Attribution of the unprecedented summer 2022 compound marine and terrestrial heatwave in the Northwest Pacific

In boreal summer (July–August) 2022, an unprecedented marine heatwave (MHW) occurred in the northwest Pacific Ocean (NWP), while a record-breaking terrestrial heatwave (THW) hit the Yangtze River Basin (YRB). The temperature anomalies caused by this compound MHW-THW event exceeded climatology by 2.5 standard deviations (SDs), breaking the historical record for nearly 100 years, with severe impacts on the ecosystems and social economy. To investigate the underlying causes, we explored the potential roles of anthropogenic forcing, atmospheric circulation, and ‘triple-dip’ La Niña on this compound event using the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model simulations. Results indicate that the 2022-like compound MHW-THW event was extremely unlikely to happen without anthropogenic warming, and that such extreme heatwaves were governed by the climatic mean temperature rather than changes in temperature variability. Notably, the anticyclone circulation patterns and associated high-pressure systems (i.e. western North Pacific subtropical high (WNPSH) and South Asian high (SAH)) increase the probability of a 2022-like MHW-THW event by 3.7 times. However, the La Niña phase has no significant effect on the occurrence probability of such events. We further estimate that the 2022-like MHW-THW event will become 7.5 and 11.4 times more likely under the SSP3-7.0 scenario by the middle and end of the 21st century, respectively. This study demonstrates the contribution of anthropogenic climate change and natural variability to the occurrence of compound MHW-THW events and highlights the urgent need to build mitigation strategies for compound MHW-THW events.


Introduction
Since the mid-20th century, the continuous emission of anthropogenic greenhouse gases has led to a considerable increase in global surface temperature, thereby enhancing the frequency of extreme weather and climate events (e.g.heatwaves, storm surges, extreme precipitation) (Cheng et al 2019, Laufkötter et al 2020, IPCC et al 2021, IPCC 2023).In addition to univariate extreme events, anthropogenic global warming also results in an increase in the frequency of compound events, which can cause more severe ecological and economic damage than single hazards (Zscheischler et al 2018(Zscheischler et al , 2020)).Significant research has delved into understanding the features, causes, and predictability of these extreme events (Boers et al 2019, Tan et al 2022, Mukherjee et al 2023).However, a critical question that has garnered widespread attention in recent years is to what extent anthropogenic climate change influences the likelihood of specific extreme events-a concept known as extreme event attribution.
Extreme event attribution is a relatively new field of climate science, with the aim to quantify and evaluate the relative contributions of anthropogenic forcing and natural variability to extreme events (Stott et al 2013, van der Wiel et al 2017, Philip et al 2018, van Oldenborgh et al 2021).Attribution methods, pioneered by Allen (2003), have spurred rapid growth in climate change attribution science globally (Stott et al 2004, Otto 2017).In recent years, extreme event attribution has gained momentum not only in the science world, such as the World Weather Attribution (WWA; www.worldweatherattribution.org) collaboration, but also in the media and public imagination (Sippel et al 2015, Noy et al 2023).Nevertheless, existing research predominantly focuses on attributing univariate extremes, neglecting compound extremes, which can have disproportionately severe impacts on ecosystems and human health (Zscheischler et al 2020, Herring et al 2022, Zscheischler and Lehner 2022, Wang et al 2023, Hao and Chen 2024).It is crucial to expand attribution research to encompass compound extremes in order to develop more effective mitigation strategies.
Heatwaves have been destructive hazards to marine and terrestrial ecosystems and biodiversity, such as mass coral bleaching, low primary productivity, water delivery shortages, reduced crop yield, and aggravated wildfires (Jacox et al 2016, Hughes et al 2018, IPCC 2019, Founda et al 2022).Anthropogenic warming has proved to be a dominant driver of the increased likelihood of either marine (Oliver 2019, Choi et al 2022, Li et al 2023, Tan et al 2023) or terrestrial (Sun et al 2014, 2018, Fischer et al 2021, Hua et al 2023) heatwaves.However, with global warming, the co-occurrence of marine heatwave (MHW) and terrestrial heatwave (THW) has increased significantly (IPCC 2023).For example, Salinger et al (2019,2020) found that New Zealand experienced the most extreme coupled ocean-atmosphere heatwaves on record in 1934/35, 2017/18, and 2018/19, which were driven by the atmosphere or combinations of atmospheric warming and oceanic heat advection.Tang et al (2023a) estimated the likelihood of 2021 concurrent MHW and THW across the northwestern Pacific to exceed about 30-fold caused by anthropogenically forced-warming.
In the boreal summer (July-August) of 2022, the northwest Pacific experienced an unprecedented compound marine and THW.The MHW struck the northwest Pacific Ocean (NWP; 22 • -47 • N, 117 • -151 • E; green rectangle in figure 1(a)), with regional average sea surface temperature anomalies (SSTA;relative to 1901-1930)  In addition to the contribution from anthropogenic forcing, natural variability also creates conditions that favor the occurrence of extreme heatwaves.A few studies have shown that anomalous anticyclones and high-pressure systems are the major atmospheric circulation patterns that result in high temperatures (Chen et al 2019, Li et al 2020, Lee et al 2020).As a remarkable case, in July-August 2022, both anomalies of the 500 hPa and 200 hPa geopotential heights (Z500 and Z200) indicate that the anomalous anticyclone circulations developed over the region, together with westward expanded WNPSH (figure 1(e)) and eastward stretched SAH (figure 1(f)).The amplified anticyclones were maintained by the stationary Rossby wave, which was caused by vorticity forcings and atmospheric diabetic heating (Liang et al 2024, Oh et al 2023), resulting in synchronized marine and THWs.Moreover, three consecutive years (2020−2022) of La Niña events were recorded as a rare 'triple-dip' , which likely contributed to the MHW and THW occurrences in the Northwest Pacific as well (Li et al 2023, Tan et al 2023).So far, however, the relative contributions of different factors in creating this record-breaking compound MHW-THW event remain poorly understood.
This study aims to answer three main questions: (1) What are the observed characteristics of the summer 2022 co-occurring NWP MHW and YRB THW in the historical context?(2) To what extent have anthropogenic forcing, atmospheric circulation, and the El Niño-Southern Oscillation (ENSO) contributed to the probability of such a compound MHW-THW event, with and without long-term warming

