-
In the previous section, the CMIP6 models showed almost no distinct difference from the CMIP5 models in terms of simulating the teleconnection between ENSO and EASR, although current models have greatly improved over their CMIP5 versions. As mentioned in the introduction, the preceding winter ENSO affects EASR through three physical processes: the effect of wintertime ENSO on TIO SST; the effect of TIO SST on PSC; and the effect of PSC on EASR. Thus, to identify the possible reasons limiting the CMIP6 models from simulating the ENSO–EASR relationship well, we analyzed the simulation of these three processes in the three generations of models.
4.1.
Simulation of the relationship between ENSO and TIO SST
-
Figure 5 shows the subsequent-summer TIO SST regressed onto the standardized DJF Niño3 index in the observations, CMIP6 MME, and individual CMIP6 models. Generally, there are three SST anomaly centers in the observations, located over the northwestern Indian Ocean, southwestern Indian Ocean, and southeastern Indian Ocean. Except for BCC-ESM1 and MCM-UA-1-0, all CMIP6 models reproduce the SST anomaly related to ENSO over the Indian Ocean, especially over the northern TIO where the SST anomaly plays a more important role in affecting the western North Pacific circulation (Xie et al., 2009; Huang et al., 2010). Most CMIP6 models fail to capture the SST anomaly over the southeastern Indian Ocean. In contrast, the ENSO-related TIO SST anomaly can be simulated in only half of the CMIP3 models (Fu et al., 2013), but it can be represented in almost all CMIP5 models, even in the “worst” CMIP5 models that have the weakest ENSO–EASR relationships (Fu and Lu, 2017). On the other hand, there is a positive SST anomaly over the eastern tropical Pacific in the observations, which is consistent with previous studies (e.g., Xie et al., 2009). Almost all the CMIP6 models represent this SST anomaly with stronger intensity than observed. In the meantime, most models simulate negative SST anomalies over the southern tropical Pacific, which do not exist in the observations. It seems that the simulated SST anomaly over the tropical Pacific cannot directly affect the ENSO–EASR relationship.
The MMEs of the three generations of models simulate almost the same spatial pattern and intensity of ENSO-related TIO SST anomalies (Figs. 6a–c). The MMEs successfully represent the ENSO-induced SST anomalies over the northwestern and southwestern Indian Ocean, but cannot reproduce the SST anomaly over the southeastern Indian Ocean. The biases of the simulated SST anomalies between the CMIP models and the observations are positive over the tropical western Indian Ocean and central southern TIO (Figs. 6d–f), indicating an overestimation compared with the observations. Additionally, the biases are negative over the southeastern Indian Ocean in the MMEs because of the unsuccessful representation of the observed positive SST anomaly in the models.
The intermodel diversity of the ENSO-related SST anomaly decreases from the CMIP3 to CMIP6 models (Figs. 6g–i). In the CMIP3 models, the intermodel StDs reach up to approximately 0.16–0.18°C over the northwestern, southwestern, and southeastern Indian Ocean. In the CMIP5 models, the largest intermodel diversity is still located over these regions, but the StDs decrease to approximately 0.10–0.12°C. In the CMIP6 models, the intermodel diversity centers over the southern TIO almost disapper, with the StDs decreasing to only approximately 0.06–0.08°C. The center over the northwestern TIO also shrinks and decreases to approximately 0.10°C. Additionally, the intermodel StD over the central Indian Ocean decreases from approximately 0.10°C in the CMIP3 models to 0.04°C in the CMIP6 models.
The improvement in the CMIP6 models in simulating ENSO’s impact on TIO SST is clearly shown in Fig. 7a, which is a reproduction of Fig.5a in Fu and Lu (2017) but with the results of the CMIP6 models added. The correlation coefficients between ENSO and TIOI tend to be stronger and are nearer to the observations (0.66) in the CMIP6 models compared with those in the CMIP3 and CMIP5 models. The correlation coefficients are within 0.70–0.80 in the largest percentage of CMIP6 models (50%), and range from 0.80 to 0.90 in the following proportion (20%). That is, 70% of the CMIP6 models simulate the ENSO–TIOI correlation coefficient within 0.70–0.90, which are stronger than the observed values. The percentage is much greater compared with that in the CMIP3 models (39%) and CMIP5 models (59%). Additionally, the correlation coefficients are stronger than 0.60 in approximately 50% of the CMIP3 models, 68% of the CMIP5 models, and 85% of the CMIP6 models.
