3 Results
3.1 Historical period
EBM2 captures the increasing temperature trend during the twentieth
century and distinguishes between high and low climate sensitivity
AOGCMs (Figure 1). In all EBM2 emulations, a proportion of the RMSE
(~ 0.07 K) arises from interannual variations in the
AOGCM ensemble means that is not captured in the emulations (there are
three members in each AOGCM historical ensemble). The performance of
EBM2, however, varies substantially between AOGCMs. There are instances
of both large and small RMSE emulations for both high and low climate
sensitivity AOGCMs. For AOGCMs where there are substantial differences
between the emulations and the AOGCM projections, the differences occur
over different time periods. Differences are large for 1925-1950
(HadGEM3-GC31-LL), for 1950-1975 (NorESM2-LM) and for 2000-2015
(HadGEM3-GC31-LL, IPSL-CM6A-LR, GFDL-ESM4 and NorESM2-LM). For
IPSL-CM6A-LR, temperatures are overestimated by the emulators throughout
1915-2014. Intriguingly, close emulation of temperatures in abrupt-4xCO2
does not guarantee close emulation for the historical period (e.g.
GFDL-ESM4), and a relatively poor emulation of abrupt-4xCO2 does not
prohibit close emulation for the historical period (e.g. CNRM-CM6-1)
(Figure S1).
The step model produces emulations with RMSEs equivalent to or less than
emulations from EBM2 in seven of the eight AOGCMs. The exception is
NorESM2-LM which has relatively large inter-annual variability and is
the only model to show an apparent cooling trend during years 20-50 of
its abrupt-4xCO2 simulation (Figure S1).
EBM3 performs better than EBM2 for abrupt-4xCO2, which is expected given
the additional timescales resolved by the third layer. The additional
degrees of freedom enable a much closer emulation of temperatures during
years 10-40 of the abrupt-4xCO2 experiment, a period when the rate of
temperature increase weakens rapidly (Figure S1). However, the
improvement of EBM3 over EBM2 in the abrupt-4xCO2 experiment does not
consistently translate to the historical experiment. Indeed there are
two AOGCMs for which EBM2 has smaller RMSEs than EBM3 (MIROC6 and
IPSL-CM6A-LR). Both EBMs overestimate temperatures for 1990-2014 in four
of the eight AOGCMs and generally produce larger RMSEs than the step
model.