Model simulations and evaluation
To quantify the role of anthropogenic forcing on this event, following the applicable compound event attribution framework (Stott et al 2016, Zscheischler andLehner 2022), we obtain monthly SST, SAT, and geopotential height outputs from 11 climate models that participated in Coupled Model Intercomparison Project Phase 6 (CMIP6) (Eyring et al 2016; see table 1 for basic information on the models used).It contains historical simulation experiments from the Detection and Attribution Model Intercomparison Project (DAMIP) (Gillett et al 2016) of CMIP6 to estimate the anthropogenic influence, including the historical all-forcing (natural + anthropogenic, ALL) and natural-only forcing (NAT), and the Shared Socioeconomic Pathway 3-7.0 (SSP3-7.0)scenario experiment for future projections (O'Neill et al 2016).
Further, we use a bivariate skill score (Ridder et al 2021) to evaluate and select whether the above CMIP6 simulations well reproduce the observed bivariate probability distributions.The skill score (S skill ) essentially compares similarities in the probability density function (PDF) of the compound MHW-THW event between observations (Z obs j,i ) and model simulations (Z mod j,i ).In detail, we calculated a score (ranging between 0 and 1) by dividing their joint PDFs into x × y (x, y ∈ N) two-dimensional bins and summing the minimum values of each bin, where x and y are chosen based on the variables and the recommended bin size according to Scott's Normal (Scott 1979).The S skill was expressed as follows: where Z obs j,i and Z mod j,i indicate the joint PDFs of NWP SSTA and YRB SATA for observations and CMIP6 models, respectively.In general, when S skill > 0.6, it means that a model performs sufficiently and passes the evaluation (see more details in the supplemental material).
This approach encompasses the entire individual distribution of both variables (i.e.SSTA and SATA) constituting the MHW-THW event, while also evaluating their correlation.Based on this method, 11 CMIP6 models with 59 ensemble members were evaluated, resulting in a total of 7 climate models with better performance, whose sets of simulations contained 28 members.That is, the 28 runs for the combined ALL forcing and SSP3-7.0scenario, as well as the 28 runs for NAT forcing (table 1).It is important to note that historical forcing runs only extend until 2014, after which the SSP3-7.0scenarios are employed to extend ALL simulations until 2034.The data over July-August 2010-2034 (a 25-year period centered on the year of the event) are then utilized as samples, representing the present-day climate conditions akin to 2022 (Wang and Sun 2022).This results in 700 samples (25 years × 28 runs) for ALL forcing simulations.For NAT simulations, 25 years (1996NAT simulations, 25 years ( -2020) ) of data closest to the year of the event are correspondingly selected, generating 688 samples because the two ensemble runs of the GISS-E2-1-G model concluded in 2014.Furthermore, future projections under the SSP3-7.0scenario are also divided into the mid-21st century (2041-2065) and late-21st century (2076-2100).To maintain consistency, all the data were bilinearly interpreted into a common 1 • × 1 • grid and converted into anomalies relative to the 1901-1930 mean.

