1. Introduction
How the tropical Pacific climate responds to rising greenhouse gases (GHGs) has been studied extensively during the past decades. Most state-of-the-art climate models project, for example, a weakened mean-state west-minus-east sea surface temperature (SST) gradient (Ying et al. 2016; Cai et al. 2021), an enhanced mean-state rainfall change in the eastern equatorial Pacific that generally follows a “warmer get wetter” paradigm (Xie et al. 2010; Huang 2014), and an increased El Niño–Southern Oscillation (ENSO)-related SST and rainfall variability (Power et al. 2013; Cai et al. 2018, 2022). These anthropogenically caused climate change signals drive important changes not only in the Pacific, but also beyond the region through ENSO teleconnections across the globe (Delage and Power 2020; Power et al. 2021), exerting potentially severe impacts on ecosystems, agriculture and social development worldwide (Collins et al. 2010). However, the robustness of these emerging signals relative to their background noise is not as well studied as have been the signals themselves. Although modest climate change signals occurring well before a “tipping point” are believed to be large and rapid enough to have profound impacts on human beings (Power 2014; Power et al. 2017), it is only when the signals emerge out of their corresponding background noise of natural variability, can we believe that such signals are detectable or emergent (Hawkins and Sutton 2012). This time of emergence (ToE)—or time of expulsion (Power 2014)—of tropical Pacific climate change signal is both important and urgent for adaptation planning and risk assessment (Giorgi and Bi 2009; Hawkins and Sutton 2012).
There have been few studies focusing on the emergence of climate change signals in the tropical Pacific. A pioneering study by Timmermann (1999) indicated that the observed changes in ENSO SST variability are within the range of natural variability, implying that such signal was, at that moment, undetectable. DiNezio et al. (2013) found that the weakened Pacific Walker circulation in response to GHGs forcing can be detected in observational record and in a subset of phase 5 of Coupled Model Intercomparison Project (CMIP5) historical simulations. Wang et al. (2015) revealed that the observed anthropogenic surface warming in the western Pacific has already emerged due to the local relatively low natural variability. Recently, Ying et al. (2022) showed, based on the latest phase 6 of CMIP (CMIP6) multimodels, that the estimated ToE of annual-mean SST is earlier than that of annual-mean rainfall, whereas the ToE of ENSO rainfall is earlier than that of ENSO SST. Although these qualitative findings are robust in the multimodel ensemble mean (MEM) result, as well as across most individual models, there remains considerable uncertainty concerning the exact ToEs for these signals. For example, the intermodel standard deviations for both the ToEs of ENSO SST and rainfall signals under a fixed emission scenario reach up to 50 years (Ying et al. 2022), which limits their applications in early climate warning. Thus, attributing and quantifying the sources of uncertainty in the projected ToEs of tropical Pacific climate change signals is required to obtain a more reliable climate projection.
In general, uncertainty in climate model projection arises from three sources: emission scenario, model differences and internal variability (Hawkins and Sutton 2009). With projections constrained on a given emission scenario, progress can be made in attributing uncertainty by focusing on the latter two. Among them, the internal variability, which is defined as the intrinsic fluctuation of climate system, plays a nonnegligible role in model projections by obscuring climate signals in response to external forcing (Hawkins and Sutton 2009; Deser et al. 2012). For example, Hawkins and Sutton (2011) found that the internal variability is the dominant source of uncertainty for decadal rainfall changes in the early few decades. Deser et al. (2012) revealed that the internal variability explains at least half of the intermodel spread in the projected surface air temperature, sea level pressure and precipitation trends during 2005–60 in the CMIP3 multimodels. Thus, it is important to investigate to what extent the internal variability impacts the estimations of exact ToEs of tropical Pacific climate change signals.
A large amount of research has proposed that internal variability could modulate both the tropical Pacific annual-mean state and ENSO changes under global warming. The eastern Pacific cooling during the past three decades are postulated to be linked to the negative phase of interdecadal Pacific oscillation (IPO), contributing to a global warming hiatus that lasted for more than a decade (Kosaka and Xie 2013; England et al. 2014). The IPO also appears to play a role in the near-future projections of tropical Pacific mean-state west-minus-east SST gradient and Walker circulation changes, which both exhibit reversed trends compared with historical simulations due to the near-future phase transition of IPO (Watanabe et al. 2020; Wu et al. 2021). Internal variability is also found to be important for ENSO changes in both observational records (Capotondi and Sardeshmukh 2017; Dieppois et al. 2021) and model simulations (Wittenberg 2009; Stevenson 2012; Wittenberg et al. 2014; Planton et al. 2021). Specifically, the projected ENSO SST change is significantly influenced by internal variability, with different initial conditions in a specified large ensemble (LE) model exhibiting both positive and negative ENSO SST variance changes (Zheng et al. 2017; Maher et al. 2018). These imply that internal variability could be a potentially important source of uncertainty in the projected ToEs of tropical Pacific climate change signals.
