1. Introduction
Weather and climate extremes like heat waves, heavy rainfall, and droughts exert devastating impacts on human society including health, food, infrastructure, and the economy (e.g., Martinich and Crimmins 2019; Son et al. 2019; Vogel et al. 2019). More frequent and severe extreme events have been occurring across the globe in line with accelerating global warming (e.g., Seneviratne et al. 2014; Du et al. 2019; Myhre et al. 2019). Given the continued warming projected in the future (Kharin et al. 2013), larger areas are expected to be affected by stronger extreme events (Coumou and Robinson 2013; Zhang et al. 2018). Thus, it is of fundamental importance to quantify the anthropogenic and natural contributions to the observed changes in climate extremes to produce more reliable future projections and inform appropriate adaptation strategies (Gillett et al. 2016).
During the past decade, there have been many detection and attribution (D&A) studies of the long-term changes in extreme temperatures at global and regional scales based on quantitative comparisons between observations and global climate model simulations (e.g., Zwiers et al. 2011; Christidis et al. 2011; Min et al. 2013; Morak et al. 2013; Kim et al. 2016; Christidis and Stott 2016; Lu et al. 2018; Yin and Sun 2018). Zwiers et al. (2011) compared observed (HadEX; Alexander et al. 2006) and multimodel simulated [phase 3 of the Coupled Model Intercomparison Project (CMIP3); Meehl et al. 2007] changes in coldest night and day (TNn and TXn) and warmest night and day (TNx and TXx) during 1961–2000 using a method based on generalized extreme value (GEV) distributions. They detected anthropogenic influence in all four indices over the global land and many continental areas. Christidis et al. (2011) detected anthropogenic signals in the warmest daily temperatures during 1951–2000 by applying an optimal detection method (which maximizes the signal-to-noise ratio based on a multiple linear regression; e.g., Allen and Stott 2003) to HadGHCND observations (Caesar et al. 2006) and HadCM3 model simulations, and also showed that the detected human influence is separable from the effects of natural forcing (solar and volcanic activities) as well as internal climate variability. Min et al. (2013) also conducted an optimal detection analysis using HadEX observations and CMIP3 multimodel runs. They identified robust human influences on the observed changes in all four indices during 1951–2000 at global and northern continental regions, in separation from natural influences. They also showed the first isolation of human influences on extreme temperature intensity at subcontinental scales. Morak et al. (2013) found detectable changes in the extreme temperature frequency indices on global and regional scales by comparing observed (HadEX) and simulated (HadGEM1 model) changes. Building on these studies, the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) concluded that “it is very likely that human influence has contributed to the observed changes in the frequency and intensity of daily temperature extremes on the global scale since the mid-20th century” (Bindoff et al. 2013).
Since AR5, there have been a few large-scale D&A studies for an extended period up to 2010. By comparing HadEX2 observations (Donat et al. 2013a) and multimodel simulations from phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) for 1951–2010, Kim et al. (2016) found an improved detection (i.e., more frequent occurrence) of anthropogenic signals, particularly in “warm extremes” (TNx and TXx, representing warmest night and day at given location for each year; Zhang et al. 2011) and also showed an improved observation–model agreement in “cold extremes” (TNn and TXn, representing coldest night and day) when removing the natural internal variability influences (the Arctic Oscillation and Pacific decadal oscillation) from observations. Christidis and Stott (2016) carried out an optimal detection analysis of 16 extreme temperature indices associated with intensity, frequency and duration, which were introduced by the Expert Team on Climate Change Detection and Indices (ETCCDI; Zhang et al. 2011). Using HadEX2 observations and HadGEM2-ES simulations for 1961–2010, they detected anthropogenic signals at the global scale and over Europe from most of the extreme indices. Recently, Lu et al. (2018) analyzed cold spell and warm spell duration indices using HadEX2 and CMIP5 datasets for 1958–2010 and found anthropogenic influences on both indices at the global scale. Anthropogenic signals were also detectable in most continental regions in the warm spell duration but less robustly in the cold spell duration. Yin and Sun (2018) performed a similar analysis focusing on fixed threshold indices (frost days, tropical nights, ice days, and summer days) for 1961–2010, and detected anthropogenic signals at global and continental scales.
Although many studies have identified anthropogenic influences on the large-scale changes in extreme temperatures, quantification of anthropogenic greenhouse gas (GHG) and aerosol contributions remains challenging. Anthropogenic aerosols (AA) are expected to have overall cooling effect on surface climate by scattering incoming solar radiation (aerosol direct effect) and changing cloud properties (aerosol indirect effect). Aerosol interactions with the meteorological state (aerosol dynamic effect) could also affect extreme temperatures depending on regions (e.g., Chen et al. 2019; Stone et al. 2019). Thus, better understanding of the relative contributions of GHG and AA forcings to the observed changes provides an important background for constraining future climate changes based on observations (e.g., Gillett et al. 2012; Shiogama et al. 2016). Although some previous studies quantified GHG and AA contributions to mean temperature changes (e.g., Najafi et al. 2015; Bilbao et al. 2019), there have been no GHG/AA quantification studies for extreme temperatures particularly at regional scales in spite of its greater impacts on human society and the ecosystem than the mean temperatures. A major difficulty was the limited availability of CMIP5 simulations forced by individual forcing factors, in particular, AA (Gillett et al. 2016). In this respect, newly released CMIP6 multimodel simulations (Eyring et al. 2016) provide a variety of external forcing experiments including GHG only, AA only, and natural (solar and volcanic activities) only forcings for an extended period up to 2020 under the Detection and Attribution Model Intercomparison Project (DAMIP; Gillett et al. 2016). This enables one to assess the relative importance of individual forcing factors in the observed changes in extreme temperatures. In this context, by conducting an updated D&A analysis using CMIP6 simulations, the present study aims at quantifying the contributions of GHG and AA to the observed global and regional changes in extreme temperature during 1951–2015.
This paper is structured as follows. Observational and model data and detection methods are described in section 2. D&A results are provided in section 3 focusing on the updated assessment of anthropogenic influences and the isolation of GHG and AA contributions. A summary and discussion are given in section 4.
