3.1 Surface warming response to changes in global SST or Arctic SIC
We first examine the annual-mean surface air temperature response to historical and future global SST (Fig. 2a, b) or Arctic SIC (Fig. 2c, d) changes shown in Fig. 1. The globe experiences relatively uniform warming in pdSST-pdSIC relative to piSST-pdSIC (Fig. 2a, referred to as historical warming) and in futSST-pdSIC relative to pdSST-pdSIC (Fig. 2b, referred to as future warming), with slightly greater magnitude in the future SST case (around 1.0–2.0 K) than the historical case (0.5-1.0 K). Thus, the SST perturbation runs show background global warming without noticeable AA. In contrast, reduced Arctic sea-ice leads to large warming over the Arctic with little temperature change south of ~ 60oN in both the historical and future perturbed SIC runs (Fig. 2b, d). Note that the local Arctic warming is larger for the future case (~ 3–5 K; Fig. 2d) than the historical case (~ 1–3 K; Fig. 2c) as the future sea-ice loss is larger (Fig. 1c-d) and that the largest historical warming (Fig. 2c) occurs over the Barents-Kara Seas region where there is large sea-ice loss (Fig. 1b).
The seasonal cycle of the surface air temperature changes averaged over the Arctic (Fig. 3a) and globe (Fig. 3b) shows different responses to global SST or Arctic SIC perturbations. Global SST perturbations produce small Arctic warming during historical (~ 0.5-1.0 K) and future (~ 1.0–2.0 K) periods for October-March and negligible summer warming (Fig. 3a). The global-mean surface temperature warms by ~ 0.8 K for the historical and ~ 1.2 K for the future SST cases, with little seasonal variation (Fig. 3b). Thus, there is small AA (< 0.8 K) during October-March while the summer Arctic warming is weaker than the global-mean warming in the SST perturbation experiments (Fig. 3c). In contrast, Arctic sea-ice loss produces large Arctic warming from October-January for the historical (~ 3 K) and future (~ 6–8 K) cases, with weak warming in summer (Fig. 3a). Note that the peak warming shifts from October in the historical case to November in the future case. The global-mean warming response to the SIC changes is weak throughout most of the year except the cold season (Fig. 3b), which is due to the large warming in the Arctic (Fig. 2c-d). As a result, AA is strong from October-January for the two perturbed SIC cases, especially for the future SIC case (up to 7 K), while the AA is weak (< 1 K) during the summer months (Fig. 3c).
3.2 Surface energy budget response to Global SST or local Arctic SIC changes
Increased upward surface energy fluxes over sea-ice retreat areas have been shown to drive large Arctic warming and AA in winter (Deser et al. 2010; Boeke and Taylor 2018; Taylor et al. 2018; Dai et al. 2019). In response to SST warming with fixed SIC, we find little change in the net surface energy flux, net surface SW, SH, and LH fluxes over the Arctic Ocean throughout the year (Fig. 4). The upward net surface LW flux decreases by ~ 1 W m− 2 for both the historical and future SST warming cases with fixed SIC (Fig. 4c). This represents a small increase in the downward LW radiation, likely due to increased water vapor and enhanced atmospheric energy convergence into the Arctic, rather than changes to surface conditions, as shown below. The suppressed Arctic surface warming and weak oceanic energy flux response to SST warming without SIC changes is consistent with Dai et al. (2019), who found similar results in model simulations with increasing CO2 concentrations and fixed Arctic sea-ice in flux calculations.
Arctic sea-ice loss greatly influences the magnitude and seasonal cycle of the Arctic oceanic heat flux. From May-August, oceanic absorption of energy increases by ~ 6–12 W m− 2 in response to historical and future SIC loss (Fig. 4f) while during October-March oceanic release of energy increases by ~ 12–18 W m− 2 (Fig. 4a). Most of the increased oceanic energy absorption from May-August is due to increased absorption of SW radiation (Fig. 4b), with negligible changes in net surface LW, SH, and LH fluxes (Fig. 4c-e) during summer. In contrast, net surface LW, SH, and LH fluxes are the main contributors to the enhanced cold-season oceanic energy release in response to Arctic sea-ice loss. Specifically, the SH (LH) flux contributes ~ 4–6 W m− 2 (~ 3–4 W m− 2) and ~ 8–10 W m− 2 (~ 8–9 W m− 2) in response to historical and future SIC loss, respectively (Fig. 4d, e). Further, the ocean surface emits ~ 1–2 W m− 2 (~ 1–3 W m− 2) more LW radiation to the atmosphere in autumn and winter in the historical (future) Arctic sea-ice loss runs (Fig. 4c). The large increases in upward surface energy fluxes in response to sea-ice loss play an important role in enhancing warming of the surface air and AA during winter (Fig. 3a).