Quantifying the contribution of internal variability to uncertainty in the projected ToE requires LE members of a climate model with different initial conditions. It has been suggested that approximately 30–40 ensemble members of a single model are needed when performing a multimember ensemble mean (MMM) to get rid of the effect of internal variability on ENSO changes under global warming (Maher et al. 2018). Accordingly, although more LE model outputs could be used (Maher et al. 2022), here we choose three LE models with 40 or more members to contrast the role of internal variability in the uncertainty of the ToEs of tropical Pacific annual-mean SST and rainfall, and ENSO-related SST and rainfall signals. We have a specific focus on the eastern equatorial Pacific (EEP) region where the key area of ENSO events is located. By comparing the intermember differences in the LE models with the intermodel differences in 29 different CMIP6 models, it is found that the internal variability plays a role in both the uncertainty in the ToEs of tropical Pacific annual-mean state and ENSO variability but with different levels of contribution.
The rest of the paper is organized as follows. Section 2 describes the datasets and methods used in the study. Section 3 displays the main results, including the role of internal variability on the uncertainty in the estimated ToEs of annual-mean SST and rainfall, and ENSO SST and rainfall. A summary and discussion is given in section 4.
2. Data and methods
a. Datasets
Outputs from the Community Earth System Model version 1 large ensemble (CESM1-LE) with 40 members are used in this study, including the historical run from 1920 to 2005 and the representative concentration pathway 8.5 (RCP8.5) run from 2006 to 2099. To make further comparison, we select another two LE simulations: the historical (1950–2005) and RCP8.5 (2006–99) runs of 50 members from the Canadian Earth System Model version 2 large ensemble project (CanESM2-LE) and the historical (1870–2005) and RCP8.5 (2006–99) runs of 100 members from the Max Planck Institute for Meteorology Earth System Model version 1.1 large ensemble project (MPI-ESM-LE). In addition, the preindustrial control (piControl) runs for the last 500 years from the above three LE models are also used. Note that model CESM1-LE contains only 319 years in the piControl run.
To compare the uncertainty arising from internal variability with the total uncertainty from both the internal variability and model, outputs from 29 CMIP6 models are used, including the historical run for 1870–2014, the shared socioeconomic pathway 5–8.5 (SSP5–8.5) emission scenario run for 2015–99, and the piControl run for the last 500 years. For each model, we use only one ensemble member run (r1i1p1f1) since many modeling groups only provide this single number of ensemble members for a given forcing scenario. Details of the models can be found in Table 1 (Eyring et al. 2016). Note that the model CESM2-WACCM, CNRM-CM6-1-HR, and KACE-1-0-G contain only 499, 300, and 450 years in the piControl run, respectively. The variables include the monthly surface temperature (ts, which is SST for the open ocean) and rainfall (pr). All the outputs from the LE models and CMIP6 models are bilinear interpolated into a 2.5° × 2.5° grid before analysis.
Information of CMIP6 models used in this study. The names, the associated institutions and countries, and ensemble members of 29 CMIP6 models used in this study (mostly r1i1p1f1, with different ensembles highlighted in boldface font).
b. Methods
The choice of reference period is somewhat arbitrary, balancing the need to see a robust global warming signal in models and availability of model outputs. We select the period of 1950–79 as the reference time for all the LE and CMIP6 models (note that the outputs from CanESM2-LE start in 1950). This reference period is different from that of 1960–89 used in our original paper (Ying et al. 2022) but the results are relatively insensitive to the choice. All the changes under global warming are obtained by subtracting the mean values in the reference time. Annual-mean values are calculated by simply averaging the monthly mean outputs from January to December, while interannual anomalies are calculated by first removing the climatological annual cycle of the chosen period and then quadratic detrending the remaining monthly values (Cai et al. 2020). To highlight the spatial difference of annual-mean SST change, the annual-mean SST for each grid point is decomposed into the tropical-mean (20°S–20°N) SST (Ttm) and the relative SST (Tr) with the tropical-mean SST removed (Xie et al. 2010).
1) Time of emergence of annual-mean SST and rainfall signals
2) Time of emergence of ENSO SST and rainfall signals
The definition of ENSO rainfall pattern is dependent on the ENSO SST pattern and the standardized first PCs. The interannual rainfall anomalies are regressed onto the standardized PCs of ENSO SST in each 30-yr time window, and the regression pattern defines the ENSO rainfall pattern (Huang and Xie 2015). The calculations for the signal, noise and ToE of ENSO rainfall change are similar to those for ENSO SST change shown above. Details about the above methods can be found in Ying et al. (2022).
3) Estimating the uncertainty
The total uncertainty of ToE under a given emission scenario, including the uncertainty that arises from both model difference and internal variability, is estimated by calculating the intermodel STD based on the selected 29 CMIP6 models. The uncertainty due to internal variability for each LE model is estimated by the corresponding intermember STD. For each LE model, the contribution of internal variability to the total uncertainty is defined as the ratio between the intermember STD and the intermodel STD. We note that the total uncertainty calculated by the intermodel STD could be underestimated due to the only one ensemble member (Table 1) of simulation from each model (Maher et al. 2021), but the relative importance of the internal variability in the total uncertainty can still be evaluated reasonably from the ratio between the intermember STD and the intermodel STD (Dong et al. 2021).
