Projections of central Arctic summer sea surface temperatures in CMIP6

This study uses data from CMIP6 model outputs obtained from the CMIP6 data archive (https://pcmdi.llnl.gov/CMIP6/) as listed in table B1 (Eyring et al 2016). The monthly mean data from the historical and future period using the shared socioeconomic pathways (SSPs) 2–4.5 and 5–8.5 simulations from these CMIP6 models are analysed. In order to assess Arctic sea surface temperature (SST) changes in response to future scenarios of global greenhouse gas emissions, we use 30 CMIP6 models. Out of total 55 CMIP6 models, 30 models provide all the necessary variables for our study: SST, sea ice concentration (SIC), sea ice thickness (SIT) and surface heat fluxes (hfls, hfss, rsds, rsus, rlds, rlus) for the calculation of net surface heat flux (NSHF). In this study, we focus on the last 30 years in the historical (1985–2014) and SSP scenario simulations (2071–2100).

The Arctic SIE is defined as the total area of grid cells, where SIC is equal or greater than 15%. Until now, the term 'ice-free Arctic' has been defined as September SIE less than 1.0 million km² (Screen and Williamson 2017). In this study, we emphasize the need for a more detailed category for September SIE. Specifically, we divide 30 climate models into three groups: (1) complete ice-free models (13 models), in which 2071–2100 average September SIE is less than 0.2 million km², (2) moderate ice-free models (9 models), in which 2071–2100 average SIE is less than 1.0 but is larger than 0.2 million km², and (3) no ice-free models (8 models), in which 2071–2100 average SIE is larger than 1.0 million km².

This study focuses on a moderate climate change scenario, SSP2-4.5, in which summer Arctic SIEs simulated by CMIP6 models are widely different among the models (Kageyama et al 2021, Keen et al 2021, Long et al 2021). In SSP1-2.6 scenario, no models simulate the ice-free Arctic, whereas almost all the models simulate ice-free Arctic in SSP5-8.5 scenario (Årthun et al 2021). Figure 1 shows Arctic September SIE of CMIP6 models for historical and SSP2-4.5 scenario simulations. Consistent with previous studies (Holland et al 2006, Singarayer et al 2006, Overland and Wang 2007, Stroeve et al 2007, Ravindran et al 2021), models simulate long-term declining trend of SIE in response to increasing greenhouse gas concentrations, although large interannual variabilities are superimposed on the trend. Out of 30 models, 22 models simulate less than 1 million km² of September SIE by the end of the 21st century. Ice-free Arctic is not defined as a complete ice-free condition in September but is defined as September SIE dropping below 1 million km². This is because sea ice of coastal Greenland and Queen Elizabeth Islands in Canada, where land-fast sea ice is relatively thick, is probably not completely melted under global warming (Agnew et al 2008, Howell et al 2013).

Zoom In Zoom Out Reset image size

However, SIC in these regions becomes nearly zero in several CMIP6 models by the end of the 21st century (see figure 2(l)). In this study, we further divide ice-free Arctic into moderate ice-free (September SIE less than 1.0 but more than 0.2 million km²) and complete ice-free (September SIE less than 0.2 million km²) conditions. Models simulating complete ice-free Arctic by the late 21st century (2071–2100 average) show the most rapid declining trend from 2015 to 2050 (red line in figure 1(a)), whereas models simulating moderate ice-free Arctic show a relatively moderate declining trend (green line in figure 1(a)). Several models do not simulate ice-free Arctic by the late 21st century and the declining trends of September SIE of these models are relatively slow (blue line in figure 1(a)).

Zoom In Zoom Out Reset image size

The September SIE is closely tied to the central Arctic SST in August–September (Aug-Sep), during which SST reaches its maximum. Figure 1(b) shows steadily increasing trends of Aug-Sep SSTs averaged over the central Arctic (74° N–90° N) in response to future sea ice loss. There is a clear difference in the SST warming trend between the models simulating the complete ice-free and the moderate ice-free summer Arctic. The central Arctic SSTs in complete ice-free models, on average, increase up to 5.2 °C by 2100 (red line in figure 1(b)) and up to 8 °C–9 °C in several models. In the moderate ice-free models, the central Arctic SST warming, on average, is ∼3 °C (green line in figure 1(b)), which is about 2 °C lower than that of the complete ice-free models. In the no ice-free models, in which ice-free Arctic is not simulated by 2100, the central Arctic SST warming is generally within ∼1.5 °C (blue line in figure 1(b)).

