the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Changing effects of external forcing on Atlantic-Pacific interactions
Evgenia Galytska
Gerald A. Meehl
Jakob Runge
Katja Weigel
Veronika Eyring
Abstract. Recent studies have highlighted the increasingly dominant role of external forcing in driving Atlantic and Pacific Ocean variability during the second half of the 20th century. This paper provides insights into the underlying mechanisms driving interactions between modes of variability over the two basins. We define a set of possible drivers of these interactions and apply causal discovery to reanalysis data, an ensemble of pacemaker simulations where the observed El Niño Southern Oscillation (ENSO) is prescribed, and a pre-industrial control simulation. We also utilize large ensemble means of historical simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to quantify the effect of external forcing and improve the understanding of its impact. By conducting a causal analysis of the historical time series, a regime switch is identified in the interactions between major modes of Atlantic and Pacific climate variability. Causal networks derived from pacemaker simulations support this finding and further demonstrate that the effect of external forcing could favor an Atlantic-driven regime between 1985 and 2014 where warming tropical North Atlantic sea surface temperatures induce a La Niña-like cooling in the equatorial Pacific during the following season through a strengthening of the Pacific Walker Circulation. This negative sign effect was not detected when the historical external forcing signal is removed in the pacemaker ensemble. The analysis of the pre-industrial control run further supports the notion that the Atlantic and Pacific modes of natural climate variability exert contrasting impacts on each other even in the absence of external forcing. We show that causal discovery can quantify previously unknown connections and thus provides important potential to contribute to a deeper understanding of the mechanisms driving changes in regional and global climate variability.
- Preprint
(7233 KB) - Metadata XML
-
Supplement
(115 KB) - BibTeX
- EndNote
Soufiane Karmouche et al.
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2023-1861', Anonymous Referee #1, 26 Oct 2023
This study investigates the Atlantic-Pacific connections using causal discovery techniques. Especially the decadal changes of connections are investigated. I am not an expert in such techniques, so my review will mere address the physical aspects of the study. In general the manuscript is well written and the research is mostly scientifically sound. However, there are several major issues that should be addressed before I can recommend publication:
1. While the connections between ENSO and TNA are properly considered, I am puzzled why another, very prominent Atlantic-Pacific connection is missing: The one related to the Atlantic Nino (measure e.g. by the ATL3 index). Numerous papers have been addressing how ENSO influences the Atlantic Nino, and how the Atlantic Nino influences ENSO, and the decadal variations of these connections have also been discussed. In a manuscript with the title Atlantic-Pacific interactions, this connection can certainly not be ignored. I strongly encourage the authors to include the ATL3 index, and perhaps other indices that potentially could act as links.
2. Proof of concept and all other connections involving ENSO: While most of the connections regarding the ENSO mechanism itself appear plausible, I am puzzled by the fact the the positive Bjerknes feedback does not properly come out. Clearly Nino3.4 should impact the central Pacific winds, which is one of the main ingredient of the Bjerknes feedback. This is probably at zero lag, therefore a straight link? By the way, I do not see any straight links in the analysis/figures.
3. Links of anything more than 1 season need special attention in their explanation. Surely, no purely atmospheric bridge could explain lags of more than 1 season (probably not even that). How can PNA and NAO impact each other at 7 month lag? There most be complex deep physical analysis here, which is completely lacking. Perhaps these links are indirect, though some SST patterns, but we can only speculate this right now. It seems that the technique used cannot answer such kind of questions.
4. Pacemaker experiments: If the SSTs in the Pacific are prescribed, how can the causal Network come to the conclusion that PWC is forcing ENSO? So, this technique is not able to detect trivial directions? Clearly the only logical explanation is that Nino3.4 is forcing PWC, but the Network does not detect this (apart from just in one case).
5. Discussion of Fig. 6b is rather hand wavy. What are the correlations between the TNA time series? ENSO should be able to impact TNA, right? This comes out from the Network, so we should see some correlation, perhaps with a 1 season lag?
Minor:
Section 4.1 For which season do you calculate the regressions shown in Figs. 3 and 4 ?