3. Results
a. Role of internal variability in the uncertainty of the ToE of annual-mean SST and rainfall signals
Given the El Niño–like response with an enhanced warming in the EEP and an equatorial enhanced background noise due to ENSO variability in both the CMIP6 multimodel ensemble mean (MEM) and CESM1-LE multimember ensemble mean (MMM) (Figs. 1a,b), the spatial patterns of ToE of annual-mean SST in both ensembles denote that annual-mean SST signals should have been detected across most of the tropical Pacific by 2020 (Figs. 1c,d). Yet there are some discrepancies in the ToE of annual-mean SST in difference models and members (Figs. 1e,f). The intermodel STDs in the ToE of annual-mean SST exceed 10 years in many parts of the tropical Pacific (Fig. 1e), while the intermember STDs in the ToE of annual-mean SST are less than 5 years across most parts of the tropical Pacific (Fig. 1f). This means that the total uncertainty in the ToE of annual-mean SST explained by the internal variability is smaller than 50% of the total uncertainty in most parts of the tropical Pacific (Fig. 1f, contours). For the EEP region, the explained variance is only 21%, indicating that the internal variability contributes only a small part to the total uncertainty in the ToE of annual-mean SST. Thus, the internal variability has a relatively minor impact on the projection of ToE of annual-mean SST, and model differences dominate the total uncertainty in the ToE of annual-mean SST.
The response of annual-mean SST to global warming in (a) CMIP6 multimodel ensemble mean (MEM) and (b) CESM1-LE multimember ensemble mean (MMM); contours in (a) and (b) denote the MEM and MMM noise of annual-mean SST (units: K, with an interval of 0.1 K), respectively. The ToE of annual-mean SST in (c) CMIP6 MEM and (d) CESM1-LE MMM; the shaded regions in (c) and (d) indicate that more than two-thirds of models or members have emergent signals by the year 2099 and are plotted by averaging all the models or members that show the local ToE by 2099. (e) The intermodel STD of ToE of annual-mean SST in CMIP6 models and (f) the intermember STD of ToE of annual-mean SST in CESM1-LE in the regions where more than two-thirds of models or members have emergent signals by the year 2099; contours in (f) are the ratio of the intermember STD of ToE of annual-mean SST in CESM1-LE to the intermodel STD of ToE of annual-mean SST in CMIP6 models, shown as percentages (units: %, with an interval of 10%). The green box in each plot represents the eastern equatorial Pacific (EEP, 2.5°S–2.5°N, 180°–90°W) region.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
On the other hand, the response of annual-mean rainfall to global warming displays a consistent enhancement along the equatorial Pacific with the largest value in the central Pacific around the date line in both CMIP6 MEM and CESM1-LE MMM (Figs. 2a,b, shaded), while the background noise of annual-mean rainfall peaks in the equatorial central-western Pacific, owing to ENSO-driven rainfall variability (Figs. 2a,b, contours). The estimated spatial patterns of ToE in most CMIP6 models and CESM1-LE members reveal that the annual-mean rainfall signal is expected to emerge earliest in the EEP region within the coming decades, while the signals in most of the other regions, including the intertropical convergence zone and South Pacific convergence zone, are undetectable before the end of this century under the high SSP5–8.5 or RCP8.5 scenarios (Figs. 2c,d). Over the EEP region where detectable signals emerge in most CMIP6 models and CESM1-LE members, the intermodel STDs in the ToE of annual-mean rainfall are generally more than 20 years (Fig. 2e), larger than those in the ToE of annual-mean SST (Fig. 1e). This indicates a larger total uncertainty in the estimated ToE of annual-mean rainfall than that of annual-mean SST. Meanwhile, the intermember STDs in the ToE of annual-mean rainfall appears to be also larger than that in the ToE of annual-mean SST (Figs. 1f and 2f). However, the total uncertainty in the ToE of annual-mean rainfall explained by internal variability over the EEP reaches 37% (Fig. 2f, contours), which is almost twice as large as that in the ToE of annual-mean SST explained by internal variability (Fig. 1f, contours). This implies the internal variability plays a more important role in the estimation of ToE of annual-mean rainfall compared with that of annual-mean SST. In addition, the less than half of the total uncertainty explained by the internal variability in CESM1-LE indicates that model differences also dominate the total uncertainty in the ToE of annual-mean rainfall.
