Figure 7. Signal-to-noise ratio (SNR) in the year 2005 for
temperature extremes in BEST, CMIP6 and CMIP5. (a) SNR in TXx for BEST;
(b) SNR in TXx for the multi-model medians in CMIP6; and (c) SNR in TXx
for the multi-model medians in CMIP5. (d-f) Same as (a-c), but for TNn.
As spatial aggregation or averaging may reduce the impact of internal
variability (Deser, Knutti, et al., 2012; Hawkins & Sutton, 2009;
Lehner et al., 2020), Figs. 8 and 9 show the times series (1950-2100) of
SNR for TXx and TNn, which are averaged over each region before the
calculation of SNR (the corresponding signal and noise are in the
supplementary Figs. S18-S20). For the temporal variations of median SNR
over the period 1950-2014, the signal and SNR for TXx in BEST can be
within the spread of the two CMIP ensembles over some regions (Fig. 8
and Fig. S18). However, for TNn the signal and SNR are usually outside
the ranges of CMIP6 and CMIP5 at the beginning of this century (Fig. 9
and Fig. S19). Despite the influence of observational uncertainty in
BEST over Australia (Deng et al., 2021), the above results suggest that
the differences between the observed and simulated signal and SNR are
mostly related to internal variability (Dai & Bloecker, 2019). In the
study by Dai and Bloecker (2019), they concluded that comparing the
trends of the observed and modelled precipitation (a variable also
exhibiting relatively large variability), which can represent the signal
in some studies (e.g., Gaetani et al., 2020), is not appropriate over
short timescales and at local and regional scales, as the observed
precipitation changes are still dominated by internal variability.