3.3 Feedback seasonal cycles and warming contributions
The contrasting surface warming responses to global SST changes or local sea-ice loss greatly influence Arctic climate feedbacks and atmospheric energy convergence changes. Under global SST warming with fixed SIC, Arctic atmospheric energy convergence (Fig. 5f) and water vapor feedback (Fig. 5b) become important contributors to the Arctic TOA flux change. Specifically, atmospheric energy convergence into the Arctic responds similarly to historical and future SST warming, with increases of ~ 4–7 W m− 2 during the winter months and ~ 2–4 W m− 2 in summer (Fig. 5f). This suggests that without changes in sea ice, increased atmospheric energy transport becomes an important contributor to small cold season Arctic warming and AA (Fig. 3). Further, the magnitude and seasonal cycle of the water vapor feedback is similar between the historical and future SST cases, with maximum water vapor feedback (1.5–1.8 W m− 2) from May-August and minimum water vapor feedback (0.4–0.7 W m− 2) during October-March (Fig. 5b). This is expected as the warm-season Arctic would see larger water vapor increases due to its warmer mean air temperatures. Arctic surface albedo (Fig. 5a), lapse rate (Fig. 5c), and Planck (Fig. 5d) feedbacks weakly respond to SST increases without sea-ice loss. Lastly, we note that the net cloud feedback produces slight cooling (-1.5~-2.0 W m− 2) in response to SST increases for June-August (Fig. 5e).
In response to sea-ice loss, Arctic surface albedo feedback increases by ~ 7–8 W m− 2 and ~ 12–18 W m− 2 for historical and future cases during the sunlit months (i.e., April-September) due to increased exposure of dark water surfaces (Fig. 5a). The ocean, rather than the atmosphere, absorbs much of the extra SW radiation (Fig. 4a), resulting in weak summer surface warming (Fig. 3a). Cloud feedback is negative in response to sea-ice loss during April-August, and the cooling is larger in the future SIC case (-1.5~-4.5 W m− 2) than the preindustrial SIC run (-1.0~-1.5 W m− 2). Lapse rate (Fig. 5c) and Planck (Fig. 5d) feedbacks weakly respond to historical or future Arctic SIC changes in summer due to small surface warming (Fig. 3a) during the sunlit season. We also find negligible water vapor feedback in response to Arctic sea-ice loss throughout the year, which differs from the noticeable water vapor feedback in response to SST warming (Fig. 5b).
The large cold-season surface warming in response to historical and future Arctic sea-ice loss enhances Arctic lapse rate (Fig. 5c) and Planck (Fig. 5d) feedbacks. When Arctic surface warming (Fig. 3a) and AA (Fig. 3c) peak from October-December, the lapse rate feedback increases the incoming TOA radiative flux by ~ 4–6 W m− 2 (~ 8–11 W m− 2) and the Planck feedback opposes warming by -6~-8 W m− 2 (-16~-20 W m− 2) due to historical (future) sea-ice loss. Note that the month of maximum (minimum) lapse rate (Planck) feedback in the historical and future SIC cases (Fig. 5c) corresponds to the month of peak Arctic surface warming (Fig. 3a), which in turn is related to peak oceanic heating (Fig. 4a) induced by sea-ice loss (Fig. 4f) in these simulations. The cloud feedback in response to future Arctic sea-ice loss also enhances the net incoming TOA radiative flux from October-January by ~ 2.5-3.0 W m− 2, but the cloud feedback is weak (< 1.0 W m− 2) during winter in response to historical sea-ice loss (Fig. 5e). In contrast to the SST change simulations, Arctic atmospheric energy convergence weakens by 7 W m− 2 and 9–13 W m− 2 in response to historical and future sea-ice loss from October-January, respectively (Fig. 5f). Enhanced Arctic warming in response to sea-ice loss in the non-summer months (Fig. 3a) weakens the temperature gradient between the midlatitudes and polar regions, thus reducing atmospheric energy convergence into the Arctic region.