The response of annual-mean rainfall to global warming in (a) CMIP6 MEM and (b) CESM1-LE MMM; contours in (a) and (b) denote the MEM and MMM noise of annual-mean rainfall (units: mm day−1, with an interval of 0.4 mm day−1), respectively; stippling in (a) and (b) indicates that more than two-thirds of models or members have the same sign. The ToE of annual-mean rainfall in (c) CMIP6 MEM and (d) CESM1-LE MMM; the shaded regions in (c) and (d) indicate that more than two-thirds of models or members have emergent signals by the year 2099 and are plotted by averaging all the models or members that show the local ToE by 2099. (e) The intermodel STD of ToE of annual-mean rainfall in CMIP6 models and (f) the intermember STD of ToE of annual-mean rainfall in CESM1-LE in the regions where more than two-thirds of models or members have emergent signals by the year 2099; contours in (f) are the ratio of the intermember STD of ToE of annual-mean rainfall in CESM1-LE to the intermodel STD of ToE of annual-mean rainfall in CMIP6 models, shown as percentages (units: %, with an interval of 10%). The pink contour in each plot denotes 7 mm day−1 MEM or MMM rainfall climatology in the reference period of 1950–79. The green box in each plot represents the EEP region.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
For the CanESM2-LE and MPI-ESM-LE, similar results can be found in terms of the ensemble mean responses of annual-mean SST and rainfall to global warming (Figs. 3a,b; Figs. 4a,b) as well as their corresponding ToEs (Figs. 3c,d; Figs. 4c,d), although differences can be found in the details. In addition, the intermember STDs in the ToE of either annual-mean SST or rainfall in the two LE models differ from those in CESM1-LE to some degree (Figs. 3e,f; Figs. 4e,f). Especially for MPI-ESM-LE, the intermember STDs of ToEs of both annual-mean SST and rainfall are much larger than the corresponding ones in CESM1-LE (Figs. 3f and 4f). Nevertheless, the result of a larger contribution of internal variability to the total uncertainty in the ToE of annual-mean rainfall than to the total uncertainty in the ToE of annual-mean SST is still robust (Figs. 3e,f and 4e,f, contours). Moreover, the total uncertainty in the ToEs of annual-mean SST and rainfall explained by internal variability in these two LEs are generally less than 50% over the EEP region, except for the ToE of annual-mean rainfall in MPI-ESM-LE that covers only half of the EEP region (Fig. 4f). This further indicates that model differences play a major part in the total uncertainty of ToE of both annual-mean SST and rainfall.
As in Figs. 1b, 1d, and 1f, but for (a),(c),(e) CanESM2-LE and (b),(d),(f) MPI-ESM-LE.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
As in Figs. 2b, 2d, and 2f, but for (a),(c),(e) CanESM2-LE and (b),(d),(f) MPI-ESM-LE.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
When a larger threshold, say, 2.0 (−2.0), is chosen, the estimated ToE of an annual-mean value would, by definition, be later. The ToEs of annual-mean SST over the EEP in most CMIP6 models and three LE members (Figs. S1a,b and S2a,b), as well as the ToEs of annual-mean rainfall over the EEP in most CESM1-LE (Fig. S1f) and CanESM2-LE members (Fig. S2e) are generally later by around 20–30 years using the 2.0 (−2.0) threshold for
Why is the total uncertainty explained by the internal variability larger in the ToE of annual-mean rainfall than in the ToE of annual-mean SST? By decomposing the gridpoint annual-mean SST into the tropical-mean and the relative SST (the deviation from the tropical-mean), and focusing on the EEP region, it is shown that the ToE of annual-mean SST is both significantly correlated with the ToE of tropical-mean SST (Figs. 5a–c) and with the relative SST (Figs. 5d,e) across members. Members with an earlier ToE of annual-mean SST corresponds to both an earlier ToE of tropical-mean SST and an earlier ToE of relative SST and vice versa. Nevertheless, the ToE of annual-mean SST is found to be more correlated with that of tropical-mean SST across members in all the three LEs (Figs. 5a–c), suggesting a dominant role of ToE of tropical-mean SST in determining the ToE of annual-mean SST in the EEP.
Intermember relationship between the ToE of the EEP annual-mean SST and (a)–(c) that of tropical-mean SST and (d),(e) EEP-relative SST in (a),(d) CESM1-LE; (b),(e) CanESM2-LE; and (c) MPI-ESM-LE. The gray and black markers in each plot indicate that members have emergent signals by the year 2099 and the MMM result based on them, respectively. The black solid line in each plot denotes the intermodel or intermember linear regression. The horizontal (vertical) red bars and the rhombic boxes in the center denote the standard deviations and means of the ToE, respectively. The red digit in the top-right corner of each plot denotes the intermember correlation coefficient, which is significant at the 99% confidence level based on the Student’s t test. The abscissa and ordinate in each plot are arranged to have the same interval to make easy comparisons with each other. Note that 40 and 47 members show detectable signals of EEP-relative SST by 2099 in CESM1-LE and CanESM2-LE, respectively, while none of them in MPI-ESM-LE show detectable signals of EEP-relative SST by 2099.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
For the annual-mean rainfall, the ToE is more correlated with the ToE of relative SST (Figs. 6d,e) rather than with the ToE of tropical-mean SST across members (Figs. 6a–c). In fact, the correlations between the ToE of annual-mean rainfall and that of tropical-mean SST across members are statistically insignificant in CESM1-LE and MPI-ESM-LE (Figs. 6a,c). Thus, members with an earlier ToE of relative SST tend to exhibit an earlier ToE of annual-mean rainfall and vice versa. This is consistent with the hypothesis that members with an enhanced local SST warming compared with the tropical-mean would cause a greater increase in local rainfall under global warming by driving a stronger local ascending circulation (Ying et al. 2022) [“warmer gets wetter” paradigm (Xie et al. 2010)]. Considering that the rainfall response to global warming over the EEP is dominated by its dynamical part through a changing atmospheric circulation response (Ying et al. 2022), it is the ToE of relative SST that mainly matters the ToE of annual-mean rainfall. Thus, the intermember STD in the ToE of annual-mean SST is mainly contributed by that of the ToE of tropical-mean SST, while the intermember STD in the ToE of annual-mean rainfall has a greater contribution from that of the ToE of relative SST. As the intermember STD in the ToE of EEP-relative SST is much larger than that in the ToE of tropical-mean SST (Fig. 5, horizontal red bars), the intermember STD in the ToE of annual-mean rainfall is larger than that in the ToE of annual-mean SST (Figs. 5 and 6, vertical red bars), leading to a larger contribution from internal variability to the total uncertainty in the ToE of annual-mean rainfall than to the total uncertainty in the ToE of annual-mean SST.
