英文摘要: | The slowdown in the rate of global warming in the early 2000s is not evident in the multi-model ensemble average of traditional climate change projection simulations1. However, a number of individual ensemble members from that set of models successfully simulate the early-2000s hiatus when naturally-occurring climate variability involving the Interdecadal Pacific Oscillation (IPO) coincided, by chance, with the observed negative phase of the IPO that contributed to the early-2000s hiatus. If the recent methodology of initialized decadal climate prediction could have been applied in the mid-1990s using the Coupled Model Intercomparison Project Phase 5 multi-models, both the negative phase of the IPO in the early 2000s as well as the hiatus could have been simulated, with the multi-model average performing better than most of the individual models. The loss of predictive skill for six initial years before the mid-1990s points to the need for consistent hindcast skill to establish reliability of an operational decadal climate prediction system.
Traditional free-running climate simulations that start in the mid-nineteenth century and proceed through the twentieth century with observed human-produced forcings, such as increasing greenhouse gases (GHGs), aerosols and ozone, along with natural forcings, such as aerosols from volcanic eruptions and solar variability, are designed to simulate the response of the climate system to those changes in external forcings. To do this, multiple realizations or ensemble members are run with each model. These are then averaged together to remove the effects of naturally occurring interannual and decadal timescale variability, leaving only the response to the external forcings. If the early-2000s hiatus is mostly a result of internally generated climate variability2, 3, 4, 5, the average of all those simulations for the early 21st century would, and indeed does, lie above the actual plateau of warming that occurred in the observations1, 6. Furthermore, the models could be overly sensitive to increasing GHGs (ref. 7), and there could have been contributions from a collection of moderate volcanic eruptions or other forcings8, 9, 10, suggesting that the forcings specified in the Coupled Model Intercomparison Project Phase 5 (CMIP5) experiments may not have been adequate to simulate all aspects of the early-2000s hiatus. But the fact that all model simulations, when averaged together, do not simulate the hiatus has been touted as a failure of any model to simulate what actually occurred in the early-2000s11, 12. However, inspection of the individual ensemble members from these same model simulations reveals that ten members actually produced the observed warming trend (defined as a trend less than 0.04 °C per decade as observed) during the period of the hiatus 2000–2013 (Fig. 1a and refs 4, 13). A composite of those ten ensemble members out of 262 possible CMIP5 realizations (Methods) shows a negative phase of the IPO, characterized by cooler-than-normal average surface temperatures over the tropical Pacific, with opposite sign anomalies in the northwest and southwest Pacific, lasting 14 years (Fig. 1b and Supplementary Fig. 1). There are 21 ensemble members that simulate a hiatus from 2000 to 2012—nine continue through 2000–2014, six from 2000 to 2015, and six from 2000 to 2016, one of which from 2000 to 2017 continues to 2018 (a hiatus of 19 years). Average hiatus composites have a negative IPO phase, as opposed to the overall average of a larger set of ensemble members showing mostly warming in the tropical Pacific14. Thus, although not specifically designed to do so, in some of the uninitialized simulations the internally generated variability associated with the IPO happens to synchronize with the phase of naturally occurring variability in the observations purely by chance. The pattern correlation of the observed IPO (Supplementary Fig. 1b) with the surface temperature trends from 2000 to 2013 in all ensemble members shows a roughly Gaussian distribution around zero pattern correlation as the internally generated variability is more or less random (Fig. 1c). The same quantity from the hiatus ensemble members shows a shift of the distribution towards statistically significant positive values greater than + 0.4 (Methods), indicating that internally generated variability with a negative IPO is tending, on average, to sync up with what happened in the observations in those members (Supplementary Fig. 1a and Fig. 1b, c). This is a compelling application of the result derived from other analyses, in that tropical Pacific surface temperatures in the negative phase of the naturally-occurring IPO can temporarily counteract the warming from increasing GHGs to produce a hiatus of warming in globally averaged surface air temperatures that can last for a decade or more2, 3, 4, 5, even as the climate system is still trapping excess heat of about 0.5–1.0 W m−2 (refs 15, 16).
|