Xu Deng^1,2* and Sarah E. Perkins-Kirkpatrick^1,2

¹School of Science, University of New South Wales, Canberra, ACT, Australia.

²ARC Centre of Excellence for Climate Extremes, University of New South Wales, Canberra, ACT, Australia.

Corresponding author: Xu Deng (xu.deng@student.adfa.edu.au)

Key Points:

Abstract

This study focuses on the projections and time of emergence (TOE) for temperature extremes over Australian regions in the phase 6 of Coupled Model Intercomparison Project (CMIP6) models. The model outputs are based on the Shared Socioeconomic Pathways (SSPs) from the Tier 1 experiments (i.e., SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5) in the Scenario Model Intercomparison Project (ScenarioMIP), which is compared with the Representative Concentration Pathways (RCPs) in CMIP5 (i.e., RCP2.6, RCP4.5 and RCP8.5). Furthermore, two large ensembles (LEs) in CMIP6 are used to investigate the effects of internal variability on the projected changes and TOE. As shown in the temporal evolution and spatial distribution, the strongest warming levels are projected under the highest future scenario and the changes for some extremes follow a “warm-get-warmer” pattern over Australia. Over subregions, tropical Australia usually shows the highest warming. Compared to the RCPs in CMIP5, the multi-model medians in SSPs are higher for some indices and commonly exhibit wider spreads, likely related to the different forcings and higher climate sensitivity in a subset of the CMIP6 models. Based on a signal-to-noise framework, we confirm that the emergence patterns differ greatly for different extreme indices and the large uncertainty in TOE can result from the inter-model ranges of both signal and noise, for which internal variability contributes to the determination of the signal. We further demonstrate that the internally-generated variations influence the noise. Our findings can provide useful information for mitigation strategies and adaptation planning over Australia.

1 Introduction

Anthropogenic climate change will lead to more severe temperature extremes, which have significant impacts on society and natural systems (Intergovernmental Panel on Climate Change, 2021). To assess possible climate futures, projections by global climate models from the Scenario Model Intercomparison Project (ScenarioMIP; O’Neill et al., 2016) as part of the Coupled Model Intercomparison Project phase 6 (CMIP6; Eyring et al., 2016) are useful resources, and may provide new insights into how temperature extremes are projected to change under climate change (e.g., Alexander & Arblaster, 2017; Grose et al., 2020; Sillmann, Kharin, Zwiers, et al., 2013; Thibeault & Seth, 2014).

Over Australia, Alexander and Arblaster (2017) indicated that significant increases (decreases) are projected for the occurrence of warm (cold) extremes by the end of this century under the intermediate- and highest-emission scenarios in CMIP5, and that these changes are most distinct in the tropics. Compared to 29 CMIP5 models, Grose et al. (2020) documented that projected changes in temperature extremes over Australia are more distinct and span narrower ranges in seven CMIP6 models. However, the smaller number of models used in this study may lead to misleading conclusions. Recently, Tebaldi et al. (2021) demonstrated that the CMIP6 ensemble projects higher warming and larger spread for global mean temperature compared with CMIP5, which could result from both a wider range of radiative forcing and higher climate sensitivity in a subset of CMIP6 models. In the present study, to obtain a more reasonable comparison with CMIP5, more models are included in the CMIP6 ensemble to analyze the projected changes of temperature extremes over Australia.

In addition, detecting the time of emergence (TOE) for extremes over Australia needs investigation. TOE is defined as the time when the externally forced climate signal (i.e., forced response) emerges from the noise (i.e., natural variability), suggesting that a significant change is detected and a novel climate regime become evident (e.g., Hawkins et al., 2020; Hawkins & Sutton, 2012; King, Donat, et al., 2015). Estimating TOE can provide insights for mitigation strategies, adaptation planning and scientific community, as the forced response relative to the background noise may be more relevant for the assessment of climate impacts, compared to the absolute change (Beaumont et al., 2011; Deutsch et al., 2008; Hawkins et al., 2020; Hawkins & Sutton, 2012; Ossó et al., 2021). For example, similar absolute changes in extreme temperature can result in different ecological impacts since extratropical ecosystems are usually more resilient than tropical ecosystems, as they are adapted to a more variable climate (Beaumont et al., 2011; Deutsch et al., 2008).