As in Fig. 5, but for the intermember relationship between the ToE of the EEP annual-mean rainfall and that of (a)–(c) tropical-mean SST and (d),(e) EEP-relative SST. The red (black) digit in the top-right corner of each plot denotes the intermember correlation coefficient, which is significant (insignificant) at the 99% confidence level based on the Student’s t test.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
b. Role of internal variability in the uncertainty of the ToE of ENSO SST and rainfall signals
The responses of both ENSO SST and rainfall to global warming in most CMIP6 models and CESM1-LE members display an enhancement over the EEP (Figs. 7a,b; Figs. 8a,b), consistent with previous studies (Zheng et al. 2017; Cai et al. 2021). The spatial patterns of ToEs of both ENSO SST and rainfall obtained by most CMIP6 models and CESM1-LE members denote that detectable signals by this century are confined to a small part over the EEP region (Figs. 7c,d; Figs. 8c,d). For individual models or members (Figs. 7e,f; Figs. 8e,f), a few of them do not show a ToE over the EEP region (e.g., INM-CM5-0, UKESM1-0-LL, and the twelfth member in CESM1-LE), indicating that the corresponding signals do not emerge by the last 30-yr time window of 2070–99. Meanwhile, there are also a few other models or members showing a ToE by the first 30-yr time window (e.g., CanESM5 and the sixth member in CESM1-LE), possibly owing to internal variability. These imply both large intermodel and intermember uncertainties are involved in the estimated ToEs of ENSO SST and rainfall.
The response of ENSO SST to global warming in (a) CMIP6 multimodel ensemble mean (MEM) and (b) CESM1-LE multimember ensemble mean (MMM); stippling in (a) and (b) indicates that more than two-thirds of models or members have the same sign. The ToE of ENSO SST in (c) CMIP6 MEM and (d) CESM1-LE MMM. The shaded regions in (c) and (d) indicate that more than two-thirds of models or members have emergent signals and are plotted by averaging all the models or members that show the local ToE. The ToE of ENSO SST over the EEP region in (e) 29 CMIP6 models as well as the MEM result and (f) 40 CESM1-LE members as well as the MMM result; missing bars indicate that signals do not emerge by the last time window of 2070–99; the horizontal dashed line in (f) denotes the first ending year of a 30-yr time window for CESM1-LE; the red error bar denotes one intermember STD, which is plotted when more than two-thirds of total models or members show emergent signals. The green boxes in (a)–(d) represent the EEP region.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
As in Fig. 7, but for ENSO rainfall in (a),(c),(e) CMIP6 models and (b),(d),(f) CESM1-LE.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
Indeed, the intermodel STD in the ToE of ENSO SST over the EEP, based on 22 CMIP6 models that have emergent signals by the last 30-yr time window of this century, reveals that the total uncertainty in the ToE of ENSO SST reaches up to around 50 years (Fig. 7e, red bar). The intermodel STD in the ToE of ENSO rainfall over the EEP, based on 27 CMIP6 models that have emergent signals by the last 30-yr time window of this century, reveals that the total uncertainty in the ToE of ENSO rainfall reaches up to around 42 years (Fig. 8e, red bar). This indicates a larger uncertainty in the estimated ToEs of ENSO SST and rainfall compared with that in the estimated ToEs of annual-mean SST and rainfall by CMIP6 models (Figs. 1e and 2e). On the other hand, the intermember STD in the ToE of ENSO SST over the EEP based on 34 out of 40 members in CESM1-LE that have emergent signals is about 40 years (Fig. 7f), explaining around 80% of the total uncertainty in the ToE of ENSO SST. The intermember STD in the ToE of ENSO rainfall over the EEP based on 39 out of 40 members in CESM1-LE that have emergent signals is about 32 years (Fig. 7f), explaining around 77% of the total uncertainty in the ToE of ENSO rainfall. These imply the internal variability plays a major role in both the total uncertainty in the ToEs of ENSO SST and rainfall.
The other two LEs, however, do not show a similar ENSO response to global warming (Figs. 9a,b; Figs. 10a,b) and thus do not show similar ToEs (Figs. 9c,d; Figs. 10c,d) as CESM1-LE, indicating a large intermodel diversity. For example, the response of ENSO SST to global warming in MPI-ESM-LE is rather weak with no intermember consistency in the EEP (Fig. 9b), while that in CanESM2-LE exhibits a consistent weakening (Fig. 9a). Moreover, the percentages of members that have detectable ENSO SST or rainfall signals by the last 30-yr time window of this century are smaller in the two LEs than those in the CESM1-LE (Figs. 9e,f; Figs. 10e,f). In more than two thirds of the total members having emergent signals, these appear only in the ENSO rainfall change in CanESM2-LE (Fig. 10e), while less than half of the total members in MPI-ESM-LE have an emergent ENSO SST signal (Fig. 9f). The limited number of members with emergent signals hampers a reliable estimation of the effect of internal variability based on intermember STD of the ToEs. Nevertheless, the intermember STD in the ToE of ENSO rainfall over the EEP, based on 42 out of 50 members in CanESM2-LE, with emergent signals by the last 30-yr time window of this century, reaches about 31 years (Fig. 9e), explaining around 74% of the total uncertainty in the ToE of ENSO rainfall based on CMIP6 models. This further verifies that the internal variability contributes a major part to the total uncertainty in the ToE of ENSO rainfall, which is inferred from CESM1-LE.
