3.4 Model-simulated WWEs, past and future
The same analysis method is applied to ten ensemble members produced by the CESM. Daily fields of 500 hPa geopotential height anomalies are mapped onto the characteristic patterns in the master SOM that were derived from reanalysis output. The closest matching node is identified for each simulated daily field. It should be noted that a master SOM created using model fields produces a very similar matrix of patterns (Fig. S9 ). Moreover, our previous work (Francis et al., 2020) demonstrated that three models participating in the Climate Model Intercomparison Project – version 5 (CMIP5) successfully captured monthly distributions of days belonging in each node as well as frequencies of LDEs during the years included in the historical period (1979-2005).
In terms of changes in WWEs over the historical period, we find CESM simulations agree reasonably well with changes observed in the reanalysis output (Fig. S10) . Given the large interannual variability in WWEs (Fig. 6a ) and the fact that the 1995-2005 period is relatively early in the era of the climate-change signal generally and the emergence of AAW specifically (node #1), it is not surprising that changes over two relative short periods do not align perfectly.
To further demonstrate the utility of this approach, next we explore projected future changes in WWEs, assuming RCP 8.5 forcing conditions in CESM. Our findings presented in Fig. 7 suggest robust changes in the number of WWEs will occur from the beginning of the 21st century (2006-2030) to the end (2076-2100) during winter and summer seasons. Nodes #1 and #3 exhibit robust increases in both seasons, while WWEs originating in node #12 are projected to decrease significantly. These changes can be attributed to a combination of changing frequency of occurrence of each node (Fig. S11 ) as well as changes in the probability (number of WWEs/number of days in a node) of a WWE originating in each node (Fig. S12 ). Patterns with positive height anomalies over northern and northeastern North America are projected to not only occur more frequently (Fig. S11 ), but also to spawn a higher frequency of WWEs in a warmer world. To check for intermodel consistency, we also analyze output from three of the models included in the CMIP5 generation. We find they project changes in WWEs similar to those from the CESM runs (Fig. S13 ). These findings point toward an increased occurrence of regime-based WWEs in the future, as patterns with positive height anomalies in high latitudes become more common and are also more likely to initiate WWEs, while those with negative anomalies in Arctic regions will occur less frequently and are less likely to spawn WWEs.
Finally, we calculate the annual mean Euclidean distances separating the two nodes participating in a WWE, based on CESM output from 2006 to 2100 (Fig. S14 ). This metric provides an indication of the severity of a WWE, as a larger Euclidean distance would represent a more dramatic shift in the pattern. We find that 7 of 12 nodes exhibit statistically significant trends; five with declining trends and two with increasing trends. The preponderance of negative trends may result from the ongoing decrease in the poleward temperature/height gradient as high latitudes warm faster than lower latitudes and the corresponding reduction in airmass contrast. This explanation is supported by the fact that patterns with low height anomalies in the Arctic (nodes #11 and #12) exhibit increasing trends, reflecting larger airmass contrasts when the Arctic is anomalously cold.