Future Meridional Wind Trends Through the Lens of Subseasonal Teleconnections

Large-scale atmospheric circulation is expected to change dramatically in the upcoming decades, and with it, the interaction between Rossby waves and the jet stream. A common feature of midlatitude wintertime variability is upper tropospheric quasi-stationary number-5 wave packets, which often propagate zonally along the jet. These are collectively referred to as the Circumglobal Teleconnection Pattern (CTP). Their likeness surprisingly emerges as a robust signal in future meridional wind trend projections in the Northern Hemisphere. 5 We attempt to elucidate this link across timescales, focusing on wave propagation in the jet waveguide in observations and a 36-member ensemble of CMIP5 models. Using EOF analysis on 300 hPa subseasonal V anomalies, we first establish the ensemble’s skill in capturing the pattern. Then, by investigating EOF phase space, we characterize the CTP’s behavior in present day climatology and how it is projected to change. Under RCP8.5 forcing, most models develop a gradual preference for monthly-mean waves with certain longitudinal phases. The ensemble is thus divided into subgroups based on region of 10 increased wave activity. This categorization strongly corresponds to the ensemble spread in local trend magnitude. Additionally, in two test-case models, this coincides with an increasing number of preferably phased wave packets at the synoptic scale. Some signs suggest that differences in CTP dynamics might stem from mean flow biases, while no evidence was found for the role of tropical diabatic forcing. Thus, we conclude that this hemisphere-wide climate change signature is actually comprised of several regional effects, 15 partly related to shifts in CTP phase distributions. The strong dynamical disagreement in the ensemble then manifests as significantly different circulation trends, which in turn might affect projected local temperature and precipitation patterns.


CTP events
We define "CTP events" in daily mean data. These are essentially Rossby Wave Packets that are nearly in phase with the 500 hPa equivalent of the preferably-phased pattern that was found in the 300 hPa monthly projections. First, we apply a 3-day running mean on the 500 hPa daily V field. After calculating the projection index and excluding low values (as in the monthly 160 case), we detect all sequences of three or more consecutive days in which |φ d − φ m | ≤ π/8, where φ d and φ m are the daily and preferred-monthly phases respectively.
When creating wave composites, we remove the Future climatology from the daily means in order to observe wave propagation and to remove the signature of other low frequency patterns. Lag 0 of a CTP event is defined as the first day of the sequence and statistical significance is determined by a 1000 member bootstrap method. 165 Additionally, we used lagged linear regression (Livezey and Chen, 1983) in an attempt to establish a causal relationship between tropical convective forcing (expressed by OLR and upper tropospheric divergence) and the excitation of CTP events.
However, the resulting patterns were not statistically significant, hence additional technical details are only provided in the supplementary materials. The CTP can be obtained by calculating the first two leading EOFs of the monthly winter (DJF) subseasonal anomalies of the upper-tropospheric meridional wind. The observational patterns for NCEP-I reanalysis can be seen in Fig. 2. Very similar patterns were also produced from ERA-interim data (not shown).
In both cases, the set of EOFs is comprised of two quasi-stationary zonal number 5 waves which are in quadrature with 175 one another. They explain 13.6 and 10.8 % of the variance (denoted by λ) in the NCEP/NCAR dataset, and 13.3 and 11.5 % in ERA-Interim. For ERA-Interim, the two EOFs are not well-separated from one another (as well as from the third EOF) according to the definition set by North et al. (1982). For NCEP-I, a longer dataset, the patterns are well-separated.
Performing the same calculations on 36 GCMs from the CMIP5 ensemble reveals the robustness of this pattern. In order to allow a comparison to observational present-day climate, the monthly data used in calculating the EOFs came from historical 180 model runs only (between the years 1900-2005). We base all of our calculations and projections hereafter on the historical EOFs, even when working with data from Future runs, as no considerable differences were found in EOFs based on RCP8.5 data (not shown). EOFs, with a stronger amplitude above North America and correctly phased patterns. As expected from theory, the waves are latitudinally confined to the area of the climatological jet stream. However, it is worth noting that model agreement on the first EOF is more robust, and that it is also closer to the observational function, as will be demonstrated quantitatively.
For most models, the variance explained by these patterns is slightly higher than in observations, with median ensemble values of 14.8 % for the first EOF, and 10.7 % for the second. Around two thirds of ensemble members have well-separated 190 EOFs as well, with EOF2 separated from both the first and third functions in most cases. This might be a result of the model runs being almost twice as long (105 winters) as the observational records (57 winters in the longer dataset, NCEP-I).
In order to quantify each model's skill in representing the CTP, each set of two EOFs was projected onto the observational patterns with cosine latitude weighting. Most models can reproduce the spatial features of the CTP fairly well. With a score of 1 representing a perfect copy, the ensemble has a median score of 0.78 and 0.5 for the first and second EOFs respectively.