Previous studies have concluded that for mean temperature there is earlier TOE over tropical regions than that in the extratropics where the noise is generally larger (e.g., Giorgi & Bi, 2009; Hawkins et al., 2020; Hawkins & Sutton, 2012; Mahlstein et al., 2012; Mahlstein et al., 2011). Furthermore, for warm and cold extremes that display larger variability, the signals for these indices tend to emerge later over both the tropics and extratropics (e.g., King, Donat, et al., 2015; Tan et al., 2018) relative to mean temperature. Currently, most studies on TOE have been conducted at global levels, with less detailed analyses over smaller-scale regions (e.g., Batibeniz et al., 2020; Gaetani et al., 2020; Ossó et al., 2021), especially for Australia (King, Donat, et al., 2015). Under different future scenarios, we aim to investigate the TOE of extreme temperatures over Australia at the subregional scale.

A variety of methods have been used in TOE assessment, which can lead to a source of uncertainty (Abatzoglou et al., 2019; Gaetani et al., 2020). A recent study (Gaetani et al., 2020) found that compared to Kolmogorov-Smirnov (KS) non-parametric test (King, Donat, et al., 2015), the signal-to-noise ratio (SNR) frameworks exhibit increased uncertainty and later times for TOE over West Africa (Gaetani et al., 2020). However, the SNR methods facilitate the separation between signal and noise, and identifying both components and their interaction physically (e.g., slow-varying ocean conditions and the modes of internal variability) can deepen our understanding in climate change (e.g., Barnes et al., 2019; Barsugli & Battisti, 1998). In this study, we adopt the method by Hawkins and Sutton (2012) and Hawkins et al. (2020) to address the TOE assessment, which is widely used and allows more cross-study comparisons (e.g., Abatzoglou et al., 2019; Gaetani et al., 2020; Hawkins et al., 2020; Hawkins & Sutton, 2012; Ossó et al., 2021). For the uncertainty in the detection of TOE in this method, it can arise from inter-model spread not only in the signal, but also from noise (Hawkins & Sutton, 2012).

Furthermore, as internal variability can also be an important source of uncertainty for regional climate (Dai & Bloecker, 2019; Deser, Knutti, et al., 2012; Deser, Phillips, et al., 2012; Hawkins & Sutton, 2009; Lehner et al., 2020), single-model initial-condition large ensembles (SMILEs; hereafter LEs) are an important tool to investigate the consequences of the intrinsic variability on the uncertainty in projected changes and TOE of extreme temperatures over Australia, of which external forcing and model structure are identical among the members (e.g., Dai & Bloecker, 2019; Deser, 2020; Deser et al., 2020; Lehner et al., 2020; Mankin et al., 2020; Perkins-Kirkpatrick et al., 2017; Xie et al., 2015).

Previous research evaluated the ability of CMIP6 models to simulate extreme temperatures over Australian regions in the historical period (1950-2014), compared these results to the CMIP5 ensemble, and investigated the effects of internal variability on the corresponding trends based on the LEs in CMIP6 (Deng et al., 2021). Following from this research, the purposes of this study are: to assess future climate changes of the extremes and the TOE over Australian regions in both the CMIP6 and CMIP5 models, and to explore the effects of internal variability on the projected changes and TOE based on LEs in CMIP6.

2 Data and Methods

2.1 Model Data

Although the scenarios in the ScenarioMIP consist of two tiers, we only use the Tier 1 experiments based on the Shared Socioeconomic Pathway (SSP) scenarios: SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, as these sample a varying range of possible emission futures and contain relatively large number of model outputs. Among them, SSP1-2.6, SSP2-4.5 and SSP5-8.5 indicate the same nominal stratospheric-adjusted radiative forcing (2.6, 4.5 and 8.5 W m⁻²) reached in 2100, compared to the scenarios based on Representative Concentration Pathways (RCPs) used in CMIP5 (i.e., RCP2.6, RCP4.5 and RCP8.5); and SSP3-7.0 fills a gap between medium and high end in the range of future forcing pathways, not included in previous CMIP generations (O’Neill et al., 2016; Tebaldi et al., 2021). Despite the similarity among the future scenarios in CMIP6 and CMIP5, it is noted that there are some differences, such as the composition of some radiatively active gases or species (e.g., CO₂ and CH₄) and aerosol emissions, making the resulting effective radiative forcing (ERF) different (Lurton et al., 2020; Riahi et al., 2017; Tebaldi et al., 2021).