The CMIP6 models exhibit smaller dispersion in the ENSO–TIOI correlation coefficients than the CMIP3 and CMIP5 models (Fig. 7b). There are only two outliers (BCC-ESM1 and MCM-UA-1-0) that simulate much lower correlation coefficients (approximately < 0.30). The correlation coefficient ranges from approximately 0.22 to 0.90 in the CMIP3 models, 0.31 to 0.90 in the CMIP5 models (except for one outlier), and from 0.55 to 0.86 in the CMIP6 models (except for two outliers). Between the 25th and 75th quartiles, the correlation coefficients are within the scope of approximately 0.35–0.84 in the CMIP3 models and 0.58–0.79 in the CMIP5 models, and the scope narrows to 0.67–0.79 in the CMIP6 models. Additionally, the MMEs of the correlation coefficients are approximately 0.59, 0.68, and 0.69 in the CMIP3, CMIP5, and CMIP6 models, respectively; and the median values are approximately 0.58, 0.74, and 0.74.
Based on the above results, we can conclude that the CMIP6 models simulate the ENSO–TIOI relationship more reasonably than the CMIP3 and CMIP5 models.
4.2.
Simulation of the relationship between TIO SST and PSC
-
Figure 8 shows the JJA precipitation regressed onto the standardized TIOI, which represents the second step of ENSO’s impact on EASR. In the observations, a positive TIO SST anomaly corresponds to a below-normal rainfall anomaly over the Philippine Sea and northwestern subtropical Pacific. This negative TIOI-related precipitation anomaly is successfully represented in the CMIP6 MME and 12 out of 20 CMIP6 models (CAMS-CSM1-0, CESM2, CESM2-WACCM, CNRM-CM6-1, CNRM-ESM2-1, FGOALS-g3, HadGEM3-GC31-LL, MCM-UA-1-0, MIROC-ES2L, MRI-ESM2-0, NorCPM1, and UKESM1-0-LL). The ratio is nearly identical to that in the CMIP5 models (12 out of 22) (Fu and Lu, 2017) and greater than that in the CMIP3 CGCMs (5 out of 18) (Fu et al., 2013). The negative precipitation anomaly in MIROC6 is relatively weak and shifts far eastward in comparison with the observations (Fig. 8o), resulting in an insignificant TIOI–PSCI correlation coefficient of approximately 0.18. Otherwise, the main body of the TIOI-related negative precipitation anomaly shifts eastward by approximately 20° in longitude compared with the observations, with the western edge located east of 130°E in six CMIP6 models (CAMS-CSM1-0, CESM2-WACCM, FGOALS-g3, HadGEM3-GC31-LL, MCM-UA-1-0, and UKESM1-0-LL).
More importantly, the well-simulated TIOI–PSCI relationship guarantees that the CMIP6 models will capture the ENSO–EASR correlation, which is quite different from the CMIP5 models. Except for MCM-UA-1-0 and NorCPM1, all the remaining 10 CMIP6 models that capture a significant TIOI–PSCI relationship of between −0.24 and −0.74 are models that realistically simulate the ENSO–EASR relationship. The two exceptions have TIOI–PSCI correlation coefficients of approximately −0.24 and −0.55, but the ENSO–EASR correlation coefficients are only 0.12 and 0.10, respectively. All remaining eight CMIP6 models that cannot reproduce the significant TIOI–PSCI relationship fail to capture the ENSO–EASR relationship. On the other hand, all 10 CMIP6 models that simulate a significant ENSO–EASR relationship are also models that realistically represent a significant TIOI–PSCI relationship. However, this phenomenon does not exist in the CMIP5 models, and no obvious connection can be found between the TIOI–PSCI and ENSO–EASR correlations (Fu and Lu, 2017).