2. Data and methods
a. Observations and model simulations
The HadEX3 dataset is used as observations (Dunn et al. 2020; https://www.metoffice.gov.uk/hadobs/hadex3). Updating the HadEX2 dataset, it provides ETCCDI extreme indices over global land from 1901 to 2018 and has a horizontal resolution of 1.875° longitude × 1.25° latitude. Four indices of extreme temperature intensity are analyzed in this study; annual minimum daily minimum and maximum temperatures (TNn and TXn, defined as cold extremes) and annual maximum daily minimum and maximum temperatures (TNx and TXx, defined as warm extremes). The analysis period is chosen as 1951–2015 considering data availability (see below for details).
Seven CMIP6 models are used (Table 1), which provide all datasets for historical (anthropogenic plus natural forcing; ALL), natural-only (NAT), greenhouse gas-only (GHG), and anthropogenic aerosol-only (AA) forcing experiments. There are 46 ALL simulations, 32 GHG simulations, 32 AA simulations, and 32 NAT simulations in total. Out of seven models analyzed, BCC-CSM2-MR used prescribed aerosol burdens (Wu et al. 2019) and the other six models implemented prescribed emissions of aerosols and its precursors (Swart et al. 2019; Voldoire et al. 2019; Yukimoto et al. 2019; Mulcahy et al. 2020; Lurton et al. 2020; Seland et al. 2020). Anthropogenic forcing signals (ANT) are estimated from differences between ALL and NAT. All forced experiments have data for 1951–2015 except ALL runs, which end in 2014 and are extended by 2015 using the corresponding Shared Socioeconomic Pathway (SSP) 2–4.5 scenario runs. Preindustrial control simulations (CTL) from 21 models are also used to estimate internal climate variability, which provide 172 nonoverlapping chunks of 65-yr length (Table 1).
CMIP6 models used in this study. Numbers indicate the number of runs for each model except for CTL, where it indicates the number of 65-yr nonoverlapping chunks.
b. Data processing
Generally, extremes do not follow normal distributions and their spatial averages can be affected more by locations with stronger values. Also, different horizontal resolutions between models and observations can affect changes in extremes and comparison results. In this respect, annual extreme temperatures are standardized prior to analysis following previous studies (Min et al. 2011, 2013). We first fit 65-yr time series of annual extreme temperatures to a generalized extreme value (GEV) distribution, which is the limiting distribution of block maxima according to the statistical extreme value theory (Coles 2001). Then we convert annual extreme values into probability-based indices (PI) ranging from 0 to 1, which are defined as cumulative density function (CDF) values from the fitted GEV distribution. This conversion into PI is done on the original grids of each model and then obtained PIs are interpolated onto the same grids of HadEX3. The interpolation using PI is more appropriate than using extreme temperatures to reduce the influence of local outliers. It should be noted, however, that CMIP6 models do not have high resolution (typically 1°–2°) enough to resolve mesoscale processes related to extremes, particularly heavy precipitation (Paik et al. 2020), although the spatial scale of extreme temperatures is generally large (Donat et al. 2013a).
Observational data coverage varies with time, which can affect area-mean values and their time series. For fair comparison with observations, we make models mimic observed variations by applying a space–time mask of observational data coverage to model data for each extreme index. Then grids with sufficient data are selected to consider long-term changes, where data exist over more than 70% of 1951–2015 plus at least 3 years during 2011–15. Here, we apply the data availability mask of TXx to the other indices, which has the least coverage (80.4%) of global land excluding Antarctica, so as to make spatial domains consistent across variables (Fig. S1 in the online supplemental material) following Kim et al. (2016).
Detection analysis is carried out for global means, five continental domains [North America (NA), South America (SA), Europe (EUR), Asia (ASI), and Oceania (OCE)], and 33 subcontinental domains with sufficient data (Fig. 1, Table 2). Subcontinental domains are based on 41 regions of Iturbide et al. (2020) prepared for the Sixth Assessment Report (AR6) IPCC Working Group I. All selected continental and subcontinental domains have more than 64% spatial data coverage. To reduce interannual variability noise, we take 5-yr nonoverlapping means for each domain, which gives a 13-dimensional vector consisting of 5-yr mean PI values over the 65 years. For South America and its subregions, an 11-dimensional vector (i.e., eleven 5-yr mean PIs) is constructed from 1961–2015 due to the poor data coverage during the first decade, which is not seen in other continents including African subregions (not shown).
The 41 subregions over global land adopted from the domains of Iturbide et al. (2020). Six continental domains are shown in different colors: North America (blue), Europe (green), Asia (red), South America (yellow), Africa (brown), and Oceania (violet). Acronyms in gray indicate subregions with spatial data coverage less than 50%, which are not used in this study, including Africa (see Table 2). Note that only land grids are considered, and Antarctica is excluded.
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
List of 33 subcontinental regions used in the analysis. HadEX3 data coverage (percentage of grid cells with data) is provided for each region based on TXx grids with sufficient data.
Because the time-mean PI is near 0.5, the resulting area-averaged PI trends can be interpreted as relative changes with respect to typical 2-yr return values (Min et al. 2013). For better interpretation of PI, we illustrate the relationship between global mean PI anomalies (note the percentage unit) and annual extreme temperatures (K) in Fig. S2. There is a strong linear relationship (correlation coefficient r = 0.98), meaning that PI well represents original extreme temperatures. Temperature–PI linear regression slopes (K per percent) are generally stronger in cold extremes than in warm extremes (see Table S1), which indicates stronger variabilities in cold extremes. To check the influence of standardization on signal detectability, we further compared signal-to-noise ratios (S/N) when using PI and raw extreme temperatures for global and all continental domains (Fig. S3). Here the S/N is defined as the ratio of variance of fingerprint (see below) to that of internal variability (DelSole et al. 2018). Results indicate that area-mean PIs have greater S/N than area-mean extreme temperatures, more strongly for warm extremes, suggesting an advantage of using a standardized index.
