Arising from A. Mamalakis et al. Nature https://doi.org/10.1038/s41467-018-04722-7 (2018)
In their recent study1, (hereafter M18) propose a new interhemispheric teleconnection linking late austral winter sea surface temperature (SST) anomalies in the New Zealand region to Southwestern US winter precipitation. They propose that warm SST anomalies propagate from the New Zealand region into the Northern Hemisphere through an atmospheric bridge mechanism (from approximately July to October) which in subsequent months (from November to March) acts to alter properties of the jet stream that influence Southwestern US winter precipitation variability. However, in this correspondence, after accounting for non-stationarity, memory and shared variability with other well-known modes of internal variability, we show a large reduction in correlation strength between SST anomalies in regions key to the proposed mechanism. Other important claims in M18 regarding the physical mechanisms involved, whereby SST anomalies in the New Zealand region enhance remote adiabatic warming and reduce cloud cover in key regions, are not robustly supported when examined in reanalysis and satellite data.
A critical aspect of the newly proposed teleconnection in M18 depends on SSTs in the New Zealand (NZI) region leading, and being a source of predictability for, western tropical Pacific SSTs in a region east of the Philippines (EPH). To reduce the likelihood of spurious correlation when searching for new teleconnections, it is crucial that any well-known confounders (i.e. other variables or processes that jointly influence SSTs in both regions) are given careful consideration. First, reasonably large trends in SSTs are known to exist in both of these regions over the time period considered in M18 (e.g. see ref. 2) (see also Supplementary Fig. 1). Second, the El Niño-Southern Oscillation (ENSO) and related oscillations, including the Southern Oscillation Index (SOI), are well known to influence SST anomalies and atmospheric circulation in both of these regions3,4 (see also Supplementary Fig. 1). Another important consideration when assessing predictability lies within the Granger causality paradigm5,6): that causation depends on the predictive strength of the independent variable (here SSTs in NZI) significantly exceeding the predictive strength of past values (i.e. memory) of the dependent variable alone (here SSTs in EPH). Lastly, it is well established that simultaneously conducting multiple hypothesis tests ensures a higher standard (i.e. lower p-values) for rejecting local null hypotheses7. The influence of each of these considerations (non-stationarity, memory, other confounders and significance testing) is examined in detail below.
The correlation strength between SSTs in NZI and EPH as a function of NZI lead time in months over the period 1982–2015 is shown in Fig. 1 (Fig. 1a is equivalent to Fig. 7 in M18). As discussed in M18, the strongest of these correlations (r = 0.85) occurs for NZI September SST anomalies leading EPH by 3–4 months and with other months/lags also displaying high correlations (r > 0.7) related to the atmospheric bridging mechanism proposed in M18 (i.e. NZI July to October SST anomalies leading EPH by 1–5 months). When the anomalies from both timeseries are linearly detrended before computing the correlation (Fig. 1b) the correlation strength during these important months/lags is reduced somewhat to the range of r = 0.52–0.76 but remains strongly statistically significant. However, EPH SST anomalies (after detrending) also display considerable memory (Fig. 1c) and in many months/lags display greater correlation strength than NZI as a predictor (comparing Fig. 1c with 1b). Furthermore, SOI is found to be very well correlated with EPH over the same months/lags of interest (comparing Fig. 1d with 1b) highlighting an important source of predictability for EPH SST anomalies that is also well correlated with SST anomalies in the NZI region (Supplementary Fig. 1).
To more directly diagnose the relative influence of SOI and EPH memory in these associations, we present results of partial correlation analysis (e.g. ref. 8,9). When the influence from EPH memory is accounted for through partial correlation (Fig. 1e), the association between NZI leading EPH is substantially reduced in all months. Over key month/lags (i.e. NZI July to October SST anomalies leading EPH by 1–5 months) the correlation strength has reduced from r = 0.7–0.85 to r = 0.15–0.58 indicating that a considerable amount of the original association can be explained by EPH memory alone. When the influence from SOI is accounted for through partial correlation (Fig. 1f) we find that none of the NZI leading EPH correlations remain statistically significant (after accounting for multiple hypothesis testing) and the correlation strength over key month/lags has reduced considerably from r = 0.7–0.85 to r = 0.22–0.51. We also found that when the proposed causal pathway is analyzed in the reverse direction (i.e. predicting NZI SST anomalies from previous EPH SST anomalies and controlling for SOI through partial correlation, Fig. 1g) the strength of association is often larger (compared to Fig. 1f) and now displays statistical significance in some months/lags. In summary, these results indicate that the strong lagged associations for NZI leading EPH SST anomalies reported in M18 are inflated by already well-known processes and variables (not accounted for in M18) that jointly influence SST anomalies in these regions.
