasing CO2 concentration across the midlatitude sector (Fig. 11a, c). Prescribed ice simulations feature a larger number of ice days than the zero ice simulation, in agreement with the weaker global temperature increase in prescribed ice simulations. The number of ice days in the active ice simulations falls in between the zero ice and prescribed ice simulations. Considered at a given temperature, over the lower midlatitude sector prescribed ice simulations have a lower number of ice days than zero ice simulations (Fig. 11b). The decrease in the number of ice days with NH temperature is larger in prescribed and zero than in the active ice simulations. In the upper midlatitudes, simulations with different ice treatments show the same response to hemispheric temperature increase (Fig. 11d).