c. Detection method
An optimal fingerprinting method (Allen and Stott 2003) is employed to compare observed and modeled extreme temperature changes, which assumes that the observed changes y are the linear sum of the responses to individual external forcings X (fingerprints) and the internally generated climate variability noise ε such that y = (X − ν)β + ε. An optimal fingerprint indicates the pattern of response to a given external forcing (fingerprint) with the highest signal relative to the internal variability and other responses. The regression coefficient β (also called “scaling factor”) is estimated using a total least squares (TLS) method, which takes account of sampling errors ν in fingerprints. Fingerprints are estimated from multimodel means (MME) for a given forcing and the internal variability is estimated from CTL chunks. To give equal weighting to each model, we first calculate each model mean using available ensemble members (Table 1) and then obtain the MME. If modeled internal variability is too small, it can result in spurious signal detection due to underestimated noise. To take into account such cases, a residual consistency test is performed following Allen and Stott (2003), in which modeled internal variability (variance from CTL simulations) is compared to the variability of observational residuals (obtained from the linear regression equation above, i.e., by removing the forcing-explaining part from the observations). The failure of the residual consistency test due to the too-small modeled variability indicates that detection results are less robust. The 172 nonoverlapping chunks of CTL simulations (Table 1) are divided into two sets. One set is used to obtain best estimates of β and the other set is used to test residual consistency and to estimate the 90% confidence interval of β following Ribes et al. (2013).
Two-signal and three-signal analyses are performed for four extreme indices over global, continental, and subcontinental domains. In the two-signal analysis, observed patterns are regressed onto ANT and NAT fingerprints at the same time and the three-signal analysis considers GHG, AA, and NAT fingerprints simultaneously in the multiple regression. It is assumed that influences of other anthropogenic such as ozone and land cover change are relatively small in the three-signal analysis. Finally, attributable changes due to individual forcings are estimated by multiplying GHG, AA, and NAT fingerprints X by the corresponding scaling factors β obtained from the three-signal analysis (Hegerl and Zwiers 2011). In our case, PI time series of each forcing (fingerprint xi, i = GHG, AA, and NAT) are first weighted by each scaling factor βi and then linear trends of scaled time series βixi are calculated for each forcing. Here, best estimates and the 5%–95% range of the attributable trends are obtained from the corresponding best estimates and 5%–95% ranges of each scaling factor. For better interpretation, the unit of attributable trend in PI (%) is converted back into degrees kelvin based on the observed extreme temperature–PI linear relationship (Table S1; see above).
3. Results
a. Observed and simulated trends
Figure 2 shows spatial distributions of the observed and simulated PI trends (percent per year) over 1951–2015. For cold extremes (TNn and TXn), the observations show overall warming trends across the globe except India and southeastern North America (for TXn only). The warm extreme indices (TNx and TXx) also exhibit increasing trends but there are more regions with cooling trends such as Central America and India for TNx and southeastern North America and southern South America for TXx. The local cooling (the so-called warming hole) in the southeastern United States during summer has been reported and analyzed by previous studies (e.g., Meehl et al. 2012; Donat et al. 2013a; Kim et al. 2016; Mascioli et al. 2017).
Spatial patterns of PI trends (% yr−1) during 1951–2015 for TNn, TXn, TNx, and TXx from HadEX3 observations and CMIP6 ALL, GHG, AA, and NAT simulations (multimodel means). Hatching indicates areas with statistically significant trends at the 5% level for HadEX3 and good intermodel agreements (at least five models out of seven have the same sign of trends) for CMIP6 results.
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
Modeled trends are less noisy because multimodel averages smooth out the intermodel differences in trends. ALL runs exhibit warming trends everywhere for all indices with stronger warming over high latitudes in cold extremes and over low to midlatitudes in warm extremes. Consistent strong warming trends are observed over northern South America for all extreme indices. GHG results display similar patterns to ALL but with stronger magnitudes, suggesting that ALL responses are dominated by GHG-induced warming. In contrast, AA runs show overall cooling responses to the increased anthropogenic aerosols. The strongest cooling occurs in India and southern China for warm extremes with good intermodel agreement (hatching), consistent with previous studies (e.g., Chen and Dong 2019). In contrast, a slight warming response is found over central-eastern Europe to western Siberia in warm extremes although the intermodel difference is large. This local warming seems to be due mostly to the reduced aerosol emissions over Europe since 1980s (Smith et al. 2011; Klimont et al. 2013; Dong et al. 2017; Undorf et al. 2018; Aas et al. 2019), which is supported by our simple comparison of the aerosol optical depth (AOD) and TXx trends between early and recent decades (Fig. S4). NAT runs show decreasing trends over a large part of the global land but with very weak amplitudes.
Figure 3 illustrates global mean times series of PI anomalies for 1951–2015. Observations show increasing trends in all four indices, larger than internal variability (gray shading, 5%–95% ranges of CTL). For cold extremes (Figs. 3a,b), observed warming trends are stronger than ALL simulations, particularly for TNn (~40% increase in PI during 65 years), which is consistent with CMIP5-based findings by Kim et al. (2016). GHG runs have monotonic increases (about 20%–25% increases) while AA runs show steady decreases (around 10% decrease) for both TNn and TXn. NAT simulations possess very weak negative trends with some fluctuations that are within the range of internal variability.
Time series of global mean PI anomalies for TNn, TXn, TNx, and TXx during 1951–2015 and its linear trends (bars) with 5%–95% confidence intervals (gray error bars) from HadEX3 observations (black) and CMIP6 ALL (green), ANT (orange), GHG (red), AA (purple), and NAT (blue) simulations. Gray shading indicates 5%–95% ranges of CTL simulations estimated using the first set of 86 total 65-yr chunks (Table 1).