195
There is however a larger spread in skill for the second function.

Projecting the MMM climate trend onto the EOFs
A quasi-stationary zonal number 5 wave in the northern hemisphere appears to be a prominent feature of the projected circulation response to GHG forcing. The similarity between the CTP and CMIP5 future circulation trends has been noted before 200 (Brandefelt and Körnich 2008;Branstator and Selten 2009;Simpson et al. 2016). However, to our knowledge no effort has yet been made to objectively link the two based on the EOF perspective that was used by B02 to first define the CTP.
Under scenario RCP8.5, the multi-model mean response in the 300 hPa meridional wind indeed takes the form of a number 5 wave (Fig. 4). This response being the Future climatology (2070-2099) minus the Past climatology . Nonetheless, when projecting each model's trend onto its historical EOFs, one gets rather low scores (absolute mean of 0.24 and 0.21 for 205 EOF1 and EOF2).
In order to further test this point, we calculated how much of the trend's spatial variance was explained by the EOFs (Fig. 4).
This was done individually for every model, as well as for the multi-model mean (using composite EOFs and the MMM trend).
For most models used, the variance is spread quite evenly across the first five EOFs (the higher-ordered of which have lower λ values). In particular, the first two leading functions, which define the CTP, explain less than half of the trend's variance for 26 210 projection scores and angles close to the preferred mean monthly phasing. These events are defined in relation to a specific region, according to the EOFs used for projection. The list of potential events is further filtered in order to remove false positive matches.
We first apply this method to observational datasets, with wave phases taken from the CMIP5 subgroups for the NA and AS 290 regions. For the NCEP reanalysis, an average of 1.5 events per season was found in the NA region, and 0.5 events per season in AS. Meanwhile, the ERA-Interim dataset seems to capture more events, with an average of 1.8 and 0.9 per winter for the two respective domains.
These results comfortably fall within the range set by Souders et al. (2014) for winter RWP frequency. In their comprehensive climatological study, they tracked ∼ 6000 RWPs that appear globally throughout the year. They found that, on average, about 295 11 RWPs form over the Pacific Ocean every winter, while 9 are formed over the Atlantic. Geographically, their domains roughly correspond to our NA and AS regions. Our CTP events likely constitute only a small subset of this climatological RWP dataset (which covers all phases and a range of wavelengths), making their result serve as a loose upper bound.
In the two GCMs, behavior of the CTP changes between the Historical and RCP8.5 periods, much along the lines of the monthly data results. For IPSL and the NA region, the number of events nearly doubles between the Past and Future periods 300 (from 1.6 to 3 events per winter). In the AS region, MIROC CTP event frequency increases from 1.1 to 1.9 for the same metric.
It's important to note that these changes only happen for a narrow range of wave phases (less than a full quadrant) within every model's domain. One interesting exception is found in MIROC, which shows the this trend for both domains, in the daily timescale only.
For both models, the additional events are not evenly spread throughout the RCP8.5 winters, but are rather concentrated 305 in a few seasons with amplified wave activity (Fig. 9). As the average number of events per winter increases, the tail of the distribution shifts as well. Thus, we begin to see winters with five or more events, which is unprecedented before 2006.
Performing the same analysis on a long Pre-industrial Control run (unfortunately available only for IPSL) further underlines how the projected RCP8.5 future differs from unforced natural variability.

RWP propagation 310
We can mark the first day of each series of CTP days within an event as lag 0 and follow the propagation of the RWPs. By definition, each such lag 0 day has a high projection score on the preferred phase, meaning that the RWP is at the peak of its life cycle. By creating a simple composite of all events for each model and domain, we observe the general features of a "typical" RWP, without tracking individual waves.
The wave packets all share realistic characteristics that have been previously affirmed by observations and models (Souders    Figure 7. Linear combination of regional EOFs: cos(γ)VEOF 1 + sin(γ)VEOF 2. γ is set to represent each group's approximate preferred phasing (−π/8 for NA, 3π/4 for AS). Contour interval is 1 ms −1 . Light shading is the MMM Future-Past V trend, and dashed lines define the domains of the regional EOFs.
24 https://doi.org/10.5194/wcd-2020-6 Preprint. Discussion started: 19 February 2020 c Author(s) 2020. CC BY 4.0 License.   Figure 11. Group bias in the jet stream. Data shown is monthly 300 hPa DJF zonal wind group composites for the Past period , with the MMM pattern removed. Contour interval is 1 ms −1 , with negative values denoted by dashed lines. Thick contours show areas where at least 90 % of group members have the same sign as the group mean. Shading represents the MMM 300 hPa climatological jet stream (U ≥ 20ms −1 ), averaged over the same period. Table 1. CMIP5 model groups based on preferred regional CTP phasing of monthly mean data projection.