3. Results
3.3 Inter-ensemble regression patterns
We next investigated the inter-ensemble relationship of atmospheric variable changes to ∆LOTC in order to check different levels of response of individual ensemble members. Changes of SAT and SLP of individual ensemble members, i.e.
∆SAT and ∆SLP, are regressed onto respective ∆LOTC (Fig. 9 a,b). There is a strong negative value over the Asian continent extending to the northwestern Pacific Ocean and a positive correlation over the northeastern Pacific Ocean particularly in the Bering Strait (Fig. 9a). Change over the continent is greater than that of the Pacific Ocean, suggesting that the change in continental temperature has a greater effect on the value of ∆LOTC. A dipole-like pattern exists over the North Pacific Ocean, extended cooling in the west and warming over the Bering Sea. This pattern is consistent with ∆SLP distribution, which shows a negative relation with
∆LOTC over the Bering Sea (Fig. 9b). This negative relation suggests that ensemble members with reduced ∆LOTC show weakening of the AL.
SLP response to SAT changes over ocean and land is investigated by defining two indexes, ∆SATocean (40°−60°N, 170°E−150°W) and ∆SATland (40°−60°N, 80°−120°E). The ∆SLP distribution regressed on ‒∆SATland (Fig. 9c) for each 40 ensembles shows a similar pattern with that on ∆LOTC (Fig. 9b). Fig. 9d is ∆ SLP distribution regressed on the ∆SATocean for each 40 ensembles. It is notable that ∆SLP over the Bering Sea is larger on ∆SATocean than ∆SATland, which is opposite to the case of Gan et al. (2017) (see their Figure 10c,d). The increase of SAT over the continent appears to have less of an effect on the change of SLP over the ocean in the KCM. We suppose that this difference could be one of reasons why the damping effect of ∆LOTC on the AL is relatively weak in the KCM result.
Fig. 9 Inter-ensemble regression patterns of the changes of (a) SAT and (b) SLP onto the LOTC changes in winter. (c), (d) as in (b), but for regressions onto the normalized SAT changes over the Asian continent and the North Pacific Ocean, respectively. Here SAT changes are calculated by the domain-mean (30°−70°N, 80°E−120°W) SAT warming in each member. Stippling indicates regions where statistical significance is greater than 95%. Contour lines show explained variance
Next, in order to examine the effect of equatorial SST changes on mid-latitude SLP distribution, we performed a Singular Value Decomposition (SVD) analysis (Bjornsson and Venegas 1997) using SST and SLP spatial distributions of 40 ensembles. The areas used for the analysis were 20°S−20°N, 120°E−79°W and 30°−70°N, 130°E−120°W, for SST and SLP, respectively. Originally, SVD
analysis calculated the covariance of SST and SLP over time, but in this study, the spatial distribution covariance of SST and SLP for 40 ensembles was calculated instead of time variation, following Gan et al. (2017). The SST expansion coefficients estimated by the SVD analysis were used to regress on tropical SST (20°S−20°N, 120°E−79°W) and North Pacific SLP (20°S−75°N, 110°E−80°W) (Fig. 10).
Regressed SST pattern shows again the El Niño-like SST warming centered in 180°W–150°W (Fig. 10a). SLP changes associated with SST increase have a negative correlation with a decrease around the Aleutian Islands (Fig. 10b). The increase of SST in the central and eastern tropical Pacific makes lower SLP to the south of the Aleutian Islands amount to 1.25 hPa. Compared with Gan et al.
(2017) (see their Figure 10a), the center of maximum negative correlation is moved to the southeast of the Aleutian Islands, outbound of the Bering Sea. In particular, the location of the maximum negative correlation with SLP is nearly identical with the area where the standard deviation of 40 ensembles is increased in future climate (Fig. 5d). From these results, the El Niño-like SST warming in future climate could be a source to strengthen the variability of the AL.
In the KCM, as we confirmed in the correlation (see Fig. 8), the sea ice loss in marginal seas of the North Pacific is the most important factor to change the AL strength. To understand the effect of sea ice loss in the Bering Sea and the Okhotsk Sea, the SAT change (∆SAT) is regressed on average SIC reduction (‒∆
SIC) in the area of 50°−70°N and 140°E−150°W (Fig. 11a) for each 40 ensembles. With the decreased sea ice in the Pacific margin, atmospheric temperatures increase in the Okhotsk Sea and the northern Bering Sea. In the northern part of the Bering Sea, it rises sharply to about 0.7°C/%. This is understood as the result of enhanced ocean-atmosphere interactions with reduced sea ice. Fig. 11b shows the ∆SLP distribution regressed on ‒∆SIC. There is a negative correlation between decreasing SLP in the North Pacific, including
marginal sea, and sea ice loss. In particular, the decrease near the Aleutian Islands is relatively larger about 0.9 hPa/%. It is quite interesting that explained variance is quite high over the Okhotsk Sea, amounting to 50%, which may indicate a contribution by ice reduction in the Okhotsk Sea to the AL deepening.
Next, we look into the effect of sea ice reduction in the Okhotsk Sea and the Bering Sea, separately. The domain (50−70°N, 135°−160°E) averaged ∆SIC in the Okhotsk Sea is different for each 40 ensembles, but all decrease within the range of ‒15 to ‒20%. The domain (50°−70°N, 170°E−150°W) averaged SIC also decreases within the range of ‒17.5 to ‒22% in the Bering Sea (not shown).