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
For warm extremes, ALL runs capture the observed long-term variations remarkably well, including stronger amplitudes in TNx (~35% increase) than in TXx (Figs. 3c,d). GHG forcing induces steady increases in both TNx and TXx. The cooling trends due to AA forcing are stronger in warm extremes than in cold extremes, particularly before 1980s, which indicates that model-simulated AA cooling effects are stronger during summer seasons when the warm extremes usually occur. Since 1990s this cooling trends become flattened, consistent with the reduced emissions of anthropogenic aerosols over Europe and North America (Fig. S4). Long-term trends are very weak in NAT simulations, but the local cooling responses can be seen clearly in the early 1980s and early 1990s, corresponding to El Chichón and Mount Pinatubo eruptions, which are shared by the observed time series. In particular, the local cooling after Mount Pinatubo eruptions is quite strong well beyond the 90% ranges of CTL values (cf. Soden et al. 2002; Bender et al. 2010; Paik and Min 2017). Due to the offsetting cooling effects by AA and NAT forcings, ALL simulations exhibit smaller positive trends than GHG and ANT for all four indices.
b. Two-signal analysis: ANT and NAT
Figure 4 shows two-signal detection results for global and continental scales (also see Fig. S5). Best estimates and 90% confidence intervals of the scaling factors for ANT and NAT are displayed for four extreme indices. For cold extremes (Figs. 4a,b), ANT signals are detected in global and most continental regions except for two cases (EUR for TNn and ASI for TXn). The best estimates of ANT generally are larger than unity in cold extremes, indicating that models underestimate the observed warming of cold extremes as found in the time series (Figs. 3a,b). The residual consistency test is passed in all cases. Compared to previous studies based on CMIP3 and CMIP5 models (Min et al. 2013; Kim et al. 2016), ANT signals are newly detected over EUR for TXn. NAT signals are also detected over NA and EUR for TXn. When comparing with results using a shorter period of 1951–2005 (Fig. S6), this improved detection seems to be due partly to the use of an extended period, which usually increases signal-to-noise ratios under continued warming trends (Fig. S5).
Two-signal detection results for TNn, TXn, TNx, and TXx at global and continental scales. Best estimates of scaling factors and its 90% confidence intervals for ANT (orange) and NAT (blue). An asterisk indicates that modeled internal variability is too small compared to that of residual observations based on a residual consistency test.
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
In warm extremes (Figs. 4c,d), the ANT response is detected over the global region (GLB) and all continental regions in both indices. Compared to cold extremes, the scaling factors for warm extremes are close to unity with narrow confidence intervals. This implies that signal amplitudes of CMIP6 models are consistent with observations for warm extremes, as found in previous studies based on CMIP5 models (Kim et al. 2016). Also, NAT signals are more detectable than in cold extremes (GLB, SA, and ASI for TNx and GLB, NA, and SA for TXx). ANT and NAT signals are newly detected for both indices over SA, which seems to be due to the improved data coverage with a limited influence of the extended period (Figs. S6c,d).
Two-signal analysis has been conducted for 33 subcontinental regions (see Fig. 1 and Table 2), and Table 3 summarizes the D&A results. ANT signals are detected in separation from NAT signals over many subregions with more frequent detection for warm extremes. NAT signals are also detected over several subregions. Figure 5 shows scaling factor distributions for warm extremes over 33 subregions. Compared to continental-scale results, confidence intervals of scaling factors become broader in part due to the increased internal variability at smaller spatial scales. Poorer projection of observed variations onto the model-derived fingerprints (with the discrepancies being averaged out at larger scales) or poor station sampling in the observations can also get smaller signal-to-noise ratios at smaller spatial scales. Nevertheless, ANT signals are detectable over more than 60% of subregions (27 regions for TNx and 20 regions for TXx). Also, only a few regions have failed the residual consistency test. For TNx, scaling factors for ANT tend to be larger than unity in several regions. Simultaneous detection of ANT and NAT occurs in some regions (EEU, RAR, RFE, TIB, EAS, and NSA) but with large uncertainty ranges. Two-signal detection results for TXx are very similar to those for TNx but the scaling factors tend to be smaller than unity in some NA and ASI subregions, consistent with the continental scale results (Figs. 4c,d).
Two-signal detection results for 33 subcontinental regions. Letters D and A indicate detection (when the 90% range of scaling factor lies above zero) and attribution (additionally the scaling factor range includes unity).
As in Fig. 4, but for 33 subcontinental regions for (a) TNx and (b) TXx. Each panel is divided into two rows showing results for (top) 20 Northern Hemisphere regions and (bottom) 13 Southern Hemisphere regions. Vertical dashed lines delineate each continent.
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
As a whole, our results show that ANT is robustly detected in separation from NAT at global, continental, and many subcontinental scales. Some improved detection is found due to the use of the extended period (in cold extremes) and the better observational coverage (over South American regions in both cold and warm extremes). Generally, warm extremes (in particular TNx) are found to have more frequent detection of ANT signals than cold extremes, supporting previous studies (Zwiers et al. 2011; Min et al. 2013) based on different models and/or methods.
c. Three-signal analysis: GHG, AA, and NAT
Three-signal analysis results for global and continental scales are shown in Fig. 6. In general, limited detection is expected in three-signal analysis with wide uncertainties in the scaling factors as signals can be more collinear (Jones et al. 2016; DelSole et al. 2018). For cold extremes (Figs. 6a,b), GHG signals are detectable in global land and all continents except EUR for TNn. Scaling factors for GHG remain larger than unity, similar to those for ANT from the two-signal analysis (Figs. 4a,b). This implies that ANT detection from the two-signal analysis is mainly contributed by GHG influences. AA influences are only detected in ASI for TNn while NAT is detectable in NA and SA for both indices. Limited AA detection might be associated with a strong collinearity between GHG and AA responses, which have monotonic increases and decreases, respectively (Fig. 3 and Fig. S5). Indeed, temporal correlation coefficients between GHG and AA time series of PI anomalies are very strong (<−0.8) in many regions for cold extremes. For warm extremes (Figs. 6c,d), GHG is detected in all continental regions except OCE for TNx, resembling ANT detection results from the two-signal analysis (Figs. 4c,d). AA signals are also detected in more cases including GLB, NA, EUR, and ASI for TNx and GLB, SA, EUR, ASI, and OCE for TXx than cold extremes. More frequent AA detection in warm extremes is likely to be related to the strong cooling responses during summer as discussed above. PI time series show that AA forcing induces strong cooling of warm extremes during early decades over most continental regions while a slight warming is simulated in EUR for recent decades (Fig. S5).