Citation: https://doi.org/10.5194/egusphere-2023-1861-RC1 -
RC2: 'Comment on egusphere-2023-1861', Anonymous Referee #2, 15 Nov 2023
This article focus on the interaction between the Pacific and Atlantic variability at interannual timescales. In particular, the authors use a novel and interesting approach based on causal discovery to investigate the causal link between the variability of the different indices from the two basins (namely ENSO, PNA, NAO, TNA, and the Pacific Walker circulation). They also investigate how those relationships are affected by external forcing and by different Pacific and Atlantic mean states. The article is overall well written. If the proposed approach is interesting and somehow promising, I have several major concerns that prevent me to recommend the article for publication.
Main comments:
1) The authors claim in the abstract (lines 14-16) that “causal discovery can quantify previously unknown connections and thus provides important potential to contribute to a deeper understanding of the mechanisms driving changes in regional and global climate variability”. However, all along the article I didn’t see clearly the adding value from the causal discovery approach used by the authors to previous knowledge.
First, in their proof of concept, the causal networks for ENSO reveals a causal link between wind stress in the western Pacific and precipitation in the Central pacific 3-month later. This link appears as independent of the central Pacific SST response to wind stress forcing. I don’t know how much importance we need to accord to this link (previously unknown?): what are its physical bases? Currently, I do have the impression that this causal link is actually an artifact and that it is due to the impacts of the Wind Stress WPAC on SST Niño3.4 (2-month lag) followed by the impacts of the SST Niño3.4 on Precip CPAC (1-month lag). This makes me wonder how confident we can be regarding the detection of “previously unknown connections” with this method. Some discussion is needed to explain this point and give credit to the method.
Second (and third), the authors split the observed record into 2 periods (1950-1983 vs 1985-2014) and apply a causal network analysis for each of these periods. This is in fact to verify whether an already proposed (from previous studies) regime switch around 1985 in the inter-connections between Atlantic and Pacific can be identified with a causal discovery algorithm. However, I don’t find this very novel nor satisfactory since the time splitting is in fact done from previously suggested period. It would have been more interesting if the author’s approach was able to detect changes in behavior from a continuous timeseries. In addition, I understand that the outputs of the causal discovery analyses are not necessarily robust (as acknowledged lines 679-682). Specifically, discarding 1 year of the 1985-2014 period leads to different results (cf. differences between Figures 5bd and S1ab). This leads me to my last comment on the method: how robust are the results discussed in the article? For example, are the differences between Figures 5a and 5b meaningful accounting for the limited sample length and the independent randomness of the variables investigated? Without giving a confidence interval in the estimated coefficients of the causal networks, I don’t think we can objectively interpret the results. (Similarly, how robust are the differences between raw and externally forced signal removed figures, e.g. Fig 5b vs 5d?)
2) The authors use a 120-year long piControl simulation performed with CESM2 to investigate the impacts of different background state in the Pacific and Atlantic on their interannual interactions. To do so, they split those 120 years into 3 periods of 40 years and perform a causal discovery analysis for each of them. They found differences and claim it is controlled by the background changes. Again, this analysis need to be done considering error on the estimates of the coefficients. Are those differences meaningful? The authors could have use more than 120 years of this piControl simulation and could have actually addressed this question computing those coefficients for every possible 40-year windows of the simulation. And then use a composite approach based on the states of the PDV and AMV, verifying whether the coefficients from those different composite pools were statistically different.
3) The authors apply a causal network on monthly (or 3-month mean) indices without considering possible changes in the interplay between those indices among different seasons. However, it has been proposed, for example, that DJF NIÑO3.4 SST anomalies are creating TNA SST anomalies of the same sign in the following MAM season, and that JJA TNA SST (as well as JJA ATL3 SST) anomalies are creating NIÑO3.4 SST anomalies of opposite sign in the following DJF season. Without accounting for this seasonality in the interactions, I believe the linked coefficients reported in the current article are actually a mixed between those interactions. Therefore, I am inviting the authors to include this seasonality in their analyses. I believe this could be done by splitting each node in 4 quadrants representing the 4 seasons. (Possibly, the lagged auto-correlation of those indices could also be represented with arrows between those quadrants, accounting for the seasonality of this auto-correlation…).
4) The authors are using pacemaker simulations in which SST is forced over Central-Eastern tropical Pacific to investigate causal relationship. I don’t understand how those kind of simulations can be useful to investigate the influence of the Atlantic on the Pacific as the Pacific SST are constrained by construction and are therefore not responding to potential Atlantic forcing. The authors should comment on that. In the same line, the causal networks of those experiments show that the PWCu is causing changes to the Nino34 SST. Could the authors explain how this is possible whereas the Nino34 SST is actually imposed?