The Arctic Ocean SST covered by sea ice is close to freezing temperature, around −1.8 °C (Thomas and Dieckmann 2002, Chin et al 2017). In the historical simulation, most of the Arctic Ocean is still covered by sea ice in September (figure 2(c)), where SSTs are close to freezing temperature (blue colour in figure 2(b)). Thus, the annual (from January to December) SST distribution averaged over the central Arctic Ocean is generally confined to below −1.0 °C (figure 2(a)). The annual SST distribution averaged over the central Arctic, which is confined to around −1.5 °–0.5 °C in historical simulation (figure 2(a)), becomes a bimodal distribution by the late 21st century in the complete ice-free models (figure 2(j)). The first peak of the annual SST distribution is around −1.0 °C–0 °C and the second peak is around 3.0 °C–4.0 °C (figure 2(j)). This bimodal SST distribution is indicative of a rapid transition from freezing winter temperature to rapid Aug-Sep warming.

This rapid seasonal SST transition in the complete ice-free models is likely associated with absence of the perennial sea ice cover. In moderate ice-free models, part of perennial sea ice remains over northern coasts of Greenland and Queen Elizabeth Islands (figure 2(i)), where SSTs are close to freezing temperature (figure 2(h)). In complete ice-free models, in which almost no sea ice remains in the central Arctic Ocean (figure 2(l)), the Aug-Sep SSTs are far above the freezing temperature, mostly above 0 °C (figure 2(k)).

In several climate models, such as CanESM5, EC-Earth3, UKESM1.0-LL, the annual SST distributions in the central Arctic exhibit distinct bimodal distributions and the annual maximum SST increases up to 5 °C–6 °C in SSP2-4.5 scenario (figure A1). These bimodal SST distributions become even more distinct in SSP5-8.5 scenario (figure A2).

Would the September SIE be a good indicator of the degree of Aug-Sep Arctic sea surface warming? the inter-model relationship between the September SIE and the Aug-Sep SST averaged over the central Arctic shows a high correlation (r = −0.91) over the historical era, from 1985 to 2014 (figure 3(a)). This is because the September SIE is dynamically tied to NSHF, which is a controlling factor of sea ice and SST. NSHF can be written as:

$$\begin{align}{\text{NSHF}} &= \left( {S{W^ \downarrow } - {\text{ }}S{W^ \uparrow }} \right) + \left( {L{W^ \downarrow } - {\text{ }}L{W^ \uparrow }} \right) \nonumber\\ &\quad - \left( {SH{F^ \uparrow } + LH{F^ \uparrow }} \right)\end{align} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

where $S{W^ \downarrow }$ and $S{W^ \uparrow }$ are downward and upward surface shortwave radiations, whereas $L{W^ \downarrow }$ and $L{W^ \uparrow }$ indicate downward and upward surface longwave radiations, respectively. $SH{F^ \uparrow }$ and $LH{F^ \uparrow }$ indicate surface sensible and latent heat fluxes, respectively. Downward surface heat fluxes, implying absorption of heat from atmosphere into ocean, are defined as positive values.

The temperature tendency of the ocean surface mixed layer can be explained by air-sea heat exchange and both vertical and horizontal advection (Hu et al 2020, Sharma et al 2023). Assuming that the vertical and horizontal heat advection averaged over the central Arctic is relatively small, the time-varying mixed layer temperature can be expressed as:

$$\begin{equation}\frac{{\partial {T_{\text{m}}}}}{{\partial t}} \approx \frac{{{\text{NSHF}}}}{{{\rho _{\text{w}}}{c_{\text{p}}}{h_{\text{m}}}}}\end{equation} \tag{ 2 }$$

where T_m is mixed-layer temperature, ρ_w and c_p denote density and specific heat of sea water, respectively. h_m represents the mixed layer depth (MLD). Figure 3(b) shows the relationship between spring-early summer (from May to July) NSHF and Aug-Sep SSTs averaged over the central Arctic Ocean in historical simulation. The correlation coefficient is quite high, r = 0.88, verifying that the seasonal SST maximum is constrained by spring-early summer NSHF.

In SSP2-4.5 simulation, however, the multi-model relationship between the September SIE and the Aug-Sep SSTs is nonlinear and the correlation coefficient (r = −0.67) is relatively small (figure 3(c)). The central Arctic SSTs rapidly increase in the case when the September SIE is less than 0.2 million km², which is consistent with figures 1(b) and 2, showing a more pronounced Aug-Sep sea surface warming in the complete ice-free models than in the moderate ice-free models. Moreover, the degree of sea surface warming is widely varying among the complete ice-free models (SIE < 0.2 million km²), as shown in figure 3(c), implying that the September SIE is not a good indicator of the Aug-Sep sea surface warming. Instead, the relationship between the spring-early summer NSHF and the Aug-Sep SST is linear (figure 3(d)) and the correlation coefficient is quite high, r = 0.86.

