英文摘要: | Reasons for the apparent pause in the rise of global-mean surface air temperature (SAT) after the turn of the century has been a mystery, undermining confidence in climate projections1, 2, 3. Recent climate model simulations indicate this warming hiatus originated from eastern equatorial Pacific cooling4 associated with strengthening of trade winds5. Using a climate model that overrides tropical wind stress anomalies with observations for 1958–2012, we show that decadal-mean anomalies of global SAT referenced to the period 1961–1990 are changed by 0.11, 0.13 and −0.11 °C in the 1980s, 1990s and 2000s, respectively, without variation in human-induced radiative forcing. They account for about 47%, 38% and 27% of the respective temperature change. The dominant wind stress variability consistent with this warming/cooling represents the deceleration/acceleration of the Pacific trade winds, which can be robustly reproduced by atmospheric model simulations forced by observed sea surface temperature excluding anthropogenic warming components. Results indicate that inherent decadal climate variability contributes considerably to the observed global-mean SAT time series, but that its influence on decadal-mean SAT has gradually decreased relative to the rising anthropogenic warming signal.
The change of global-mean SAT during the first decade of the twenty-first century was less than 0.05 °C, indicating a considerably slower rate of warming than during the late twentieth century3, 6. The causes of this global warming hiatus, which are still under debate, can be categorized into either internal or external processes of the climate system. The principal candidates for external drivers of the hiatus are the weakening of solar activity7 and increase in stratospheric aerosols8 plausibly associated with accumulation from minor volcanic eruptions9. However, these effects are quantitatively insufficient to explain the warming hiatus. Indeed, satellite measurements of the top of atmosphere (TOA) radiative budgets for 2001–2010 indicate excess energy of about 0.5 W m−2 received by the Earth10, suggesting that the prime cause of the hiatus is internal to the climate system. Concurrently with the stall of surface warming, despite the energy storage to the system, observational studies have shown evidence that ocean interior warming has occurred continuously11, 12, 13, 14, 15. This indicates the strengthening of global ocean heat uptake16, which acts to increase the ocean temperature below 700 m. Whereas historical climate simulations reproduce neither the warming hiatus nor the strengthening of ocean heat uptake during the past decade3, 16, multi-century control simulations with prescribed pre-industrial radiative conditions do reveal intermittent occurrences of pauses in warming and intensification of heat uptake in phase with the Interdecadal Pacific Oscillation (IPO), an inherent low-frequency variability of the Pacific atmosphere–ocean system17, 18. Modelling evidence that supports the crucial role of the Pacific atmosphere–ocean variability in the hiatus has been provided by numerical experiments of a climate model in which sea surface temperature (SST) was nudged to observations in the eastern equatorial Pacific since 1861. An ensemble of the historical runs reproduced the hiatus remarkably well4. Similarly, prescribing linear trends in tropical wind stresses into a climate model simulated a slowdown of surface warming as well as an increase in heat uptake triggered by the pronounced acceleration of the Pacific trade winds5. The above modelling results not only suggest an origin of the hiatus attributable to prolonged La Niña-like conditions or negative IPO, but also advocate the usefulness of coupled atmosphere–ocean general circulation model (CGCM) experiments partially driven by observed atmosphere–ocean states. One caveat regarding these experiments is that observations that constrain the CGCM are provided a priori, which hampers quantitative separation of the internally generated and externally forced components of climate change. Such attribution studies have been performed instead using Coupled Model Intercomparison Project (CMIP) multi-model ensembles19, 20; however, they cannot reproduce correctly the past interannual and decadal fluctuations3, 16. In this study, we devised a combined modelling approach of using a CGCM forced by observed wind stresses, called partial wind overriding (PWO) experiments, and attribution experiments using an atmospheric general circulation model (AGCM), which provided robust estimates of the extent to which natural decadal fluctuations contributed to the global-mean SAT history over the past five decades, including the period of the global warming hiatus. The model used here is the updated Model for Interdisciplinary Research on Climate version 5 (MIROC5) together with its atmosphere component21. For the PWO experiments, surface wind stress anomalies (τ) in MIROC5 were replaced with daily values derived from the 55-year Japanese Reanalysis (JRA55) data set22 over 30° S–30° N oceans for 1958–2012 (Methods). SST could have been nudged instead, as tropical winds and SST are tied to each other, but the described experiment is advantageous in avoiding artificial heat input into the climate system. A five-member ensemble, called ASYM-H, was branched off from a historical simulation from 1 January 1958. It had slight differences in initial conditions, but identical radiative forcing and other boundary conditions (land use and aerosols) following the CMIP phase 5 (CMIP5; ref. 23) historical and Representative Concentration Pathways 4.5 (RCP4.5) scenario runs before and after 2006. Another set of the five-member ensemble, called ASYM-C, was generated similarly, but with radiative forcing and boundary conditions fixed at the 1850 level. Figure 1 shows the global-mean SAT time series for ASYM-H and ASYM-C, compared with observations and combined simulations of the 46 CMIP phase 3 (CMIP3; ref. 24) and CMIP5 models (Supplementary Table 1). As in the previous SST-nudged experiment4, our ASYM-H run reproduces the observed temperature history well (the correlation is r = 0.89 for the entire period). Its peaks on interannual time scales are found coincident with the occurrence of El Niño and La Niña events, which are not captured by the CMIP simulations by definition. The linear temperature trend for 2003–2012 is slightly negative in the observations, but slightly overestimated in the ASYM-H run, which might be improved were the natural radiative forcing data updated25. A striking feature in Fig. 1 is that ASYM-C generates persistent positive SAT anomalies from the late 1970s to 1999, but negative anomalies beyond; values for the respective decades are 0.11 ± 0.09, 0.13 ± 0.11 and − 0.11 ± 0.09 °C for the 1980s, 1990s and 2000s, suggesting a significant contribution of inherent decadal variability. The cooling seen in ASYM-C is clearly reflected by the hiatus in ASYM-H, and the anthropogenic warming during the 1980s and 1990s is amplified by the imposed forcing in τ. A similar comparison between the SST-nudged historical and control experiments was performed in a previous study4, but, in this work, further quantitative attribution of the history of the global-mean SAT is undertaken. Because there is no input of artificial heat into the system in our PWO experiments, the TOA radiative budget for 2001–2010 is roughly zero in ASYM-C, but it is 0.64 W m−2 in ASYM-H, which is close to the latest satellite estimate10 (Supplementary Fig. 1).
|