As in Fig. 4, but for three-signal detection results using GHG (red), AA (purple), and NAT (blue).
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
To quantify each forcing contribution to extreme temperature warming, we analyze attributable trends in TXx to GHG, AA, and NAT forcings in comparison with the observed (Fig. 7; see section 2c for method details). Global mean TXx has increased by 0.96 K in the observations and GHG makes a large contribution to the warming by +1.76 K (90% confidence interval: from +1.20 to +2.38 K). AA cooling effect offsets the GHG-induced warming by −0.88 K (from −0.52 to −1.30 K) while NAT exerts negligible influence with −0.02 K (from −0.06 to +0.02 K). Results for five continental regions support the dominant contributions of GHG and AA forcings to the observed TXx changes with amplitudes varying across domains. NA experienced a relatively weak observed warming (+0.44 K) and GHG and AA contributions are estimated as +0.75 and −0.29 K, respectively. In contrast, SA had a strong warming (+1.24 K) to which GHG and AA contributions are estimated as +2.58 and −1.43 K, respectively, indicating a large cancelation. Interestingly, EUR shows the strongest observed increase in TXx (+1.58 K) due to a strong GHG (+2.09 K) with much weaker AA contribution (−0.40 K). The latter seems to be due to the large cancelation between the early period increase and the recent period decrease of AA emissions over the region as discussed above (Fig. S4). Asian results are very similar to the global case, with AA offsetting about one-third of GHG-induced warming. Over OCE, there is a large cancelation between GHG warming (+1.66 K) and AA cooling (−1.35 K) with large uncertainties, resulting in a weaker observed warming (+0.67 K). When checking the spatial patterns of AA forcing trends and temperature responses (Fig. S4), the strong AA cooling is observed in the Southern Hemisphere land areas including SA and OCE as well as in the Northern Hemisphere midlatitudes (also see Fig. 2). This indicates that temperature responses to AA forcing extend well beyond the emission locations through large-scale atmospheric and oceanic circulation changes (e.g., Rotstayn et al. 2012; Shindell et al. 2010, 2015). The AA-induced remote cooling response is stronger during the early decades (1951–80) than the recent decades (1981–2015; Fig. S4), detailed mechanisms of which need further investigation.
Attributable trends of TXx (K per 65-yr period) to GHG (red), AA (purple), and NAT (blue) signals compared with the observed trends (OBS; black) for global (GLB) and continental domains.
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
The three-signal analysis is conducted for subcontinental regions and results are summarized in Table 4. GHGs are detected over several regions for cold extremes but over more than 60% of the 33 regions for warm extremes. AA influences are detected less frequently than GHG, mainly in warm extremes. Figure 8 shows subregional details of GHG, AA, and NAT scaling factors for warm extremes. For TNx, GHG is detected across the continents, mainly over the Asian and the North and South American regions. AA is detected in several North American and Asian regions while NAT influences are limitedly detected in a few regions. Results for TXx are generally similar to those for TNx. A couple of differences are noticeable. GHG scaling factors for TXx tend to include unity in more regions over Asia and Europe, indicating better agreement with the observations. Also, AA signals are detected in Southern Hemisphere regions for TXx (NES, SES, SWAF, SEAF, NAU, and CAU). CEU, TIB and EAS (for TXx) are the only regions with all three signals simultaneously detected.
As in Fig. 5, but for three-signal detection results using GHG (red), AA (purple), and NAT (blue).
Citation: Journal of Climate 34, 3; 10.1175/JCLI-D-19-1023.1
Overall, our results provide a first quantitative assessment of GHG and AA contributions to the observed changes in warm extremes at continental and subcontinental scales based on the three-signal detection analysis using CMIP6 individual forcing experiments. We find for the first time that GHG signals are detectable in warm extremes over many subcontinental regions with AA signals jointly detected over several subregions.
4. Conclusions and discussion
This study presents updated detection and attribution analysis results using four indices of annual extreme temperature intensity (TNn, TXn, TNx, and TXx). HadEX3 observations over 1951–2015 are compared with CMIP6 simulations that were performed under different forcings (ALL, GHG, AA, and NAT) using an optimal fingerprinting technique, considering global, five continental, and 33 subcontinental domains. A particular focus is placed on the separation of ANT, GHG, and AA signals and the quantification of their contributions to the observed changes using two-signal and three-signal detection analyses.
Results from two-signal analysis show that ANT signals are robustly detected at global and most of the northern continental scales for all four indices, supporting previous findings. More frequent detection was obtained compared to previous studies, including the first detection of ANT signal over Europe (for TXn only) and South America, which seems to be due to the use of extended period and the improved spatial data coverage, respectively. The ANT influence is also detected over many subcontinental regions, particularly for warm extremes (more than 60% of the considered domains), which provides a new evidence for human influences on extreme temperatures at regional scales. In spite of overall greater signal detection, model underestimation of the observed warming of cold extremes remain in the CMIP6 models, particularly for TNn. Further investigation is warranted into the causes of the observation–model discrepancy.
Three-signal analysis demonstrates for the first time that GHG influence on extreme temperature intensity is detected at global, continental, and even subcontinental scales, in separation from AA and NAT influences. GHG forcing is found to contribute dominantly to the observed warming, which is more frequently identified in warm extremes than cold extremes. In addition, AA contributions are jointly detected in several continental and subcontinental domains in warm extremes. Interestingly, AA is found to not only contribute to the observed cooling during the early decades over globe, Europe, and Asia but induce the slight warming over Europe as well during the recent decades.