The TIOI-related precipitation anomaly over the Philippine Sea and northwestern subtropical Pacific in the CMIP6 MME is relatively stronger than those in the CMIP3 and CMIP5 MMEs (Figs. 9a–c). It also shows that the simulated PSC shifts eastward in all three MMEs, with the western edge located east of 130°E. Accordingly, the biases, with the maximum located over 120°–140°E, decrease from the CMIP3 to CMIP5 models (Figs. 9d–f). Different from the MMEs and biases, the intermodel spread in the CMIP6 models, however, increases. Over the PSC region, the intermodel StDs are approximately 0.2–0.3 mm d−1 in the CMIP3 models (Fig. 9g), 0.3–0.4 mm d−1 in the CMIP5 models (Fig. 9h), and larger than 0.4 mm d−1 in the CMIP6 models (Fig. 9i).
Figure 7c displays a histogram of the TIOI–PSCI correlation coefficients in the CMIP3, CMIP5, and CMIP6 models, which is a reproduction of Fig.5b in Fu and Lu (2017) but with the results of the CMIP6 models added. Generally, the intensity of the TIOI–PSCI correlation coefficients in the CMIP5 and CMIP6 models exhibit almost no obvious difference, and they are both stronger than that in the CMIP3 models. In the largest percentage of the CMIP6 models (20%), the correlation coefficients are within the scope of −0.20 to −0.30. In the CMIP5 models, the correlation coefficients of the largest proportion (27%) range from −0.50 to −0.40, which is stronger than that in the CMIP6 models, while the correlation coefficients of the largest proportion (28%) are weaker, at only −0.20 to −0.10, in the CMIP3 models. Additionally, 60% of the CMIP6 models reasonably represent the TIOI–PSCI correlation coefficient (< −0.20). The ratio is comparable to that in the CMIP5 models (55%) and larger than that in the CMIP3 models (28%).
Figure 7d quantitatively shows that the intermodel diversity increases from the CMIP3 to CMIP6 models. The scope of the TIOI–PSCI correlation coefficients is from approximately −0.59 to 0.26 in the CMIP3 models, −0.58 to 0.32 in the CMIP5 models, and increases to −0.74 to 0.42 in the CMIP6 models. Between the 25th and 75th quartiles, the correlation coefficients change from approximately −0.25 to 0.10 in the CMIP3 models, −0.45 to 0.03 in the CMIP5 models, and −0.44 to 0.04 in the CMIP6 models. The MME/median values are −0.12/−0.17, −0.21/−0.22, and −0.21/−0.24 in the CMIP3, CMIP5, and CMIP6 models, respectively. Additionally, the observed TIOI–PSCI relationship (−0.49) is underestimated in almost all three generations of models.
In summary, the most important improvement is that the well-simulated TIOI–PSCI relationship guarantees that the CMIP6 models will realistically capture the ENSO–EASR correlation, but this is not the case in the CMIP5 models. However, the CMIP6 models show no obvious changes in terms of simulating this relationship. The TIOI–PSCI correlation coefficients in the CMIP6 models are almost the same as those in the CMIP5 models and stronger than those in the CMIP3 models.
4.3.
Simulation of the relationship between PSC and EASR
-
Figure 10 shows the summer precipitation regressed onto the standardized PSCI in the observations, CMIP6 MME, and individual CMIP6 models. In the observations, a positive PSCI induces a negative EASR anomaly, which indicates a representation of the Pacific–Japan pattern (Lu, 2004; Kosaka and Nakamura, 2006). The below-normal precipitation anomaly is simulated in 18 out of 20 CMIP6 models (all except BCC-ESM1 and NESM3), although it is much weaker in most models than that in the observations. In contrast, 14 out of 18 CMIP3 models (Fu et al., 2013) and 17 out of 22 CMIP5 models (Fu and Lu, 2017) can represent the PSCI–EASRI relationship. Therefore, most CMIP models can reproduce the inherent relationships of the East Asian summer monsoon well.