As one aim of this study is to compare the two CMIP ensembles in projected changes and TOE in extremes, we do not consider the interdependence among the models and use emergent constraints or any other ways of model weighting to reduce the differences between CMIP6 and CMIP5 (e.g., Tokarska et al., 2020), which is similar to the practice by Seneviratne and Hauser (2020). Similar to Deng et al. (2021), only one ensemble member (typically the first member) in each model is considered for the main part of analysis. There are 25 models in CMIP6 and 26 models in CMIP5 for at least one of the future scenarios. In addition, two LEs under SSP5-8.5 and SSP1-2.6 in CMIP6 are used to investigate the impacts of internal variability on the projected changes and TOE of the extremes: CanESM5-LE and MIROC6-LE, which contain 25 members and 50 members, respectively. Detailed information on the simulations from CMIP6 and CMIP5 models are listed in the Tables S1 and S2, respectively.

2.2 Temperature indices

As in Deng et al. (2021), based on daily maximum and minimum temperatures (TX and TN), the annualized temperature extremes defined by the Expert Team on Climate Change Detection and Indices (ETCCDI; Zhang et al., 2011) are used, which forms a continuous and comprehensive investigation of changes in extremes, similar to other studies for CMIP5 (e.g., Alexander & Arblaster, 2017; Sillmann, Kharin, Zhang, et al., 2013; Sillmann, Kharin, Zwiers, et al., 2013; Thibeault & Seth, 2014). Besides diurnal temperature range (DTR), other extreme indices for temperatures are classified into four categories: absolute indices (hottest day [TXx], coldest day [TXn], warmest night [TNx] and coldest night [TNn]), threshold indices (summer days [SU], tropical nights [TR] and frost days [FD]), percentile-based indices (warm days [TX90p], cold days [TX10p], warm nights [TN90p] and cold nights [TN10p]), and duration indices (warm spell duration index [WSDI] and cold spell duration index [CSDI]). The bootstrap resampling procedure by Zhang et al. (2005) is applied to the percentile-based and duration indices, among which the spells crossing year boundaries are taken into consideration for WSDI and CSDI. Since the definitions of growing season length (GSL) and ice days (ID) are not suitable over most of Australia (Alexander & Arblaster, 2017), we do not use them in this study. Detailed information on the indices can be found in Table S3.

2.3 Time of Emergence

The TOE is determined using the signal-to-noise framework as detailed by Hawkins and Sutton (2012) and Hawkins et al. (2020), which is considered as the first year when the signal-to-noise ratio (SNR) is larger than nominated thresholds (e.g., 1 and 2). As suggested by Frame et al. (2017), we consider SNR=1 as the threshold for an “unusual” climate and SNR=2 as “unfamiliar”. This approach linearly regresses annual local variations in temperature extremes onto global mean surface temperature change (\(GMST\)), relative to the base period:

where \(\hat{L}\left(t\right)\ \)represents the regressed\(L\left(t\right)\), denoting annual local changes in extremes over time; \(G\left(t\right)\) is a smoothed version of \(GMST\) over the same period; \(\alpha\) defines the linear scaling between\(\hat{L}\left(t\right)\ \)and \(G\left(t\right)\); and \(\beta\) is a constant. \(GMST\) is smoothed with a “Locally Weighted Scatterplot Smoothing” filter (LOWESS; Cleveland, 1979) of 21 years, which filters out interannual variability (though retaining multi-decadal variability). The signal of local climate change described by \(GMST\)is \(\text{αG}\left(t\right)\), and the noise is defined as the standard deviation of the residuals (\(L\left(t\right)\) –\(\text{αG}\left(t\right)\)). The method implies that local variations for some variables scale well with \(GMST\) (Fischer et al., 2014; Seneviratne & Hauser, 2020; Sutton et al., 2015). It is also noted that internal variability can contribute to the determination of signal, which may introduce further uncertainty in the estimate of TOE (Gaetani et al., 2020; Kumar & Ganguly, 2018; Lehner et al., 2020).