Our results of GHG and AA signal detection are intriguing, considering the difficulty in separating multiple signals under the situation of possible collinearity between GHG and AA responses (Ribes and Terray 2013; DelSole et al. 2018). To further improve detection power, combining spatial information into the analysis would be useful as a way of reducing or avoiding the collinearity. Also, more sensitivity tests are needed to assess robustness of our conclusions by using other observations and more available models. To assess observational uncertainties, we have repeated two- and three-signal analyses using an independent extreme dataset from GHCNDEX (Donat et al. 2013b) for global and continental domains, and compared results with those from HadEX3. A 60-yr period of 1951–2010 is used here and South America is excluded due to a poor data coverage of GHCNDEX during recent years. Overall results remain very similar between the two datasets (Figs. S7 and S8) with the ANT signals detected for all extreme temperature indices and the GHG and AA signals detected for warm extremes over most of the domains, suggesting insensitiveness of our main findings to the observational uncertainties. However, a comprehensive analysis using other available observations that provide recent records is needed especially for subcontinental regions to better assess associated impacts and inform risk management (Seneviratne et al. 2012). Regarding the model uncertainty, a caveat remains in the model resolutions, which are not high enough to resolve detailed thermodynamic and dynamic processes responsible for the detected changes in extreme temperatures at local scales as discussed above. In this regard, a systematic model evaluation using process-based metrics is warranted (e.g., Eyring et al. 2019), particularly, for smaller scales where different mechanisms work depending on regional forcings and associated feedbacks like anthropogenic aerosols.
Acknowledgments
We thank Dáithí Stone and two anonymous reviewers for their useful comments. This study is supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (NRF-2018R1A5A1024958). We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access (https://esgf-node.llnl.gov/projects/cmip6/), and the multiple funding agencies who support CMIP6 and ESGF.
REFERENCES
Aas, W., and Coauthors, 2019: Global and regional trends of atmospheric sulfur. Sci. Rep., 9, 953, https://doi.org/10.1038/s41598-018-37304-0.
Alexander, L. V., and Coauthors, 2006: Global observed changes in daily climate extremes of temperature and precipitation. J. Geophys. Res., 111, D05109, https://doi.org/10.1029/2005JD006290.
Allen, M. R., and P. A. Stott, 2003: Estimating signal amplitudes in optimal fingerprinting, part I: Theory. Climate Dyn., 21, 477–491, https://doi.org/10.1007/s00382-003-0313-9.
Bender, F. A. M., A. M. L. Ekman, and H. Rodhe, 2010: Response to the eruption of Mount Pinatubo in relation to climate sensitivity in the CMIP3 models. Climate Dyn., 35, 875–886, https://doi.org/10.1007/s00382-010-0777-3.
Bilbao, R. A. F., J. M. Gregory, N. Bouttes, M. D. Palmer, and P. Stott, 2019: Attribution of ocean temperature change to anthropogenic and natural forcings using the temporal, vertical and geographical structure. Climate Dyn., 53, 5389–5413, https://doi.org/10.1007/s00382-019-04910-1.
Bindoff, N., and Coauthors, 2013: Detection and attribution of climate change: From global to regional. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 867–952.
Caesar, J., L. Alexander, and R. Vose, 2006: Large-scale changes in observed daily maximum and minimum temperatures: Creation and analysis of a new gridded data set. J. Geophys. Res., 111, D05101, https://doi.org/10.1029/2005JD006280.
Chen, W., and B. Dong, 2019: Anthropogenic impacts on recent decadal change in temperature extremes over China: Relative roles of greenhouse gases and anthropogenic aerosols. Climate Dyn., 52, 3643–3660, https://doi.org/10.1007/s00382-018-4342-9.
Chen, W., B. Dong, L. Wilcox, F. Luo, N. Dunstone, and E. J. Highwood, 2019: Attribution of recent trends in temperature extremes over China: Role of changes in anthropogenic aerosol emissions over Asia. J. Climate, 32, 7539–7560, https://doi.org/10.1175/JCLI-D-18-0777.1.
Christidis, N., and P. A. Stott, 2016: Attribution analyses of temperature extremes using a set of 16 indices. Wea. Climate Extremes, 14, 24–35, https://doi.org/10.1016/j.wace.2016.10.003.
Christidis, N., P. A. Stott, and S. J. Brown, 2011: The role of human activity in the recent warming of extremely warm daytime temperatures. J. Climate, 24, 1922–1930, https://doi.org/10.1175/2011JCLI4150.1.
Coles, S., 2001: An Introduction to Statistical Modeling of Extreme Values. Springer, 209 pp.
Coumou, D., and A. Robinson, 2013: Historic and future increase in the global land area affected by monthly heat extremes. Environ. Res. Lett., 8, 034018, https://doi.org/10.1088/1748-9326/8/3/034018.
DelSole, T., L. Trenary, X. Yan, and M. K. Tippett, 2018: Confidence intervals in optimal fingerprinting. Climate Dyn., 52, 4111–4126, https://doi.org/10.1007/s00382-018-4356-3.
Donat, M. G., and Coauthors, 2013a: Updated analyses of temperature and precipitation extreme indices since the beginning of the twentieth century: The HadEX2 dataset. J. Geophys. Res. Atmos., 118, 2098–2118, https://doi.org/10.1002/jgrd.50150.
Donat, M. G., L. V. Alexander, H. Yang, I. Durre, R. Vose, and J. Caesar, 2013b: Global land-based datasets for monitoring climatic extremes. Bull. Amer. Meteor. Soc., 94, 997–1006, https://doi.org/10.1175/BAMS-D-12-00109.1.
Dong, B., R. T. Sutton, and L. Shaffrey, 2017: Understanding the rapid summer warming and changes in temperature extremes since the mid-1990s over western Europe. Climate Dyn., 48, 1537–1554, https://doi.org/10.1007/s00382-016-3158-8.
Du, H. B., and Coauthors, 2019: Precipitation from persistent extremes is increasing in most regions and globally. Geophys. Res. Lett., 46, 6041–6049, https://doi.org/10.1029/2019GL081898.
Dunn, R. J. H., and Coauthors, 2020: Development of an updated global land in-situ-based data set of temperature and precipitation extremes: HadEX3. J. Geophys. Res. Atmos., 125, e2019JD032263, https://doi.org/10.1029/2019JD032263.