Warmer SSTs enhance poleward atmospheric energy transport at all latitudes for each model for the historical (Fig. 6a) and future (Fig. 6b) SST warming cases, with slightly larger increases in the northern hemisphere than southern hemisphere from October-March. All models, except CESM2, show enhanced cold season northward energy transport with peak increases of ~ 0.18 (~ 0.22) PW around ~ 45°-50°N for the historical (future) SST warming cases. In CESM2, atmospheric energy transport shows peak increases of 0.27 (0.40) PW around 30°N for October-March. Thus, without large Arctic warming related to sea-ice loss, the atmosphere displaces energy surpluses poleward. For the SIC perturbation experiments, there is a net decrease in poleward atmospheric energy transport around 30°-90°N with a maximum decrease of -0.05~-0.12 PW around 60°N but little change south of 30°N for both historical (Fig. 6c) and future (Fig. 6d) sea-ice loss, consistent with Deser et al. (2015). Therefore, SST-induced background warming enhances atmospheric poleward energy transport into the polar regions, while large Arctic warming in response to sea-ice loss weakens atmospheric poleward energy transport over the northern mid-high latitudes.
Figure 7 shows the potential warming contributions of the climate feedbacks over the Arctic and the tropics. A larger (smaller) Arctic than tropical warming contribution indicates that the process contributes positively (negatively) to AA (Pithan and Mauritsen 2014). We focus on the warming contributions from October-March (rather than the annual-mean) as surface warming and AA is largest during the cold season (Fig. 3a) and warm-season positive contributions (such as that from surface albedo feedback) do not lead to significant surface warming as the energy is stored in the upper ocean. Atmospheric energy convergence is the largest contributor to cold-season AA under historical (Fig. 7a) and future (Fig. 7b) global SST warming, as it redistributes the energy from the lower latitude oceans, where SSTs increase, to the Arctic region. In contrast, oceanic heat release opposes AA in response to global SST warming (Fig. 7a-b) because the warmer SSTs produce a greater ocean-to-atmosphere energy flux outside the Arctic, thus causing more warming in the tropics than in the Arctic. Water vapor feedback makes a small contribution to Arctic warming due to low October-March mean temperatures but contributes to ~ 1 K of warming in the tropics in response to global SST warming (Fig. 7b), opposing AA. Without sea-ice loss, lapse rate feedback contributes little to Arctic warming but produces weak tropical cooling in response to historical (Fig. 7a) and future (Fig. 7b) SST increases from October-March. The local Planck feedback (relative to the global-mean Planck feedback) slightly contributes to AA in the SST warming runs because the cooling effects from Planck feedback are slightly less in the Arctic region than over the rest of the world (Fig. 7a-b). Surface albedo and cloud feedbacks contribute little to Arctic warming or AA in response to global SST increases and fixed Arctic SIC during October-March for historical (Fig. 7a) and future (Fig. 7b) cases.
In response to Arctic sea-ice loss with fixed global SSTs, oceanic heat release is the largest contributor to AA from October-March in historical (Fig. 7c) and future (Fig. 7d) SIC cases, followed by the positive lapse rate feedback. This supports previous studies that showed that sea-ice loss and oceanic energy release during Arctic winter are necessary to trigger large surface warming and thus strong positive lapse rate feedback in the Arctic (Feldl et al. 2020; Jenkins and Dai 2021; Dai and Jenkins 2023). The local Planck feedback (relative to the global-mean Planck feedback) also contributes to Arctic warming and AA in response to historical (Fig. 7c) and future (Fig. 7d) Arctic SIC changes by cooling the Arctic region less than the tropics. Additionally, positive cloud feedback makes a slight contribution to cold-season Arctic warming and AA in response to future Arctic SIC loss (Fig. 7d), but the contribution is negligible in the historical SIC loss run (Fig. 7c). Water vapor feedback is suppressed over the Arctic and globe in the historical (Fig. 7c) and future (Fig. 7d) SIC runs, suggesting that local sea-ice loss and water vapor feedback are decoupled, as found previously (Jenkins and Dai 2021). In contrast to the perturbed SST runs, the atmosphere displaces energy away from the Arctic in response to cold season sea-ice loss (Fig. 7c-d), thus opposing AA.