As in Figs. 7b, 7d, and 7f, but for (a),(c),(e) CanESM2-LE and (b),(d),(f) MPI-ESM-LE. The horizontal dashed line in (e) denotes the first ending year of a 30-yr time window for CanESM2-LE.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
As in Figs. 8b, 8d, and 8f, but for (a),(c),(e) CanESM2-LE and (b),(d),(f) MPI-ESM-LE. The horizontal dashed line in (e) denotes the first ending year of a 30-yr time window.
Citation: Journal of Climate 36, 8; 10.1175/JCLI-D-22-0554.1
The estimated ToEs of ENSO SST and rainfall would be later when a more demanding threshold, say the 97.5th (2.5th) or 99th (1st) percentile, is chosen (Figs. S3 and S4). Specifically, more models or members do not show a ToE of ENSO SST by the last 30-yr time window of 2070–99 when the threshold is changed from 95th (5th) to 97.5th (2.5th) or 99th (1st) percentile (Figs. S3c,g; Figs. S4c,g). Nevertheless, the intermodel STDs, as well as the intermember STDs in both the ToEs of ENSO SST and rainfall over the EEP based on a 97.5th (2.5th) or 99th (1st) percentile threshold are comparable to those based on a 95th (5th) percentile threshold, and the intermember STDs still account for a dominant part of the intermodel STDs (Figs. S3c,d,g,h; Figs. S4d,g,h, red bars). This indicates that the dominant role of internal variability in the total uncertainty of ToEs of ENSO SST and rainfall is insensitive to the threshold.
4. Summary and discussion
Reliable estimations of the ToEs of climate change signals in the tropical Pacific are of great significance for risk assessment and early climate warning. Here we show that large uncertainties are involved in the estimated ToEs of tropical Pacific annual-mean SST and rainfall, and ENSO SST and rainfall based on CMIP6 multimodel ensemble. Uncertainties are on the order of decades for emergence of annual mean signals and on the order of multidecades for the ENSO-related signals. Internal variability is found to play a role in both the uncertainties in the ToE of tropical Pacific annual-mean state and ENSO variability based on three large ensembles, but with different levels of importance. For the annual-mean SST and rainfall, internal variability generally explains less than 50% in the total uncertainty of the two ToEs over the EEP in all three LEs, hence model difference plays a more important role. While for the ENSO SST and rainfall, internal variability plays a dominant role in the total uncertainty of the two ToEs over the EEP in only one LE model (CESM1-LE) that has sufficient members to show emergent signals. Such conclusions are insensitive to the threshold of defining a ToE.
A comparison of the total uncertainty explained by the internal variability between the ToE of annual-mean SST and that of annual-mean rainfall reveals that the internal variability explains more uncertainty of the latter than the former. It is shown that the intermember uncertainty in the ToE of annual-mean SST is mainly contributed by that of the tropical-mean SST, while the intermember uncertainty in the ToE of annual-mean rainfall is more contributed by that of the relative SST. Due to a much smaller intermember uncertainty in the ToE of tropical-mean SST than that in the ToE of relative SST, the intermember uncertainty in the ToE of annual-mean SST is smaller than that in the ToE of annual-mean rainfall, leading to a larger contribution from internal variability to the total uncertainty in the ToE of annual-mean rainfall than to the total uncertainty in the ToE of annual-mean SST.
The dominant roles of internal variability in both the total uncertainty of ToEs of ENSO SST and rainfall are most prominent in CESM1-LE simulations. The two other LEs—CanESM2-LE and MPI-ESM-LE—however, do not show as clear a result as in CESM1-LE due to too few members showing emergent signals, indicating a large intermodel diversity in estimating the ToEs of ENSO SST and rainfall. While an improvement on the estimations of ToE of ENSO SST and rainfall can be expected if more LE simulations are available, due to the dominant role of internal variability exhibited by CESM1-LE, model diversity is still important.
Given the difference in importance of model and internal variability on the modeled ToEs of annual-mean and ENSO signals, and the fact that both the observed signals of annual-mean SST and ENSO SST are undetectable in historical record (Ying et al. 2022), what are the implications for this analysis on the detection of these signals in the real world? If an emergent annual-mean signal is observed in the next, say, 20 years, we could claim that climate change has impacted the annual-mean state, and the anthropogenic signal is detectable at that time, with relatively high confidence. This is due to the effect of internal variability on the ToE of annual-mean SST and rainfall being small. On the other hand, if an emergent signal of ENSO variability is observed in the future, we may not have a high level of confidence to claim that climate change has impacted ENSO variability, as we can easily mistake the signal due to a modulation of internal variability. Further studies should be focused on quantifying these uncertainties, so that changes in tropical Pacific climate can be confidently detected, should they occur.