The intensity and spatial characteristics of the PSCI-related EASR anomaly are similar to each other in the MMEs, and all are much weaker than those in the observations (Figs. 11a–c). Accordingly, the biases of the precipitation anomaly in the MMEs exhibit nearly the same pattern and intensity (Figs. 11d–f). The positive biases are mainly located over central China and the Pacific that east of Japan. The intermodel diversity exhibits almost no difference from each other in the three generations of models, with intermodel StDs of approximately 0.2 mm d−1 over the EASR region (Figs. 11g–i). Therefore, the three generations of models have similar skills in representing the PSCI–EASRI relationship.
Figure 7e shows that the PSCI–EASRI correlation coefficient tends to become weaker in the CMIP6 models, especially compared with that in the CMIP5 models. Approximately 90% of the CMIP6 models simulate a significant PSCI–EASRI relationship (< −0.20) that is statistically significant at the 5% level. The ratio is slightly larger than that in the CMIP3 (78%) and CMIP5 (77%) models. However, strong correlation coefficients (< −0.40) are simulated in only 25% of the CMIP6 models, which is lower than that in the CMIP3 (39%) and CMIP5 (41%) models. The correlation coefficients change from −0.30 to −0.20 with the peak proportion (35%) in the CMIP6 models, which is weaker than that of −0.50 to −0.40 (23%) in the CMIP5 models and identical to that of −0.30 to −0.20 (33%) in the CMIP3 models. Additionally, the PSCI–EASRI relationship is weaker than observed (−0.62) in almost all the analyzed CMIP3, CMIP5, and CMIP6 models.
Except for the outliers with correlation coefficients markedly stronger (MIROC-ES2L and UKESM1-0-LL) or weaker (BCC-ESM1 and NESM3) than those of the other models, the simulated PSCI–EASRI relationship tends to be more concentrated in the CMIP6 models than in the CMIP3 and CMIP5 models (Fig. 7f). The correlation coefficients spread from −0.56 to −0.21 in the CMIP6 models (except four outliers), while they range from −0.56 to −0.06 in the CMIP3 models and from −0.61 to 0.07 in the CMIP5 models. Between the 25th and 75th quartiles, the correlation coefficients change from approximately −0.50 to −0.20 in the CMIP3 models, from −0.46 to −0.20 in the CMIP5 models, and from −0.42 to −0.27 in the CMIP6 models. Additionally, almost all three generations of models underestimate the PSCI–EASRI relationship.
In summary, the three generations of models exhibit essentially identical capabilities in representing the PSCI–EASRI relationship, with almost the same spatial pattern, intensity, bias, and intermodel diversity of the PSC-related precipitation anomaly. However, the correlation coefficient is weaker in the CMIP6 models, although it is more concentrated after excluding the outliers.
The above study evaluated the three physical processes related to the delayed impact of winter ENSO on the subsequent EASR in the CMIP6 models and compared the results with those in the CMIP3 and CMIP5 models. According to the analysis in section 4.2, except MCM-UA-1-0 and NorCPM1, all remaining 10 CMIP6 models that capture a significant TIOI–PSCI relationship are identical to the models that reproduce a significant ENSO–EASR relationship, and the eight CMIP6 models that cannot reproduce a significant TIOI–PSCI relationship fail to capture the ENSO–EASR relationship. This suggests that the TIOI–PSCI relationship is the key teleconnection determining whether the CMIP6 models can simulate the ENSO–EASR relationship. Unfortunately, the CMIP6 models fail to offer any improvement in simulating the TIOI–PSCI relationship, although they simulate a more realistic ENSO–TIOI relationship. The failure likely explains the fact that there is no obvious progress in simulating the ENSO–EASR relationship, as identified in section 3. Additionally, the ENSO–EASR relationship shows a slightly larger intermodel uncertainty in the CMIP6 models than in the CMIP5 models (Fig. 4b), which is attributable to the increased intermodel spread in the TIOI–PSCI relationship (Fig. 7d) since the intermodel spread for the remaining two physical processes is reduced (Figs. 7b and f). This result further supports the conclusion that the TIOI–PSCI relationship is the key process in determining the reproduction of the ENSO–EASR relationship in the CMIP6 models.