Attribution methods
For large-scale atmospheric circulation, the regions with WNPSH (17 in Z200 were first selected as the key circulation regions (green rectangles in figures 1(e) and (f)), which have proved to be important factors influencing the occurrence of this compound MHW-THW event (Chen andLi 2023, Oh et al 2023).The corresponding correlation analysis also reveals that both SSTA and SATA exhibit significant positive correlations (p < 0.01) with the circulation anomalies (Z500 and Z200) (figure 2).Then we calculated the spatial correlation coefficients between simulated circulation anomalies from ALL and NAT forcings and observed patterns in July-August 2022 over the key regions.Finally, based on the circulation similarity method (Christidis and Stott 2015), the model simulations were divided into two sub-samples: high-correlation (r > 0.5) (including samples ALL_high, NAT_high) and low-correlation (r < 0.1) (including samples ALL_low, NAT_low).For ENSO influences, El Niño is defined as when the July-August SSTA (calculated relative to a 30year running mean) in the Niño-3.4region 5 • S-5 • N, 120 • -170 • W is above or equal to 0.75 SD.Conversely, La Niña is identified when this index value in the same region is below or equal to −0.75 SD (Sun et al 2023).This generates samples with different ENSO phases namely, ALL_El Niño, ALL_La Niña, and ALL_Neutral.We also conducted sensitivity tests on ENSO events defined using different criteria (0.75 SD, 1.0 SD, and 1.5 SD) for ALL forcing and found very similar results.
As in previous work (Min et al 2022, Tang et al 2023a), the univariate (MHW and THW) and joint (MHW-THW) exceedance probabilities were estimated empirically by counting the number of events with NWP SSTA and YRB SATA exceeding their observed respective thresholds and dividing it by the Table 1.Details of CMIP6 simulations used in this study.The realizations of the ALL and NAT simulations that passed the evaluation are marked in bold.Note that r * i1p1f2 ensemble member is available and used for CNRM-CM6-1 and GISS-E2-1-G models.total number of samples.The risk ratio (RR; Fischer and Knutti 2015) was used to assess the influences of anthropogenic warming, atmospheric circulation, and ENSO on the likelihood of the 2022-like compound MHW-THW event.RR was defined as:

Institute/Country
where P 0 is the joint exceedance probability of extreme events under counterfactual climate conditions, and P 1 is that under factual climate conditions.For instance, regarding anthropogenic effects, P 0 and P 1 indicate the occurrence probabilities of similar events equaling or exceeding the previous recordbreaking value (i.e., observed thresholds in July-August 2022; dashed lines in figure 4) for the NAT and ALL samples, respectively.Then, the comparison between both probabilities is defined as RR ALL/NAT .Similarly, RR ALL_high/NAT_high and RR ALL_high/ALL_low were calculated to estimate the contribution of atmospheric circulation under anthropogenic forcings.For ENSO effects, the RRs were calculated as RR ALL_La Niña/ALL_El Niño and RR ALL_La Niña/ALL_Neutral .In addition, the associated RR with 95% confidence intervals was estimated via bootstrapping 1,000 times (Christidis et al 2013).

Unprecedented compound MHW-THW event in 2022
During From a perspective of the dynamic environment, this compound MHW-THW event was associated with quasi-barotropic anticyclonic circulation strengthened by the overlap of the two high-pressure systems (WNPSH and SAH) (figures 1(e) and (f), Oh et al (2023), Tang et al (2023b)), accompanied by descending motion and increasing surface solar radiation.Intuitively, the WNPSH, defined by the 5880 -gpm contour in Z500, strengthened and extended westward in the summer 2022.Meanwhile, the SAH, characterized by the 12 560-gpm contour in Z200, demonstrated an enhancement and eastward expansion.As the key circulation systems for this event, the WNPSH and SAH are moving in the opposite direction and connecting to cover most of the northwestern Pacific Ocean, favoring maintaining the vertical high-pressure anomaly structure, and jointly leading to long-lived extreme heatwaves here.As a result, the NWP SSTA and YRB SATA are significantly spatially related (r = 0.59, p < 0.01) (figure 1(d)).Even after detrending, the correlation remained significant (r = 0.42, p < 0.01), albeit with a slight decrease in the correlation coefficient.Moreover, both SSTA and SATA show significant positive correlations (p < 0.01) with Z500 (figures 2(a) and (b)) and Z200 (figures 2(c) and (d)) anomalies, further confirming the intimate relationship between the 2022 MHW-THW event and anomalous atmospheric circulation patterns.