Acknowledgments.
S. Z. and J. Y. were supported by the National Natural Science Foundation of China (Grant 42227901), the Scientific Research Fund of the Second Institute of Oceanography, Ministry of Natural Resources (Grant QNYC2001), and the Innovation Group Project of the Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai) (Grant 311021001). M. C. was supported by a grant from the U.K. Natural Environment Research Council (NE/S004645/1). We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP6, and the climate modeling groups (listed in Table 1) for producing and making available their model outputs. We also thank the three anonymous reviewers for their constructive comments that helped improve the paper.
Data availability statement.
The CMIP6 model outputs are derived from: https://esgf-node.llnl.gov/projects/cmip6/. The large ensembles of CESM1 are derived from http://www.cesm.ucar.edu/projects/community-projects/LENS/. The large ensembles of CanESM2 are derived from https://open.canada.ca/data/en/dataset/aa7b6823-fd1e-49ff-a6fb-68076a4a477c. The large ensembles of MPI-ESM are derived from https://esgf-data.dkrz.de/projects/mpi-ge/.
REFERENCES
Cai, W., and Coauthors, 2018: Increased variability of eastern Pacific El Niño under greenhouse warming. Nature, 564, 201–206, https://doi.org/10.1038/s41586-018-0776-9.
Cai, W., B. Ng, T. Geng, L. Wu, A. Santoso, and M. J. McPhaden, 2020: Butterfly effect and a self-modulating El Niño response to global warming. Nature, 585, 68–73, https://doi.org/10.1038/s41586-020-2641-x.
Cai, W., and Coauthors, 2021: Changing El Niño–Southern Oscillation in a warming climate. Nat. Rev. Earth Environ., 2, 628–644, https://doi.org/10.1038/s43017-021-00199-z.
Cai, W., B. Ng, G. Wang, A. Santoso, L. Wu, and K. Yang, 2022: Increased ENSO sea surface temperature variability under four IPCC emission scenarios. Nat. Climate Change, 12, 228–231, https://doi.org/10.1038/s41558-022-01282-z.
Capotondi, A., and P. D. Sardeshmukh, 2017: Is El Niño really changing? Geophys. Res. Lett., 44, 8548–8556, https://doi.org/10.1002/2017GL074515.
Collins, M., and Coauthors, 2010: The impact of global warming on the tropical Pacific Ocean and El Niño. Nat. Geosci., 3, 391–397, https://doi.org/10.1038/ngeo868.
Delage, F. P. D., and S. B. Power, 2020: The impact of global warming and the El Niño–Southern Oscillation on seasonal precipitation extremes in Australia. Climate Dyn., 54, 4367–4377, https://doi.org/10.1007/s00382-020-05235-0.
Deser, C., A. Phillips, V. Bourdette, and H. Teng, 2012: Uncertainty in climate change projections: The role of internal variability. Climate Dyn., 38, 527–546, https://doi.org/10.1007/s00382-010-0977-x.
Dieppois, B., A. Capotondi, B. Pohl, K. P. Chun, P.-A. Monerie, and J. Eden, 2021: ENSO diversity shows robust decadal variations that must be captured for accurate future projections. Commun. Earth Environ., 2, 212, https://doi.org/10.1038/s43247-021-00285-6.
DiNezio, P. N., G. A. Vecchi, and A. C. Clement, 2013: Detectability of changes in the Walker circulation in response to global warming. J. Climate, 26, 4038–4048, https://doi.org/10.1175/JCLI-D-12-00531.1.
Dong, L., L. R. Leung, F. Song, and J. Lu, 2021: Uncertainty in El Niño–like warming and California precipitation changes linked by the Interdecadal Pacific Oscillation. Nat. Commun., 12, 6484, https://doi.org/10.1038/s41467-021-26797-5.
England, M. H., and Coauthors, 2014: Recent intensification of wind-driven circulation in the Pacific and the ongoing warming hiatus. Nat. Climate Change, 4, 222–227, https://doi.org/10.1038/nclimate2106.
Eyring, V., S. Bony, G. A. Meehl, C. A. 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.
Giorgi, F., and X. Bi, 2009: Time of emergence (TOE) of GHG-forced precipitation change hot-spots. Geophys. Res. Lett., 36, L06709, https://doi.org/10.1029/2009GL037593.
Hawkins, E., and R. Sutton, 2009: The potential to narrow uncertainty in regional climate predictions. Bull. Amer. Meteor. Soc., 90, 1095–1108, https://doi.org/10.1175/2009BAMS2607.1.
Hawkins, E., and R. Sutton, 2011: The potential to narrow uncertainty in projections of regional precipitation change. Climate Dyn., 37, 407–418, https://doi.org/10.1007/s00382-010-0810-6.
Hawkins, E., and R. Sutton, 2012: Time of emergence of climate signals. Geophys. Res. Lett., 39, L01702, https://doi.org/10.1029/2011GL050087.
Huang, P., 2014: Regional response of annual-mean tropical rainfall to global warming. Atmos. Sci. Lett., 15, 103–109, https://doi.org/10.1002/asl2.475.