Eyring, V., S. Bony, G. A. Meehl, C. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Eyring, V., and Coauthors, 2019: Taking climate model evaluation to the next level. Nat. Climate Change, 9, 102–110, https://doi.org/10.1038/s41558-018-0355-y.
Gillett, N. P., V. K. Arora, G. M. Flato, J. F. Scinocca, and K. von Salzen, 2012: Improved constraints on 21st-century warming derived using 160 years of temperature observations. Geophys. Res. Lett., 39, L01704, https://doi.org/10.1029/2011GL050226.
Gillett, N. P., and Coauthors, 2016: The Detection and Attribution Model Intercomparison Project (DAMIP v1.0) contribution to CMIP6. Geosci. Model Dev., 9, 3685–3697, https://doi.org/10.5194/gmd-9-3685-2016.
Hegerl, G., and F. Zwiers, 2011: Use of models in detection and attribution of climate change. Wiley Interdiscip. Rev.: Climate Change, 2, 570–591, https://doi.org/10.1002/wcc.121.
Iturbide, M., and Coauthors, 2020: An update of IPCC climate reference regions for subcontinental analysis of climate model data: Definition and aggregated datasets. Earth Syst. Sci. Data, https://doi.org/10.5194/ESSD-2019-258, in press.
Jones, G. S., P. A. Stott, and J. F. B. Mitchell, 2016: Uncertainties in the attribution of greenhouse gas warming and implications for climate prediction. J. Geophys. Res. Atmos., 121, 6969–6992, https://doi.org/10.1002/2015JD024337.
Kharin, V. V., F. W. Zwiers, X. Zhang, and M. Wehner, 2013: Changes in temperature and precipitation extremes in the CMIP5 ensemble. Climatic Change, 119, 345–357, https://doi.org/10.1007/s10584-013-0705-8.
Kim, Y.-H., S.-K. Min, X. Zhang, F. W. Zwiers, L. V. Alexander, M. G. Donat, and Y. S. Tung, 2016: Attribution of extreme temperature changes during 1951–2010. Climate Dyn., 46, 1769–1782, https://doi.org/10.1007/s00382-015-2674-2.
Klimont, Z., S. J. Smith, and J. Cofala, 2013: The last decade of global anthropogenic sulfur dioxide: 2000–2011 emissions. Environ. Res. Lett., 8, 014003, https://doi.org/10.1088/1748-9326/8/1/014003.
Lu, C., Y. Sun, and X. Zhang, 2018: Multimodel detection and attribution of changes in warm and cold spell durations. Environ. Res. Lett., 13, 074013, https://doi.org/10.1088/1748-9326/aacb3e.
Lurton, T., and Coauthors, 2020: Implementation of the CMIP6 forcing data in the IPSL-CM6A-LR Model. J. Adv. Model. Earth Syst., 12, e2019MS001940, https://doi.org/10.1029/2019MS001940.
Martinich, J., and A. Crimmins, 2019: Climate damages and adaptation potential across diverse sectors of the United States. Nat. Climate Change, 9, 397–404, https://doi.org/10.1038/s41558-019-0444-6.
Mascioli, N. R., M. Previdi, A. M. Fiore, and M. Ting, 2017: Timing and seasonality of the United States ‘warming hole.’ Environ. Res. Lett., 12, 034008, https://doi.org/10.1088/1748-9326/aa5ef4.
Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor, 2007: The WCRP CMIP3 multimodel dataset: A new era in climate change research. Bull. Amer. Meteor. Soc., 88, 1383–1394, https://doi.org/10.1175/BAMS-88-9-1383.
Meehl, G. A., J. M. Arblaster, and G. Branstator, 2012: Mechanisms contributing to the warming hole and the consequent U.S east–west differential of heat extremes. J. Climate, 25, 6394–6408, https://doi.org/10.1175/JCLI-D-11-00655.1.
Min, S.-K., X. Zhang, F. W. Zwiers, and G. C. Hegerl, 2011: Human contribution to more-intense precipitation extremes. Nature, 470, 378–381, https://doi.org/10.1038/nature09763.
Min, S.-K., X. Zhang, F. W. Zwiers, H. Shiogama, Y. S. Tung, and M. Wehner, 2013: Multimodel detection and attribution of extreme temperature changes. J. Climate, 26, 7430–7451, https://doi.org/10.1175/JCLI-D-12-00551.1.
Morak, S., G. C. Hegerl, and N. Christidis, 2013: Detectable changes in the frequency of temperature extremes. J. Climate, 26, 1561–1574, https://doi.org/10.1175/JCLI-D-11-00678.1.
Mulcahy, J. P., and Coauthors, 2020: Description and evaluation of aerosol in UKESM1 and HadGEM3-GC3.1 CMIP6 historical simulations. Geosci. Model Dev., https://doi.org/10.5194/GMD-2019-357,accepted.
Myhre, G., and Coauthors, 2019: Frequency of extreme precipitation increases extensively with event rareness under global warming. Sci. Rep., 9, 16063, https://doi.org/10.1038/s41598-019-52277-4.
Najafi, M. R., F. Zwiers, and N. Gillett, 2015: Attribution of Arctic temperature change to greenhouse gas and aerosol influences. Nat. Climate Change, 5, 246–249, https://doi.org/10.1038/nclimate2524.
Paik, S., and S.-K. Min, 2017: Climate responses to volcanic eruptions assessed from observations and CMIP5 multimodels. Climate Dyn., 48, 1017–1030, https://doi.org/10.1007/s00382-016-3125-4.
Paik, S., S.-K. Min, X. Zhang, M. G. Donat, A. D. King, and Q. Sun, 2020: Determining the anthropogenic greenhouse gas contribution to the observed intensification of extreme precipitation. Geophys. Res. Lett., 46, e2019GL086875, https://doi.org/10.1029/2019GL086875.