Minor comments:
- reference for Walker Circulation (Bjeknes, 1969) is introduced line 42 instead of line 35.
- line 62: “While debate over the precise attribution of recent warming trends”, the authors may be referring to the global warming hiatus of the 2000-2014 period, but I am not completely sure. Please clarify.
- line 68: “differences in the considered timescales”, I don’t understand to what this is related. Remove?
- Section 2.4: NAO indices defined from seasonal mean EOF. I am surprised by this choice as the locations of the NAO centers of action are not the same between the 4 seasons. Given the higher variance of the winter SLP, it is likely that the EOF is capturing the winter NAO pattern. I am then wondering the physical meaning of this for the other seasons, especially in summer. This comment could also apply for the PNA index.
- lines 128 and 130: please defined what you mean exactly by “seasonally averaged”, i.e. 3-month.
- line 132: “for the pacemaker simulations”, I believe this comment is valid for the piControl simulation too.
- lines 139-144: definition of the AMV and PDV indices. The variations of the global averaged SST index are driven by external forcing but also by the expression of regional internal variability. For example, if the SST over the whole globe were showing no anomaly except in the North Atlantic, the global SST average anomaly would be due to the North Atlantic. By removing the global SST average to local SST, one not only “detrend” the local data, but also remove the global expression of the regional variability. I understand that this approach might be used in same circumstances. However, in the present article the authors are estimating the effect of external forcing from a CMIP6 model multi-ensemble mean (cf. Section 3.1), detrending the observed data with it. Therefore, I would advice to not remove the global SST average to estimate the AMV and PDV indices.
- line 162: “Consequently, the discrepancies in each pacemaker simulation relative to MEM can be attributed to internal variability”, please note that the differences between the CESM2 pacemaker simulations and the MEM can also come from: (i) the real climate sensitivity to external forcings in the restoring region being different from the one estimated by the MEM, (ii) the specific climate sensitivity of CESM2 to external forcings (i.e. externally forced signal in CESM2 is different from MEM) outside of the restoring region and, (iii) a mixed between (i) and (ii).
- line 262-263: “The changes in precipitation patterns further affect the atmospheric and oceanic conditions, reinforcing the warming in the Niño3.4 region”. This does not appear on Figure 2, remove? Or at least makes it clear that it is an information capture from the causal network but coming from literature.
- Figures 2, 5, 7 and 8: why not using the same scale for “auto-coeff.” and “link coeff.”? Are those coefficient of different nature? If it is the case, please make it clearer in the method section, if not, I would advice to use only one color bar. Also, please use the same expression in all those figures. In the current state, some figure labels are saying “coeff.” whereas other are saying “corr.” and “strength”.
- line 509: please substitute “two members in each experiment” by “two members in each period”.
- line 509: please substitute “in the four experiments” by “in the four cases”.
- lines 534-535: “small differences, most likely originating from observational uncertainty, ERSSTv5 vs HadISST”. The differences can also be due to the efficiency of the SST restoring, since the SST are not imposed as in atmospheric only simulation but they are nudged.
- lines 535-536: “The pacemaker ensemble mean TNA (red lines in Fig. 6b) implies an important role of ENSO in shaping SSTAs over the Atlantic”. I don’t see the basis for this comment, could you explain why you consider that ENSO play an important role in shaping SSTAs in the Atlantic from Figure 6b?
- line 592-593: “The warming SST trends in the Atlantic favor a strengthened PWC which ultimately cools the SSTs over the Niño3.4 due to enhanced upwelling”. According to P3 in igure 8b, this is happening with a 1-year lag. What physical process can explain that tropical SST anomalies are changing tropical atmospheric circulation 1 year later?
- lines 601-605: how meaningful are those changes? See my main comments 1 and 2.
Citation: https://doi.org/10.5194/egusphere-2023-1861-RC2
Soufiane Karmouche et al.
Soufiane Karmouche et al.
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
127 | 78 | 11 | 216 | 22 | 7 | 7 |
- HTML: 127
- PDF: 78
- XML: 11
- Total: 216
- Supplement: 22
- BibTeX: 7
- EndNote: 7
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1