3.4 Physical processes underlying climate feedbacks
Water vapor feedback is complicated in high latitudes due to local temperature inversions and low amounts of water vapor (Curry et al. 1995; Sejas et al. 2018). Global maps reveal that SST warming (Fig. 8a, b) has a larger effect than local sea-ice loss (Fig. 8c, d) on water vapor feedback in both the Arctic and remote areas. Specifically, water vapor feedback is largest near the equator at ~ 2–5 W m− 2 in response to historical (Fig. 8a) and future (Fig. 8b) SST warming and decreases poleward to ~ 0.5-1.0 W m− 2 in the Arctic region (Fig. 8a, b). The cold-season water vapor feedback is weak in response to Arctic sea-ice loss (Fig. 8c, d), including over the Arctic where low-level specific humidity increases (Fig. 9c, d). This is due to low or negative values of the October-March LW and net (i.e., LW + SW) water vapor kernel in the Arctic lower troposphere (Fig. 10a, c). Because the water vapor feedback is most sensitive to upper tropospheric water vapor content (Shell et al. 2008; Soden et al. 2008; Pendergrass et al. 2018), the low-level water vapor increases in response to Arctic sea-ice loss do not lead to large TOA flux changes.
Slight positive water vapor feedback occurs over sea-ice loss areas in the historical SIC loss run (~ 0.50–0.75 W m− 2; Fig. 8c) but there are negligible water vapor feedback effects in the Arctic under future SIC conditions (Fig. 8d). As the October-March LW and net water vapor kernel is negative near the surface (Fig. 10a, c), any increase in moisture in the lower troposphere will result in enhanced radiative emission to space (i.e., a negative water vapor radiative effect). In response to future Arctic SIC (Fig. 9d), there are greater increases in the natural logarithm of specific humidity [∆ln(q)] in the lower troposphere than in the historical case (Fig. 9c). Thus, greater future lower tropospheric moistening in the Arctic region produces a more negative water vapor radiative effect at the TOA. We also note that there is a large spread (as shown by the standard deviation) among the PAMIP models and individual ensemble members in upper tropospheric moistening in the perturbed Arctic SIC runs, where there is little change in the mean ∆ln(q) (Fig. 10c, d). Thus, some ensemble members may have experienced a slight decrease in upper tropospheric ∆ln(q) in response to Arctic sea-ice loss with fixed global SST, enhancing outgoing LW radiation at the TOA. In contrast, the historical (Fig. 10a) and future (Fig. 10b) perturbed SST runs experienced slightly greater ∆ln(q) in the upper troposphere than the lower troposphere for both warm and cold seasons. Due to positive values of the TOA LW and net Arctic water vapor kernel in the upper troposphere (Fig. 10a, c), top-heavy moistening in response to global SST warming produces a positive water vapor feedback from the TOA perspective.
Arctic low cloud amount has been suggested to increase during the cold season in response to sea-ice loss due to decreased lower tropospheric stability (Kay and Gettelman 2009; Jenkins et al. 2023), thus affecting Arctic cloud feedback (Vavrus 2004; Morrison et al. 2019; Jenkins and Dai 2022). We find weak October-March cloud feedback in response to perturbed SST with fixed Arctic SIC for historical (Fig. 11a) and future (Fig. 11b) cases, suggesting that remote processes do not greatly impact Arctic cloud feedback. On the other hand, Arctic sea-ice loss produces a large positive cloud feedback response in winter, especially in regions with large sea-ice loss and surface warming (Fig. 11c, d). For the run with historical SIC loss, cloud feedback enhances the TOA radiative flux by ~ 2–5 W m− 2 in the Barents-Kara Seas region and by ~ 0.5-1.0 W m− 2 in the Chukchi Sea, where the largest sea-ice loss and surface warming occurs. Under future Arctic sea-ice loss, cold-season cloud feedback is largest in the Barents-Kara Seas (~ 3–5 W m− 2) except the warming effects from clouds extend into the Central Arctic Ocean. This is likely related to the greater area with large sea-ice loss (Fig. 1b, d) and surface warming (Fig. 2c-d) in the future case than in the historical case.