To compare observed SNR with the simulations, Berkeley Earth Surface Temperatures (BEST; Rohde, Muller, Jacobsen, Muller, et al., 2013; Rohde, Muller, Jacobsen, Perlmutter, et al., 2013) is used in this study. Although TN in BEST is biased over Australia (Deng et al., 2021), the TX and TN in BEST show higher correlation compared to Australian gridded climate data (AGCD, previously termed Australian Water Availability Project [AWAP]; Jones et al., 2009), which is better than other global datasets, including National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis 1 (NCEP1; Kalnay et al., 1996), NCEP/Department of Energy (DOE) Reanalysis 2 (NCEP2; Kanamitsu et al., 2002), Twentieth Century Reanalysis (20CR; Compo et al., 2011), and European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis version 5 (ERA5) with preliminary extension to 1950 (Bell et al., 2021; Hersbach et al., 2020) (not shown).

2.4 Regional Assessment

According to climatological and geographical conditions (Perkins et al., 2014; http://www.bom.gov.au/climate/change/about/temp_timeseries.shtml), Australia is divided into nine sub-regions: AUS (Australia), NA (Northern Australia), SA (Southern Australia), SEA (South East Australia), MEA (Middle Eastern Australia), TA (Tropical Australia), SWA (South West Australia), SSA (Southern South Australia), CAU (Central Australia), and MWA (Mid-Western Australia), shown in Table S4 and Fig. S1, which allows a detailed assessment over smaller subregions. And the base period is from 1961 to 1990, which is commonly used and allows us to analyze TOE with respect to a recent period. Still, we regrid TX and TN to 1° × 1° resolution using bilinear interpolation, and then calculate extreme indices. In addition, grid boxes containing less than 75% land are masked out (King, van Oldenborgh, et al., 2015).

In the next section, temporal variations from 1950 to 2100 for the ETCCDI indices in different future scenarios are first analyzed, followed by the spatial patterns of the changes in the indices over 2071-2011 and 2031-2060. Then, the SNR and TOE for TXx and TNn is calculated to address when a novel climate for temperature extremes emerges. For consistency among CMIP6, CMIP5 and BEST, we calculate the noise in SNR for the period 1950-2005, as the estimation of noise can stabilize over longer timescale (Dai & Bloecker, 2019; Santer et al., 2011). Finally, we use two LEs to check the effects of internal variability on the projected responses of extremes and TOE.

3 Results

3.1 Projected changes

Relative to the base period 1961-1990, Figs. 1 and 2 indicate time series of the anomalies for the 14 ETCCDI indices averaged over Australia (10-45°S, 110-155°E) during the period 1950-2100 under different future scenarios in CMIP6 (SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5) and CMIP5 (RCP2.6, RCP4.5 and RCP8.5). For the multi-model medians (Fig. 1), consistent with RCPs in CMIP5 (Fig. 2), the Tier 1 experiments in ScenarioMIP show projected increases in the absolute indices (TXx, TXn, TNx and TNn) and in the warm extremes for percentile-based, duration and threshold indices (TX90p, TN90p, WSDI, SU and TR); in contrast, there are decreases in other cold extremes (TX10p, TN10p, CSDI and FD).

Among the scenarios, the indices under SSP5-8.5 and RCP8.5 generally show larger warming evolution, especially by the end of the century. Moreover, except for DTR, CSDI and FD (Fig. 1e, k and n), extremes under the SSP3-7.0 fill the gap between SSP2-4.5&RCP4.5 and SSP5-8.5&RCP8.5. For example, in the year 2100, the median of TXx under SSP3-7.0 is 4.58°C, lower than 5.78°C&5.82°C in SSP5-8.5&RCP8.5 and higher than 3.24°C&2.67°C in SSP2-4.5&RCP4.5. In the lower emission scenarios (SSP1-2.6&RCP2.6) there is a stabilization for the extremes in the second half of 21^st century, achieving lowest warming (e.g., 2.23°C&1.92°C for TXx in 2100). This result implies the benefits of mitigation strategies associated with these scenarios (O’Neill et al., 2016). However, the separation for the adjacent pathways (e.g., SSP5-8.5&SSP3-7.0, SSP3-7.0&SSP2-4.5 and SSP2-4.5&SSP1-2.6) usually occurs after 2060s for most indices over Australia. In particular, compared to SSP5-8.5&RCP8.5, if a more aggressive mitigation policy is undertaken (e.g., SSP1-2.6&RCP2.6), it may still take one or two decades to notice its effects on projected changes in temperature extremes over Australia.