M18 suggested that the apparent association between NZI and EPH SSTs over the period 1982–2015 (but not before 1982) may be due to recent tropical expansion under anthropogenic forcing or due to poor data quality prior to 1982 masking the true correlation strength. To shed light on these considerations we analyzed multiple runs from the fully coupled CESM LENS historical experiment (see ref. 10. for details). Each run in the ensemble has the same historical forcings and parametrizations but slightly perturbed initial conditions. We found that many runs (e.g. Supplementary Fig. 2) display very weak lagged correlations (e.g. r = 0.2) over relevant months/lags but can strengthen (r > 0.5) and become strongly statistically significant in selected model runs (e.g. Supplementary Fig. 3). Since these runs differ only in terms of initial conditions, this suggests that any apparent lagged correlations found between these regions are not likely a consistent or emerging feature of the climate system, but instead occasionally arise due to stochastic internal climate variability.
Lastly, we examine the proposed physical mechanism in M18 detailing how SST anomalies in the NZI region might propagate north to the EPH region. According to M18, this occurs when warm SST anomalies in the NZI region are sufficient in magnitude to induce an anomalous Hadley circulation resulting in descending motion over EPH, suppressed cloud formation and positive SST anomalies. By analyzing ERA-Interim vertical velocity at the 500-hPa level during the 5 warmest NZI years (as in Fig. 6 of MH18), we show that strong ascending motion remains apparent over the EPH region with no consistent reduction in strength of ascending motion (Fig. 2a, d, g). There is also no consistent reduction in total cloud cover (combined day and night) over the EPH region compared to climatology (Fig. 2b, e, h); instead total cloud cover slightly increased during NZI warm years. This result is supported in CERES EBAF satellite data (Fig. 2c, f, i) in the years of available data (2000-present).
In summary, a critical underpinning for the proposed teleconnection in M18 depends on an atmospheric bridge mechanism whereby SST anomalies propagate from NZI to EPH beginning in austral winter. However, we have shown that after accounting for significant non-stationarity, memory, multiple hypothesis testing, and other confounders there is currently little observational evidence that NZI SST anomalies offer a unique source of predictability for EPH SST anomalies. Second, using the CESM LENS model ensemble, we have illustrated that apparent lagged associations of the magnitude reported in M18 can occasionally arise from stochastic internal variability. Lastly, other mechanistic claims in M18 for how SST anomalies propagate between these key regions, are not robustly supported in reanalysis or satellite data. In combination, we argue that the physical basis for the proposed new teleconnection, derived solely from observations in M18, remains unclear. Without this, there is no indication that the proposed NZI index will be a useful source of predictability in future years (i.e. out-of-sample) for Southwestern US winter precipitation.
All data used in this study are freely availability. SOI data used were from: ftp://ftp.bom.gov.au/anon/home/ncc/www/sco/soi/soiplaintext.html provided by the Australian Bureau of Meteorology. Observational SST data were from JRA COBE-SST2 from: https://www.esrl.noaa.gov/psd/data/gridded/data.cobe2.html. CESM LENS SST data were obtained from http://www.cesm.ucar.edu/projects/community-projects/LENS/data-sets.html. ERA-Interim omega and tcc data were from: http://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc. CERES EBAF v4 tcc data were from https://doi.org/10.5067/TERRA+AQUA/CERES/EBAF-TOA_L3B.004.0.
Mamalakis, A., Yu, J. Y., Randerson, J. T., AghaKouchak, A. & Foufoula-Georgiou, E. A new interhemispheric teleconnection increases predictability of winter precipitation in southwestern US. Nat. Commun. 9, 2332 (2018).
Purich, A. et al. Tropical Pacific SST drivers of recent Antarctic sea ice trends. J. Clim. 29, 8931–8948 (2016).
Chen, X. & Wallace, J. M. ENSO-like variability: 1900–2013. J. Clim. 28, 9623–9641 (2015).
Trenberth, K. E. & Caron, J. M. The Southern Oscillation revisited: Sea level pressures, surface temperatures, and precipitation. J. Clim. 13, 4358–4365 (2000).
Granger, C. W. J. Investigating causal relations by econometric models and cross-spectral methods. Econmetrica 37, 424–438 (1969).
McGraw, M. C. & Barnes, E. A. Memory matters: a case for Granger causality in climate variability studies. J. Clim. 31, 3289–3300 (2018).
Wilks, D. S. “The stippling shows statistically significant grid points”: how research results are routinely overstated and overinterpreted, and what to do about it. Bull. Am. Meteorol. Soc. 97, 2263–2273 (2016).
Cai, W., Van Rensch, P., Cowan, T. & Hendon, H. H. Teleconnection pathways of ENSO and the IOD and the mechanisms for impacts on Australian rainfall. J. Clim. 24, 3910–3923 (2011).
Hendon, H. H., Wheeler, M. C. & Zhang, C. Seasonal dependence of the MJO–ENSO relationship. J. Clim. 20, 531–543 (2007).
Kay, J. E. et al. The Community Earth System Model (CESM) large ensemble project: a community resource for studying climate change in the presence of internal climate variability. Bull. Am. Meteorol. Soc. 96, 1333–1349 (2015).
The research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. Copyright 2018 California Institute of Technology. U.S. Government sponsorship acknowledged
The authors declare no competing interests.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Gibson, P.B., Waliser, D.E. & DeFlorio, M.J. A critical examination of a newly proposed interhemispheric teleconnection to Southwestern US winter precipitation. Nat Commun 10, 2687 (2019). https://doi.org/10.1038/s41467-019-10528-y