To better understand the process causing pronounced sea surface warming appearing in several climate models, we divided the complete ice-free models into two groups: one with the Aug-Sep SST higher than 4 °C (warm model group, red dots in figure 4(a)) and the other one with the Aug-Sep SST less than 4 °C (cold model group, blue dots in figure 4(a)). While both groups simulate complete ice-free Arctic (SIE < 0.2 million km²) in September, they show different seasonal cycle of SIE (figure 4(b)). The seasonal SIE of the warm model group is about 30%–70% smaller than that of the cold model group from winter to spring.

Zoom In Zoom Out Reset image size

In response to reduced sea ice cover, the seasonal cycle of NSHF is amplified in the warm model group (figure 4(c)). In the warm model group, more heat is absorbed into the central Arctic Ocean from May to July, whereas more heat is released into the atmosphere from November to May. Despite increased heat loss to atmosphere, the increased ocean-to-ice heat fluxes in response to global warming may suppress sea ice growth (Cornish et al 2022). The average difference of downward heat flux between the warm and cold model groups is about 30 W m⁻² in May and June, and 3–4 W m⁻² in July (figure 4(d)). Observational studies suggest that the Arctic Ocean MLDs are around 35 meters in May and 25 meters in June (Peralta-Ferriz and Woodgate 2015, Zhai and Li 2023). Assuming that the May-June average MLD of the Arctic Ocean is 30 meters, equation (2) indicates that an anomalous downward NSHF of 30 W m⁻² for two consecutive months (May-June) can lead to ∼1.5 °C SST difference in summer $\left( {30\,W\,{m^{ - 2}}/{\rho _{\text{w}}}{c_{\text{p}}}30\,m \approx 1.5{\,^ \circ }C/2\,{\text{month}}} \right)$. The difference of May–July NSHF between the warmest model (CMCC-CM2-SR5) and the coldest model (CESM2-WACCM) among the models simulating the ice-free summer Arctic is ∼70 W m⁻² in May and June, leading to ∼4 °C SST difference in Aug-Sep.

Further analysis indicates that the increased shortwave absorption explains the majority of the increased downward surface heat flux from May to July (green line in figure 4(d)), highlighting the effectiveness of spring ice-albedo feedback associated with preceding winter SIT. In April, the increased downward shortwave radiation associated with less sea ice cover in the warm models is largely compensated by increased turbulent heat fluxes (figure 4(d)). These results suggest that the September SIE is not the best indicator of the Aug-Sep sea surface warming in the central Arctic Ocean. Instead, the degree of SST increase in future climate is more associated with sea ice cover in spring and early summer during which sunlight is strong and sea ice cover can effectively modulate net shortwave radiation (figure 3(d)). Further analysis shows that the interannual correlation between SIE and NSHF in spring–early summer (May–July) is quite high, with r = −0.9 (see figure A3). This result underscores the importance of spring–early summer SIE in modulating the NSHF.

Because SIT anomalies can strongly affect the seasonal SIE anomalies for many months ahead (Blanchard-Wrigglesworth et al 2011, Ordoñez et al 2018, Bushuk et al 2020), the earlier sea ice retreat may imply relatively thinner sea ice in winter. Figure 5(a) shows the relationship between future winter (from December to March) SIT and NSHF averaged in May–July, in which both SSP2-4.5 and SSP5-8.5 scenario simulations (black vs white dots in figure 5(a)) are lumped together. Here, both SSP2-4.5 and SSP5-8.5 scenario simulations are added to increase the statistical degree of freedom (many of CMIP6 models do not provide SIT as outputs). The inter-model relationship between winter SIT and the May–July NSHF is high: the correlation coefficient is −0.71, verifying that the spring-early summer NSHF is closely tied to winter SIT. In SSP2-4.5, in which 26 models provide SIT as an output variable, the correlation coefficients between winter SIT and spring-early summer NSHF is −0.73.

Zoom In Zoom Out Reset image size

Because the reduced sea ice cover in spring and early-summer is accompanied by Arctic sea surface warming, the inter-model relationship between winter SIT and the central Arctic SST in Aug-Sep is also linear (figure 5(b)): the correlation coefficient is −0.77. In each scenario simulation, the correlation coefficients between winter SIT and Aug-Sep SST are −0.78 and −0.62 for SSP2-4.5 and SSP5-8.5, respectively. These results are consistent with previous studies indicating that winter SIT can have a significant impact on the subsequent spring and summer Arctic sea ice cover. Specifically, late winter and early spring SIT initializations have been shown to substantially improve seasonal Arctic sea ice cover forecast in ocean-sea ice models, such as HadGEM1.2 (Day et al 2014), CanCM3 (Dirkson et al 2017), and NEMO3.4 (Balan-Sarojini et al 2021). Observational studies also suggest that thinner winter Arctic SIT can lead to a reduction in spring and early summer sea ice cover by lowering surface ice albedo (Nasonova et al 2018, Petty et al 2023).