Memory effects of Eurasian land processes cause enhanced cooling in response to sea ice loss

Amplified Arctic warming and its relevance to mid-latitude cooling in winter have been intensively studied. Observational evidence has shown strong connections between decreasing sea ice and cooling over the Siberian/East Asian regions. However, the robustness of such connections remains a matter of discussion because modeling studies have shown divergent and controversial results. Here, we report a set of general circulation model experiments specifically designed to extract memory effects of land processes that can amplify sea ice–climate impacts. The results show that sea ice–induced cooling anomalies over the Eurasian continent are memorized in the snow amount and soil temperature fields, and they reemerge in the following winters to enhance negative Arctic Oscillation-like anomalies. The contribution from this memory effect is similar in magnitude to the direct effect of sea ice loss. The results emphasize the essential role of land processes in understanding and evaluating the Arctic–mid-latitude climate linkage.

Nonlinear response depending on the initial land condition. There is a strong nonlinearity in the responses between the initial land surface conditions. Strong and obvious negative AO-like circulation patterns and mid-latitude cooling anomalies are found in the memory effect with LICE's land condition ( Interestingly, significant anomalies only appear when the experiments are performed with the LICE land condition. The atmosphere and land conditions adjusted to the high sea ice condition appear to be more stable with respect to the sea ice change than those adjusted to the low ice condition. To shift the climate regime from one adjusted to the high ice condition to one adjusted to the low ice condition might require repeated forcing from persistently low sea ice conditions, as have been observed in recent decades. Figure 1. Simulated winter atmospheric responses to sea ice reduction. This is the same as Fig. 2 in the main text except that the Sea ice and Memory effects are separately evaluated by the procedure using a and c LICE † and b and d HICE † runs, corresponding to the left and right panels in Fig. 1, respectively. Red and blue indicate positive and negative anomalies, respectively, and light and heavy grey shadings indicate statistical significance exceeding 95% and 99%, respectively.

Supplementary Note 2.
Characteristics of the seasonal cycle of soil conditions to form the memory effect. We performed additional 'initialized run' experiments that extended the integration period by three years. Specifically, we examined the characteristics of the memory effect, particularly the way in which they evolve beyond interannual cycles.
As we discuss in the main text, circulation anomalies due to the sea ice effect bring cold anomalies over Siberia during winter. This Siberian cooling forces the soil to cool in the late winter of every year (Supplementary Figure 2a). Cold soil temperature anomalies intensify year by year, and the coldness peak moves later each year (see Fig. 6 in the main text). The largest difference between the sea ice and memory effects is anomalous soil coldness in the late spring to early autumn.
Next, we examined the causes of this soil coldness. In the sea ice effect, during late winter to early spring, anomalous soil coldness is largely led by the anomalous upward radiation forcing (Supplementary Figure 2b, dashed lines). This corresponds to a shortage of short-wave radiation due to the increased snow cover (Fig. 5 in the main text) because the contribution of long-wave radiation is small (not shown). In turn, anomalous soil coldness cools the atmosphere through turbulent heat flux anomalies (Supplementary Figure 2b, solid lines). In the memory effect, as a result of the delayed coldness peak, the anomalous soil coldness cools the atmosphere through turbulent heat flux during late spring. After that, heat flux anomalies are nearly zero during summer to autumn, while the anomalous coldness gradually weakens. This indicates that the air temperature that adjusted to the cold ground condition in spring persists until autumn ( Fig. 2c and 2d in the main text). This, in turn, possibly brings anomalous coldness and early snowfall in the autumn and early winter as a memory effect (Fig. 5 in the main text).

Supplementary Note 3. Relationship between ground conditions and the Eurasian winter climate in AMIP-like simulations.
We analysed the output of the historical simulation using AFES ( 1 Ogawa et al., 2018). There are two types of 30-member ensemble simulations for the 1979-2014 period. One is a historical simulation forced by historical monthly mean SST and sea ice boundary conditions based on the Merged Hadley/OI SST dataset ( 2 Harrell et al., 2008). The other is similar to the historical simulation, but SST was fixed as the climatological mean annual cycle. Details of the simulation configuration are described by 1 Ogawa et al. (2018). Here, we evaluate the relationship between long-term trends of preceding ground conditions (October-November snow cover and July-August-September soil temperature) and winter (December-January-February) surface temperature averaged over the eastern part of the Eurasian continent (60-120°E, 40-60°N). In this historical simulation using AGCM, the long-term trend varies among ensemble members. Such intra-ensemble differences could occur due to random noise from the atmospheric internal variation or variation of ground conditions. We then evaluate the intra-ensemble relationship.
Autumn snow cover seems to have no relation to Siberian/East Asian winter temperature in the historical simulation (Supplementary Figure 2a These results suggest that the ground conditions (autumn snow and summer soil temperature) have the potential to affect the winter climate even in the long term. However, other external conditions, such as SST variations, largely disturb this relationship because the relationship is stronger in varying ICE than in varying SST&ICE.

Supplementary Note 4.
Observational evidence of the relationship between sea ice variation and ground temperature anomalies.
To evaluate observation-based evidence supporting our simulation results, we conducted an additional analysis using borehole temperature data from the GTN-P dataset (  shown. Note that the sign of coefficient is reversed to show anomalies associated with a decrease in sea ice.
the data period is 10 years, these in-situ observations are quite helpful in examining the role of ground processes. We calculated the correlation between preceding sea ice anomalies over the Barents/Kara Seas and annual mean borehole temperatures at individual stations.
Correlations are generally positive, indicating a ground temperature increase after sea ice loss in accordance with the long-term tendency due to the global warming (Supplementary Figure 4a). On the other hand, when we looked at the relationship of interannual variations by conducting a detrend analysis, negative correlations appear over the eastern part of Eurasia, which indicates the occurrence of cold ground temperature anomalies after the sea ice loss, (Supplementary Figure 4b), while two stations near the Barents/Kara Seas still had positive correlations. Although the correlations are not statistically significant, these signals roughly correspond to our simulation results showing temperature response to sea ice loss ( Fig. 2b and Fig.  3d in the main text).