全球变化知识资源中心(globalchange): Robust warming projections despite the recent hiatus

The recent slowdown or 'hiatus' in global average surface air temperature (SAT) rise^{1, 2, 3, 4} has been used in some studies as evidence to argue that current models overestimate the climate response to increasing concentrations of greenhouse gases^{5, 6}. Other studies suggest instead that the recent hiatus merely reflects interdecadal variability superimposed on a long-term warming trend^{2, 3, 4, 7, 8, 9}. However, because climate models seem to underestimate the magnitude of observed interdecadal variability⁸, and as this variability may be linked to longer-term sequestration of heat into the deep ocean, the question arises as to what extent future projections need to be re-examined in light of the present hiatus. Here we assess whether twenty-first-century warming projections are altered in any way when considering only simulations that capture a slowdown in global surface warming, as observed since 2000.

We assessed individual global SAT projections in those climate models that participated in the Coupled Model Intercomparison Project Phase 5 (CMIP5), separating the model ensemble members into those that do, and those that do not, capture a slowdown in surface warming during the period 1995–2015. This time window was selected to incorporate the approximate period of the present hiatus (2000–2013), with a slight extension in time to include models whose natural variability is not synchronized precisely with observations. The window did not extend to pre-1995 as the cooling impacts of the Mount Pinatubo eruption dominate the forcing before this time. In principle, the allowable window could have been extended to beyond 2015; however, under high-emissions scenarios, fewer and fewer models capture a hiatus as time progresses through the twenty-first century¹⁰. Although the time window to find model hiatus periods was fixed at 1995–2015, the magnitude of the hiatus SAT trend and the duration of the hiatus, were both varied to test a suite of hiatus criteria in recalculating projections for the end of the twenty-first century. For further details of the methods used and analyses presented, see the Supplementary Information.

Figure 1 shows the results of such an analysis wherein projections are reassessed using only hiatus ensemble members. Although the projected warming distributions are shown for the calendar year 2100 in Fig. 1, these were also recalculated for both the end-of-century decade mean (2091–2100) and multi-decade mean (2081–2100), and robust results were obtained. The hiatus criterion chosen here requires a 14-year period within 1995–2015 when SAT rise is no more than 0.096 °C per decade. This warming rate corresponds to the trend in global average SAT during 2000–2013 in the corrected HadCRUT4 data set¹¹. The hiatus subsampling leaves only 19 (of 90) and 19 (of 108) experiment runs under the IPCC Representative Concentration Pathway (RCP) scenarios RCP8.5 and RCP4.5, respectively (see Fig. 1 insets), with most of these ensemble members appearing at the low end of warming at the end of their respective hiatuses. Indeed, at this stage of the simulations (year 2015), the warming of the subsampled hiatus set is weaker than the all-ensemble distribution; significant at >94% and >99.5% confidence levels, respectively, for the RCP4.5 and RCP8.5 scenarios. Yet the resulting warming projections for the end of the twenty-first century show no significant differences compared to the all-ensemble set, with the mean and distribution under each RCP not significantly altered when excluding the non-hiatus experiments. For example, multi-model mean warming at century's end is hardly changed, reaching 2.49 °C and 4.93 °C for RCP4.5 and RCP8.5, respectively (hiatus experiments), compared to 2.63 °C and 4.99 °C in the all-experiment mean. Thus, the difference in projected warming between the two RCP scenarios is much greater than the difference between the hiatus and all-model ensemble means. For example, global mean projected warming is roughly doubled under RCP8.5 compared with RCP4.5, yet reduced by only < 0.1 °C when just considering the hiatus runs in the high-emissions scenario.