The lapse rate feedback experiences large seasonal and spatial variations in the Arctic in response to SST warming or Arctic SIC loss. From October-March, the lapse rate feedback is negative-neutral in response to the global SST warming (Fig. 12a, b) due to relatively uniform vertical warming profiles (Fig. 13a, b). We note that without changes in SIC, there are negligible changes in Arctic oceanic heat uptake or surface warming in the cold season, leading to suppressed lapse rate feedback (Fig. 12a, b). In contrast, cold-season sea-ice loss enhances Arctic lapse rate feedback for historical (Fig. 12c) and future (Fig. 12d) SIC cases when surface and lower tropospheric warming outpaces warming in the mid-upper troposphere (Fig. 13c, d). We note that lapse rate feedback strengthens (~ 6–10 W m− 2) in regions with the greatest October-March oceanic heat release and surface warming in response to historical (Fig. 12c) and future (Fig. 12d) sea-ice loss, consistent with previous studies (Dai et al. 2019; Feldl et al. 2020; Boeke et al. 2021; Jenkins and Dai 2021, 2022; Dai and Jenkins 2023). Thus, sea-ice loss is necessary to produce bottom-heavy warming and trigger Arctic positive lapse rate feedback during winter, as shown previously by Dai and Jenkins (2023) using coupled model experiments.
3.5 Response to simultaneous SST and SIC changes
We compare the Arctic vs. tropical October-March potential warming contributions of climate feedbacks, changes in atmospheric energy convergence and oceanic heat release in response to historical global SST warming and historical polar sea-ice loss together (i.e., pdSST-pdSIC minus piSST-piSIC; Fig. 14a; referred to as TOTAL) and the sum of the separate responses to historical SST warming (i.e., pdSST-pdSIC minus piSST-pdSIC) and historical polar sea-ice loss (i.e., pdSST-pdSIC minus pdSST-piArcSIC and pdSST-piAntSIC) (Fig. 14b; referred to as SUM). The warming contributions of the lapse rate, water vapor, cloud, and Planck feedbacks in TOTAL match SUM well, with the lapse rate feedbacks making the largest contribution to AA (Fig. 14). Except for CESM2 in TOTAL, the change in atmospheric energy convergence makes roughly equal warming contributions to Arctic and tropical warming from October-March, suggesting that remote SST warming and Arctic sea-ice loss have opposing effects on the horizontal atmospheric energy flux. The oceanic heat release changes in IPSL-CM6A-LR makes a greater contribution to Arctic than tropical warming, but there are slight discrepancies between CanESM5 and CESM2 oceanic heat release between TOTAL and SUM. In TOTAL, CanESM5 and CESM2 oceanic heat release changes contributes roughly the same amount to Arctic and tropical warming; however, CESM2 (CanESM5) produces slightly greater Arctic (tropical) warming in SUM. The surface albedo feedback is inactive from October-March due to lack of sunlight and is not a major contributor to large cold-season AA.
The northward atmospheric energy transport response to the SST and SIC perturbations is similar among TOTAL (Fig. 15a) and SUM (Fig. 15b), with little difference between the two cases (Fig. 15c). In the tropical regions (i.e., 30°S-30°N), remote SST warming enhances poleward atmospheric energy transport by ~ 0.1–0.15 PW in the southern hemisphere and ~ 0.1–0.35 PW in the northern hemisphere. Around 60°-90°N, there is little net change in atmospheric energy transport in response to simultaneous SST and SIC changes, suggesting that remote warming due to SST changes and local Arctic warming related to sea-ice loss have opposing effects on Arctic atmospheric energy transport. The similarity of climate feedbacks (Fig. 14) and the atmospheric energy transport (Fig. 15) response between TOTAL and SUM suggest that the effects of SIC or SST changes can be linearly separated. In other words, the individual responses to SST or SIC perturbations approximately sum to the combined influence of changes in SST and SIC.