Ensemble mean values are 18.3% and 20%, respectively. Correlation coefficient between ∆NPI and ∆SIC of the Okhotsk Sea is 0.42 and ∆SIC of the Bering Sea is 0.47; both correlation coefficients have a statistical significance of over 99%.
We regressed ∆SAT and ∆SLP of individual ensembles on the average SIC reduction (‒∆SIC) of the Okhotsk Sea and the Bering Sea, separately (Fig. 11c-f).
The inter-ensemble regression pattern of SAT onto ∆SIC of the Bering Sea shows greater warming over the northeastern Bering Sea, where the SLP decrease (Fig.
11e). This may indicate that the Bering Sea ice reduction could be a result of the AL deepening such that increased cyclonic circulation brings warm air into the Bering Sea. Differently, the SAT response to ∆SIC of the Okhotsk Sea shows large warming over the Okhotsk Sea (Fig. 11c) although a similar deepening of the AL occurs (Fig. 11d). Strengthening of the AL, by its associated wind pattern, should increase cold air inflow to the Okhotsk Sea from the Arctic Sea, leading to an increase of sea ice there. However, the regression pattern indicates SIC decreases in the Okhotsk Sea even though AL is strengthened, suggesting that the response of Okhotsk SIC may not be due to changes of the AL. Therefore, we hypothesize that sea ice changes over the Okhotsk Sea and the Bering Sea may have different causes.
Fig. 10 Patterns of inter-ensemble co-variability between the changes in the tropical Pacific SST and the North Pacific SLP in winter. This is derived by (a) regressing SST change and (b) SLP change onto the SST expansion coefficient taken from
the first mode of inter-ensemble SVD. The red dashed box denotes the SLP domain used for the SVD analysis. Stippling indicates regions where the statistical
significance is greater than 95%. Note that SST change is normalized by the tropical mean (20°S−20°N, 120°E−70°W) SST warming in each ensemble
As shown in Figs. 10c and 10e, SAT increases with respect to sea ice is about 0.6 to 0.8°C/% of each section area. It also can be conjectured that sea ice loss in the Okhotsk Sea and the Bering Sea warm surface air through vertical diffusion of heat fluxes. This is consistent with Deser et al. (2010): the reduction in sea ice cover results in an anomalous net upward surface energy flux (Qnet) from the Arctic Ocean to the atmosphere. The effect on the SLP in the North Pacific by the Bering Sea’s SIC reduction correlated negatively at ‒0.8 hPa/%, limited to the Bering Sea region, suggesting also that the SIC change in the Bering Sea may be a response to the AL change.
On the other hand, Fig. 11d shows decreasing SLP from the Okhotsk Sea to the southeastern part of the Aleutian Islands due to the SIC reduction in the Okhotsk Sea. This means that the Okhotsk Sea’s SIC reduction drives the SLP change remotely, especially around the Aleutian Islands. According to Deser et al. (2016), the Arctic sea ice loss including Pacific marginal seas, the Bering Sea and the Okhotsk Sea, results in deepening the AL and thereby a significant increase in precipitation over the eastern North Pacific in the fully coupled system although they did not elucidate the contributions of sea ice loss in different regions to strengthening the AL. In this study, we found that the change of AL under global warming condition is highly correlated with sea ice loss in the Bering Sea and the Okhotsk Sea and is remotely affected by the latter.
Fig. 11 Inter-ensemble regression pattern of the winter (DJF) SAT (a,c,e) onto the area-averaged SIC reduction (∆SIC). (b) As in (a), but for the SLP. SIC reduction
area averaged over (a, b) the Pacific (50°−70°N, 140°E−150°W). (c,d) the Okhotsk Sea (50°−70°N, 135°−160°E) and (e,f) the Bering Sea (50°−70°N, 170°E−150°W). Stippling indicates regions where the statistical significance is
greater than 95%. Contour lines in Fig. 11. show explained variance
Coefficient STD.error T-value P-value(>|t|) VIF
(Intercept) 4.57 3.10 1.47 0.15
△LOTC -0.11 0.29 -0.38 0.70 1.22
△SST
(EEP–WEP) -0.49 1.43 -0.34 0.73 1.34
△SIC 0.36 0.152 2.44 0.02* 1.50
Table 1 Multiple linear regression of Land Ocean Thermal Contrast (△LOTC), El Niño-like SST warming (∆(EEP SST‒WEP SST)) and Sea ice reduction (△SIC)
on the NPI change
To quantitatively estimate the contribution of ∆LOTC, ∆SST, and ∆SIC on the NPI change, we used the multiple ordinary least square (OLS) linear regression (Stone and Brooks 1990) (Table 1). The coefficients for ∆LOTC, ∆SST, and ∆ SIC are ‒0.11, ‒0.49, and 0.36 respectively. It should be noted that only ∆SIC has statistical significance (P-value: 0.02). The variance inflation factor (VIF) of the three variables is quite low, supporting no multicollinearity problems between them, which indicates that the three variables are independent of each other. We also performed the analysis of variance (Fisher RA 1992) to check the reasonableness of the OLS analysis and found a consistent result with the OLS’s (not shown).
Therefore, we confirm again that the contribution of the ∆SIC to the AL deepening is more critical than the other factors.