Model performance assessment
To further evaluate the performance of selected CMIP6 models, we compare the time series and PDFs of regional average July-August NWP SSTA and YRB SATA based on the observations and simulations over 1901-2022 (figure 3).It is found that the CMIP6 models can capture the temporal evolutions of NWP SSTA and YRB SATA (figures 3(a) and (b)).Except for a few very extreme years, their observed ranges are encapsulated by the 5%-95% model spread.Specifically, the standard deviations of SSTA and SATA in simulations are roughly similar to those in observations.The long-term trend of SSTA for ALL simulations (0.075 • C/decade) is slightly lower than the observed trend (0.092 • C/decade).CMIP6 simulations underestimated the warming trend of SSTA, possibly due to the pronounced low estimations of SST rising rates before the late 1970s (Deng et al 2023).In contrast, the trend of SATA is higher for ALL simulations (0.075 • C/decade) than for observations (0.042 • C/decade), which could be a result of the performance and high equilibrium climate sensitivity of some models (Christidis andStott 2022, Yang et al 2023).Furthermore, these selected CMIP6 ALL models better represent the bivariate distribution of SSTA and SATA along with their correlation, via a Cram´er-von Mises (C-VM) test (p > 0.01) (Genest et al 2009) (figure 3(c)).Overall, CMIP6 simulations can reproduce the observed distribution characteristics of July-August NWP SSTA and YRB SATA, which can be used in the subsequent attribution analysis.

Attribution of 2022 compound MHW-THW event
Using selected simulations from CMIP6, we found that not only the univariate distributions of NWP SSTA and YRB SATA but also the bivariate MHW-THW distribution showed an apparent warming shift from NAT to ALL forcing (figure 4(a)).The probabilities of NWP MHW, YRB THW, and MHW-THW were 0.1429, 0.2329, and 0.0871 in ALL forcing simulations, respectively.Notably, no anomaly value exceeded the 2022 thresholds in the NAT world.It suggests that 2022-like individual and compound heatwaves would be extremely impossible to occur without human influence (figure 4(a); table 2).
We further estimate how often a 2022-like MHW-THW event will occur in the future anthropogenic warming (SSP3-7.0scenario).In the mid-21st century (2041-2065), the joint exceedance probability is up to 0.6543 and the associated RR increases to 7.51 (95% confidence intervals: 5.93, 10.04, the same below) (figure 4(d); table 2).In the late 21st century (2076-2100), such events are projected to become much more common, occurring once almost every year and increasing by 11.41 (9.08, 15.13) times the 2020s level (table 2).Though the exceeding probabilities of compound MHW-THW are less than that of univariate event, however, the bivariate RRs are larger than any of the univariate RRs in the future, indicating that anthropogenic forcing has strongly contributed to 2022-like co-occurring heatwave occurrences.