Ribes, A., and L. Terray, 2013: Application of regularized optimal fingerprinting to attribution. Part II: Application to global near-surface temperature. Climate Dyn., 41, 2837–2853, https://doi.org/10.1007/s00382-013-1736-6.
Ribes, A., S. Planton, and L. Terray, 2013: Application of regularized optimal fingerprinting to attribution. Part I: Method, properties and idealised analysis. Climate Dyn., 41, 2817–2836, https://doi.org/10.1007/s00382-013-1735-7.
Rotstayn, L. D., S. J. Jeffrey, M. A. Collier, S. M. Dravitzki, A. C. Hirst, J. I. Syktus, and K. K. Wong, 2012: Aerosol- and greenhouse gas-induced changes in summer rainfall and circulation in the Australasian region: A study using single-forcing climate simulations. Atmos. Chem. Phys., 12, 6377–6404, https://doi.org/10.5194/acp-12-6377-2012.
Seland, Ø., and Coauthors, 2020: The Norwegian Earth System Model, NorESM2—Evaluation of the CMIP6 DECK and historical simulations. Geosci. Model Dev. Discuss., https://doi.org/10.5194/GMD-2019-378,in review.
Seneviratne, S., and Coauthors, 2012: Changes in climate extremes and their impacts on the natural physical environment. Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation, C. B. Field et al., Eds., Cambridge University Press, 109–230.
Seneviratne, S., M. G. Donat, B. Mueller, and L. V. Alexander, 2014: No pause in the increase of hot temperature extremes. Nat. Climate Change, 4, 161–163, https://doi.org/10.1038/nclimate2145.
Shindell, D., M. Schulz, Y. Ming, T. Takemura, G. Faluvegi, and V. Ramaswamy, 2010: Spatial scales of climate response to inhomogeneous radiative forcing. J. Geophys. Res., 115, D19110, https://doi.org/10.1029/2010JD014108.
Shindell, D., G. Faluvegi, L. Rotstayn, and G. Milly, 2015: Spatial patterns of radiative forcing and surface temperature response. J. Geophys. Res. Atmos., 120, 5385–5403, https://doi.org/10.1002/2014JD022752.
Shiogama, H., and Coauthors, 2016: Predicting future uncertainty constraints on global warming projections. Sci. Rep., 6, 18903, https://doi.org/10.1038/srep18903.
Smith, S. J., J. van Aardenne, Z. Klimont, R. J. Andres, A. Volke, and S. Delgado Arias, 2011: Anthropogenic sulfur dioxide emissions: 1850–2005. Atmos. Chem. Phys., 11, 1101–1116, https://doi.org/10.5194/acp-11-1101-2011.
Soden, B. J., R. T. Wetherald, G. L. Stenchikov, and A. Robock, 2002: Global cooling after the eruption of Mount Pinatubo: A test of climate feedback by water vapor. Science, 296, 727–730, https://doi.org/10.1126/science.296.5568.727.
Son, J.-Y., J. C. Liu, and M. L. Bell, 2019: Temperature-related mortality: A systematic review and investigation of effect modifiers. Environ. Res. Lett., 14, 073004, https://doi.org/10.1088/1748-9326/ab1cdb.
Stone, D. A., and Coauthors, 2019: Experiment design of the international CLIVAR C20C+ detection and attribution project. Wea. Climate Extremes, 24, 100206, https://doi.org/10.1016/j.wace.2019.100206.
Swart, N. C., and Coauthors, 2019: The Canadian Earth System Model version5 (CanESM5.0.3). Geosci. Model Dev., 12, 4823–4873, https://doi.org/10.5194/gmd-12-4823-2019.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Undorf, S., M. A. Bollasina, and G. C. Hegerl, 2018: Impacts of the 1900–74 increase in anthropogenic aerosol emissions from North America and Europe on Eurasian summer climate. J. Climate, 31, 8381–8399, https://doi.org/10.1175/JCLI-D-17-0850.1.
Vogel, E., M. G. Donat, L. V. Alexander, M. Meinshausen, D. K. Ray, D. Karoly, N. Meinshausen, and K. Frieler, 2019: The effects of climate extremes on global agricultural yields. Environ. Res. Lett., 14, 054010, https://doi.org/10.1088/1748-9326/ab154b.
Voldoire, A., and Coauthors, 2019: Evaluation of CMIP6 DECK experiments with CNRM-CM6-1. J. Adv. Model. Earth Syst., 11, 2177–2213, https://doi.org/10.1029/2019MS001683.
Wu, T., and Coauthors, 2019: The Beijing Climate Center Climate System Model (BCC-CSM): The main progress from CMIP5 to CMIP6. Geosci. Model Dev., 12, 1573–1600, https://doi.org/10.5194/gmd-12-1573-2019.
Yin, H., and Y. Sun, 2018: Detection of anthropogenic influence on fixed threshold indices of extreme temperature. J. Climate, 31, 6341–6352, https://doi.org/10.1175/JCLI-D-17-0853.1.
Yukimoto, S., and Coauthors, 2019: The Meteorological Research Institute Earth System Model version 2.0, MRI-ESM2.0: Description and basic evaluation of the physical component. J. Meteor. Soc. Japan, 97, 931–965, https://doi.org/10.2151/jmsj.2019-051.
Zhang, W., T. Zhou, L. Zou, L. Zhang, and X. Chen, 2018: Reduced exposure to extreme precipitation from 0.5°C less warming in global land monsoon regions. Nat. Commun., 9, 3153, https://doi.org/10.1038/s41467-018-05633-3.
Zhang, X., L. Alexander, G. C. Hegerl, P. Jones, A. Klein Tank, T. C. Peterson, B. Trewin, and F. W. Zwiers, 2011: Indices for monitoring changes in extremes based on daily temperature and precipitation data. Wiley Interdiscip. Rev.: Climate Change, 2, 851–870, https://doi.org/10.1002/WCC.147.
Zwiers, F. W., X. Zhang, and Y. Feng, 2011: Anthropogenic influence on long return period daily temperature extremes at regional scales. J. Climate, 24, 881–892, https://doi.org/10.1175/2010JCLI3908.1.