Huang, P., and S.-P. Xie, 2015: Mechanisms of change in ENSO-induced tropical Pacific rainfall variability in a warming climate. Nat. Geosci., 8, 922–926, https://doi.org/10.1038/ngeo2571.
Kosaka, Y., and S.-P. Xie, 2013: Recent global-warming hiatus tied to equatorial Pacific surface cooling. Nature, 501, 403–407, https://doi.org/10.1038/nature12534.
Maher, N., D. Matei, S. Milinski, and J. Marotzke, 2018: ENSO change in climate projections: Forced response or internal variability? Geophys. Res. Lett., 45, 11 390–11 398, https://doi.org/10.1029/2018GL079764.
Maher, N., S. B. Power, and J. Marotzke, 2021: More accurate quantification of model-to-model agreement in externally forced climatic responses over the coming century. Nat. Commun., 12, 788, https://doi.org/10.1038/s41467-020-20635-w.
Maher, N., T. P. Tabarin, and S. Milinski, 2022: Combining machine learning and SMILEs to classify, better understand, and project changes in ENSO events. Earth Syst. Dyn., 13, 1289–1304, https://doi.org/10.5194/esd-13-1289-2022.
Planton, Y. Y., and Coauthors, 2021: Evaluating climate models with the CLIVAR 2020 ENSO metrics package. Bull. Amer. Meteor. Soc., 102, E193–E217, https://doi.org/10.1175/BAMS-D-19-0337.1.
Power, S. B., 2014: Expulsion from history. Nature, 511, 38–39, https://doi.org/10.1038/511038a.
Power, S. B., F. Delage, C. T. Y. Chung, G. Kociuba, and K. Keay, 2013: Robust twenty-first-century projections of El Niño and related precipitation variability. Nature, 502, 541–545, https://doi.org/10.1038/nature12580.
Power, S. B., F. P. D. Delage, C. T. Y. Chung, H. Ye, and B. F. Murphy, 2017: Humans have already increased the risk of major disruptions to Pacific rainfall. Nat. Commun., 8, 14368, https://doi.org/10.1038/ncomms14368.
Power, S. B., and Coauthors, 2021: Decadal climate variability in the tropical Pacific: Characteristics, causes, predictability, and prospects. Science, 374, eaay9165, https://doi.org/10.1126/science.aay9165.
Stevenson, S. L., 2012: Significant changes to ENSO strength and impacts in the twenty-first century: Results from CMIP5. Geophys. Res. Lett., 39, L17703, https://doi.org/10.1029/2012GL052759.
Timmermann, A., 1999: Detecting the nonstationary response of ENSO to greenhouse warming. J. Atmos. Sci., 56, 2313–2325, https://doi.org/10.1175/1520-0469(1999)056<2313:DTNROE>2.0.CO;2.
Wang, G., S. B. Power, and S. McGree, 2015: Unambiguous warming in the western tropical Pacific primarily caused by anthropogenic forcing. Int. J. Climatol., 36, 933–944, https://doi.org/10.1002/joc.4395.
Watanabe, M., J.-L. Dufresne, Y. Kosaka, T. Mauritsen, and H. Tatebe, 2020: Enhanced warming constrained by past trends in equatorial Pacific sea surface temperature gradient. Nat. Climate Change, 11, 33–37, https://doi.org/10.1038/s41558-020-00933-3.
Wittenberg, A. T., 2009: Are historical records sufficient to constrain ENSO simulations? Geophys. Res. Lett., 36, L12702, https://doi.org/10.1029/2009GL038710.
Wittenberg, A. T., A. Rosati, T. L. Delworth, G. A. Vecchi, and F. Zeng, 2014: ENSO modulation: Is it decadally predictable? J. Climate, 27, 2667–2681, https://doi.org/10.1175/JCLI-D-13-00577.1.
Wu, M., T. Zhou, C. Li, H. Li, X. Chen, B. Wu, W. Zhang, and L. Zhang, 2021: A very likely weakening of Pacific Walker Circulation in constrained near-future projections. Nat. Commun., 12, 6502, https://doi.org/10.1038/s41467-021-26693-y.
Xie, S.-P., C. Deser, G. A. Vecchi, J. Ma, H. Teng, and A. T. Wittenberg, 2010: Global warming pattern formation: Sea surface temperature and rainfall. J. Climate, 23, 966–986, https://doi.org/10.1175/2009JCLI3329.1.
Ying, J., P. Huang, and R. Huang, 2016: Evaluating the formation mechanisms of the equatorial Pacific SST warming pattern in CMIP5 models. Adv. Atmos. Sci., 33, 433–441, https://doi.org/10.1007/s00376-015-5184-6.
Ying, J., M. Collins, W. Cai, A. Timmermann, P. Huang, D. Chen, and K. Stein, 2022: Emergence of climate change in the tropical Pacific. Nat. Climate Change, 12, 356–364, https://doi.org/10.1038/s41558-022-01301-z.
Zheng, X.-T., C. Hui, and S.-W. Yeh, 2017: Response of ENSO amplitude to global warming in CESM large ensemble: Uncertainty due to internal variability. Climate Dyn., 50, 4019–4035, https://doi.org/10.1007/s00382-017-3859-7.
