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.