Conditional attributions to circulation, ENSO and mean warming
In terms of conditional attributions for highcorrelation circulation patterns (ALL_high and NAT_high), we also find that human influence has caused major shifts in the distributions toward a hotter regime (figure 4(b); table 2).The likelihood of a 2022-like MHW-THW event is highly unlikely to occur in the NAT_high simulations, consistent with the unconditional results.While it increased to about 0.1277 in the ALL_high simulations, which is greater than the unconditional attribution since anticyclonic circulations favor extreme heatwaves, especially for THW, reaching around 0.4043.Moreover, we compared the distributions of simulated temperature anomalies between high-and low-correlation samples (ALL_high and ALL_low) under ALL forcing (figure 4(c)).The RRs for compound MHW-THW event show that anomalous circulation (considering double conditions of Z500 & Z200) enhanced its occurrence probability by 3.72 (1.24, 11.15) times and stronger than univariate heatwaves (table 2).We also found that the relative contribution to extreme heatwaves is different under the conditions of a single atmospheric circulation pattern (Z500 or Z200) (table S1).Specifically, the Z500 anomaly contributes more to the occurrence of THW, increasing to 2.02 (1.43, 2.80) times.For the Z200 anomaly condition, it affects both MHW and THW, making them 2.43 (1.59, 3.86) and 2.16 (1.57, 3.01) times more likely to occur, but more intensely for increases in MHW-THW with a significant increase of 3.79 (2.17, 8.01).
In terms of ENSO effects, we compared the ALL runs in La Niña years with those in El Niño and Neutral years.The resulting RRs are 1.12 (0.58, 2.24) and 1.45 (0.77, 2.52) for RR ALL_La Niña/ALL_El Niño and RR ALL_La Niña/ALL_Neutral , respectively (figure S3; table 2).It follows that the above RR results were not significantly different from 1.0, implying that the developing phase of La Niña is not likely to change its probability significantly.Finally, a similar attribution analysis was conducted with detrended NWP SSTA and YRB SATA.The distributions of SSTA and SATA exhibit almost coincident patterns (figure S2), with estimated RR ALL/NAT of 1.31 (0.44, 4.91) (figure S2(a)), revealing that detrended temperature anomalies cannot be attributed to human-induced warming.It means that the increased likelihood of the 2022 compound MHW-THW event is determined by the warming mean climate instead of changes in the variance of SST and SAT.The same result could be achieved from the projected future changes in the middle of the 21st century, with an RR of 0.75 (0.20, 2.67) (figure S2(d)).On the contrary, the more intense RRs for compound MHW-THW event under the conditions of high-and low-correlation circulations, the RR ALL_high/NAT_high and RR ALL_high/ALL_low were 3.13 (0.20, +∞) and 16.72 (0, +∞), respectively (figures S2(b) and (c)).

Conclusions and discussion
In this study, we first detected and defined the extreme characteristics of compound MHW-THW event across the northwest Pacific in the summer of 2022, and then evaluated the contributions of anthropogenic forcing, atmospheric circulations, and ENSO on the compound event from CMIP6 multimodel simulations.
The results show that the unprecedented July-August 2022 compound MHW-THW event was extremely rare without anthropogenic forcing (figure 4(a); table 2), and that the increased likelihood Our results are generally consistent with recent studies that reported the significant combined roles of anthropogenic forcing and natural variability on the likelihood of MHW or THW in 2022, respectively (Hua et al 2023, Oh et al 2023, Tan et al 2023).For instance, Oh et al (2023) revealed that an anomalous anticyclone pattern with strengthened high pressure played a major role in the record-breaking longlasting ECS MHW during summer 2022 rather than a 'triple-dip' La Niña.Hua et al (2023) suggested that the 2022-like THW in the middle reaches of the YRB was favored by atmospheric circulation and persistent high-pressure anomalies, and further enhanced by anthropogenic forcing, which together increase the probability of such events by 6-7 times.Additionally, we also considered the effect of the Indian Ocean Dipole (IOD) on the likelihood of these extreme heatwaves.Recent research has suggested that the ECS MHWs are closely linked to the negative IOD phase (Tan and Cai 2018, Tan et al 2023).In contrast, the positive IOD phase dominates in enhancing THWs across China, particularly in the west and north regions (Wei et al 2020, 2023, Pan et al 2024).For the compound MHW-THW event, we found that the IOD is not likely to significantly affect the probability of the 2022-like compound heatwave (figure S4; table S2).
However, for the first time, our study quantitatively assesses the contribution of different external forcing factors (anthropogenic and natural-only forcings) and internal natural variability (atmospheric circulation and ENSO) to the occurrence of a 2022-like compound MHW-THW event, based on the compound event attribution framework (Zscheischler and Lehner 2022).We also applied univariate and bivariate skill score methods to evaluate the models in capturing both marginal distributions and bi-variate dependence, which improve the attribution reliability for the compound event (Ridder et al 2021).Overall, our study promotes the understanding of drivers for the MHW-THW event, more importantly, it highlights the urgent need for the stakeholders and policymakers to build mitigation strategies for the type of compound event.
Nonetheless, it is essential to acknowledge several potential limitations and uncertainties in this analysis.Firstly, inherent biases in the response of climate models to forcings may impede a comprehensive explanation of the weather systems associated with extreme events, as recognized in previous literature (Trenberth et al 2015, van Oldenborgh et al 2021, Hua et al 2023).Moreover, uncertainty arises from estimating the likelihood of exceptionally rare events using a limited data sample.Therefore, further studies should consider using simulations from the single model initial-condition large ensembles or reanalysis to reconstruct and expand samples (Bevacqua et al 2020, 2023, Batibeniz et al 2023), which can increase our confidence in attribution statements.It is worth mentioning that the further extension of a combined framework integrating the storyline-probability approach is a promising method in the field of event attribution, which allows us to address challenges associated with low probability, event characterization, model limitations, uncertain climate change impacts on circulations, and the physical interpretability of attribution results (Fischer et al 2023, Qian et al 2023, Hao and Chen 2024).
reaching +1.72 • C (red bars in figure 1(c)).Meanwhile, the THW impacted the Yangtze River Basin (YRB; 23 • -38 • N, 92 • -122 • E; green rectangle in figure 1(b)), featuring surface air temperature anomalies (SATA) up to +1.71 • C (black solid line in figure 1(c)).In the study region, SSTA and SATA simultaneously exceeded 2.5 SDs, which is the greatest value recorded since 1901 (figures 1(c) and (d)).This compound MHW-THW event affected over 39 million people, damaged 4.27 million hectares of crops, and caused direct economic losses of 5.7 billion USD (MEMC 2023).Previous studies have separately revealed the contribution of multiple drivers to these MHWs (Oh et al 2023, Tan et al 2023) or THWs (Jiang et al 2023, Zhang et al 2023, Tang et al 2023b) over the northwest Pacific in summer 2022.For example, Tan et al (2023) reported that the record-breaking long-lasting MHW occurred in the East China Seas (ECS) during the summer 2022 will be more frequent and intense over the coming decades due to anthropogenic climate change.Jiang et al (2023) and Liang et al (2024) each indicated that the long-term trend in global warming may explain about 30%-40% of total THW days.

