全球变化知识资源中心

Observations and reanalysis data.

SAT is taken from the Goddard Institute for Space Studies (GISS) climatology^{31, 32}. Recent observed winds and SLP are taken from the interim European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-Interim³³). Winds from earlier in the twentieth century are from the National Oceanic and Atmospheric Administration (NOAA) reanalysis³⁴. The 1992–2011 trade wind trend is robust across various climatologies²⁶. Sea surface height is taken from the Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) data set. SST is from the Hadley Centre climatology³⁵. Various reanalysis products^{36, 37, 38, 39, 40} are used to assess recent interior ocean temperature and circulation trends. The IPO index (Fig. 1) is calculated as per ref. 19.

Wind-forced ocean model experiments.

A fully dynamic ocean circulation model coupled to an atmospheric energy–moisture balance model⁴¹ is used to assess the impact of wind trends on ocean heat uptake, in isolation of atmospheric circulation feedbacks. The ocean component is purposefully configured to match the resolution and parameters of typical models used in the Coupled Model Intercomparison Project Phase 5 (CMIP5). Atmospheric heat and moisture are fully coupled to the other model components, whereas wind fields are prescribed, which has the advantage of allowing us to quantify the impact of Pacific wind trends on ocean heat uptake and cooling of the atmosphere, in isolation of other feedbacks. As such, for this model, single runs are sufficient and no multi-experiment ensemble analysis is required.

After the model is equilibrated with atmospheric CO₂ fixed at pre-industrial, a control experiment is integrated from 1900 to 2012 forced by historical reconstructed greenhouse gas concentrations and other natural and anthropogenic factors, including anthropogenic aerosols (volcanic aerosols are not included). The model’s depth-integrated ocean heat content change in the upper 2000 m from 1990 to 2012 (19 × 10²² J; equivalent to 0.51 W m⁻²) compares well to observed estimates¹⁴ (~16 × 10²² J or 0.43 W m⁻², noting this estimate is considered conservative⁴²). This suggests that the overturning circulation and large-scale mixing of heat into the interior is well reproduced by the model. The model is also able to capture the overall twentieth-century SAT trend, giving confidence that it resolves the broad-scale cycling of heat between the atmosphere, ocean, ice and land surface systems.

The Pacific wind perturbation experiments replicate the control, but there is a linear wind anomaly trend applied between 45° N and 45° S (smoothed to zero between 40–45° N and 40–45° S), beginning in 1992 and reaching a maximum at the end of 2011, based on the observed linear best fit trends in the ERA-Interim monthly wind stress climatology³³ during 1992–2011 (as depicted in Fig. 2a).  Wind speed trends during this time are also applied to the ocean–atmosphere latent and sensible heat fluxes. The direct wind stress impacts are found to dominate the forcing response, however, with wind-altered buoyancy fluxes secondary.

Beyond 2012, the experiments are integrated following the Representative Concentration Pathway (RCP) 6 emission scenario⁴³. In addition, three different future Pacific wind trend scenarios are run from 2012 to 2100, wherein (1) there is a linear ramp-down of the Pacific Ocean winds back to climatological values by 2030, (2) the wind anomalies remain constant from 2012 onwards, and lastly (3) the imposed upward wind trend is continued until 2020, before stabilizing and then remaining constant until the end of the model run.

Wind-forced coupled climate model experiments.

A full coupled climate model^{44, 45} is used to further assess the role of the recent observed Pacific wind trends on the ongoing warming hiatus. The model resolves the interactions between the atmosphere, ocean, land and sea-ice components; although during the wind-forced phase, the ocean component instead ‘feels’ prescribed winds in the Pacific between 45° N–45° S, in all other ways the ocean–atmosphere system freely evolves during the model integration. The model is equilibrated for more than 3,000 years with atmospheric CO₂ fixed at pre-industrial and then integrated during 1780–2030 following historical CO₂ forcing (1780–2000) and then the CMIP3 A2 emissions scenario⁴⁶ from the year 2000 onwards. The model’s overall climate sensitivity is at the low end range of CMIP3 models (reaching 2.1 °C warming by 2090–2099 relative to 1980–1999); however, as our interest is on the cooling impact of the wind-perturbed experiments relative to the control, low climate sensitivity has little bearing on the results. Two sets of 12 ensemble members were generated including full ocean–atmosphere coupling, only with the ocean component forced by: first, seasonally varying climatological model wind stresses (referenced over the period 1980–2012) over the Pacific domain (45° N–45° S); and second, as per the first, only with the additional observed Pacific wind trends applied during 1992–2011. Three future wind trend scenarios are also explored to 2030, as per the ocean–EBM model. Throughout all experiments, the atmosphere is free to respond to the resulting changes at the sea surface. We also conducted similar experiments with the Pacific winds freely evolving. In those experiments, the SAT cooling is even stronger due to the combined effect of both the prescribed wind trends and the enhanced wind anomalies owing to the Bjerknes feedback (figure not shown).

Mid-twentieth-century hiatus experiment.

A final wind anomaly ocean model experiment is assessed over the twentieth century to investigate the role of interdecadal Pacific wind variability in contributing to the 1940–1975 hiatus. In this experiment CO₂ tracks as per the control, but wind anomalies are now derived from a regression of the IPO index¹⁹ onto the ERA-Interim monthly mean wind climatology. Aerosol impacts are purposefully not included in this experiment to isolate the impacts of the mid-twentieth-century IPO evolution separately from the cooling impact of aerosols. As per the 1992–2011 wind trend simulations, these experiments include wind anomalies applied over just the Pacific Ocean (45° S–45° N) and applied to both the air–sea momentum and buoyancy fluxes.

Coupled climate model intercomparison.

We also analysed multi-model ensemble global SAT hindcasts and future projections in climate models that participated in CMIP3 and CMIP5 (Fig. 5a and Supplementary Fig. 3). Furthermore, 48 CMIP5 climate model experiments were assessed in terms of their simulation of Pacific Ocean multidecadal wind trends (Fig. 5b and Supplementary Fig. 9) and how these wind trends relate to decadal global SAT trends (Supplementary Fig. 7). The CMIP5 simulations employ historical anthropogenic and natural forcings until 2005, thereafter the forcing follows RCP4.5 wherever possible; otherwise experiments adopting RCP8.5 are used. As the hiatus decade analysis runs only until 2013 (Supplementary Fig. 7) and the SAT projections to 2020 (Fig. 5a), this choice of scenario is largely arbitrary.