Figure 1 .
Figure 1.Spatial distribution of (a) SSTA and (b) SATA in the summer (July-August) of 2022.The green rectangles indicate the NWP (22 • -47 • N, 117 • -151 • E) and YRB (23 • -38 • N, 92 • -122 • E).(c) Time series of regionally averaged July-August NWP SSTA (red bars) and YRB SATA (black solid line) during 1901-2022.The dashed lines represent their respective 2 standard deviations.(d) Scatter plot depicting the relationship between July-August NWP SSTA and YRB SATA, with a focus on the year 2022.Linear regression fits are denoted by red lines, and correlation coefficients and significance levels are provided in the upper-left corner, with significance levels in brackets.The 90th-percentile thresholds for univariate extremes are indicated by gray lines.Anomalies of (e) Z500 and (f) Z200 (shaded) in July-August 2022 are displayed.The green rectangles indicate the key circulation regions, including the western North Pacific subtropical high (17 • -36 • N, 104 • -146 • E) in Z500 and the South Asian high (23 • -39 • N, 78 • -126 • E) in Z200.Red and black dashed lines show the 5880 (12 560) gpm line for 2022 and climatology (1901-2021), respectively.All anomalies are relative to the 1901-1930 climatology.

Figure 2 .
Figure 2. Scatter plot for the (a) July-August NWP SSTA and (b) YRB SATA with observed Z500 anomaly (relative to 1901-1930), respectively.(c), (d) As in (a), (b), but for Z200 anomaly.Red line denotes linear regression fits, correlation coefficients and significance levels calculated before and after detrending displayed on the upper-left corner, with the latter in brackets.The red stars indicate the case of 2022.

Figure 3 .
Figure 3.Time series of (a) July-August NWP SSTA and (b) YRB SATA for observation (black), and CMIP6 simulations with ALL (red) and NAT (blue) from 1901 to 2022.Solid lines denote the results of the multi-model ensemble mean.The shaded areas represent the 5%-95% ranges of the individual model members.(c) Univariate and bivariate distributions of the NWP SSTA and YRB SATA from observation (black) and CMIP6 ALL (red) simulations.The contour lines, from innermost to outermost, represent 5%, 25%, 50%, 75%, and 95% of the bivariate distributions.

Table 2 .
The univariate (for SSTA and SATA) and bivariate exceedance probabilities exceed the observed 2022 thresholds and corresponding risk ratio (RR) values under the effects of anthropogenic forcing, atmospheric circulations, and ENSO.Above the effects, Roman numerals (e.g.I, II, III-1) label the different simulations or conditions.Parentheses indicate the 5%-95% uncertainty ranges of RR.