In this section, we will investigate the dominant factors to the causes of different prediction skills of the leading modes of the A-AM from two perspectives: (a) remote impact from tropical SST forcing and (b) local air-sea interaction processes, including the seasonal persistence of MC SSTA and the corresponding structure of atmospheric circulation during two periods.
a. Teleconnections of SST in the tropical equatorial Pacific during two epochs
To explain the spatial differences between the leading modes of A-AM in section 3 and the corresponding differences in prediction techniques in section 4, we will explore their mechanisms from the perspective of tropical SST signals. As an early signal, ENSO contributes to the better performance of multi-model prediction in capturing the spatiotemporal variations of the leading modes of the A-AM. (Wang et al. 2003). Figure 7 shows the seasonal mean SSTA regressed to PC1. It is evident that the warming of Niño3.4 is stronger before the year 2000 and the warming of the eastern equatorial Pacific and Indian Ocean persists until the following summer, enhancing the lagged relationship between A-AM and ENSO. This indicates a synchronization between the seasonal evolution of the S-EOF1 and the temporal transitions of ENSO, potentially leading to more pronounced abnormal Walker circulation. During the decaying phase of ENSO, El Niño events dissipated more rapidly during the period from 2000 to 2017. In MAM(1), the eastern equatorial Pacific SST has begun to transform into a cold anomaly, and the warm anomaly in the central and eastern equatorial Pacific completely disappears by the following summer.
To evaluate whether there are differences in the ability of the NMME models to capture the seasonal evolution of tropical Pacific SST in years with strong and weak memory, leading to differences in predictions for the A-AM system. Figure 7 and Fig. 8 show the regression of the first two PCs and NMME pattern projection coefficients with increasing leading time onto tropical Pacific SSTAs for the two epochs. Here we focus on the results of the MME. It is observed that S-EOF1 of A-AM corresponds to the decaying phase of ENSO. During 1982–1999, warming in the eastern equatorial Pacific persists until MAM(1) or even JJA(1). The MME captures the decaying phase of ENSO well within a lead time of 1–3 months with a PCC of around 0.9. During 2000–2017, ENSO events decay rapidly, and cooling in the eastern Pacific begins from MAM(1), becoming a cold phase by JJA(1), weakening the lagged relationship between A-AM and ENSO. Moreover, compared to 1982–1999, the predictive skills are weaker, particularly at a lead time of 5 months when the skill becomes negative and essentially fails. It is worth noting that during the El Niño developing phase, MME exhibits a higher pattern correlation with the seasonal evolution of SSTAs. However, during the El Niño decaying phase (JJA(1)), the similarity to observations rapidly decreases, especially for weak memory epoch, indicating a significant impact of SST forcing on the accuracy of model predictions.
The pattern of S-EOF2 mainly corresponds to the developing phase of ENSO (Fig. 8). During periods with a strong memory, warming amplitudes in the equatorial central-eastern Pacific are more intense, corresponding to a stronger westerly anomalies in S-EOF2 observed in JJA(1) and a precursor signal for ENSO. The relationship between ENSO and S-EOF2 has changed Remarkably. As depicted in Table 2, during 1982–1999, the lagged correlation coefficient between the second principal component (PC2) and the Niño3.4 index was found to be 0.827, indicating a significant relationship at a confidence level of at least 99% according to a Pearson correlation test. In contrast, during 2000–2017, this lagged correlation coefficient decreased to only 0.33, which did not reach statistical significance at a confidence level of at least 95%. During the period from 1982 to 1999, the model simulation accurately captured the transition of the Eastern Pacific from a near-normal SST mode to warming during the ENSO developing phase (PCC around 0.92–0.99). From 2000 to 2017, the warming amplitude during the developing stages of ENSO was lower compared to that observed in the preceding period, and the MME overestimated the warming amplitude of the eastern equatorial Pacific (PCC around 0.58–0.85). When the MME struggles to accurately capture the evolution characteristics of tropical SST during the mature and decay phases of ENSO, it will correspondingly lead to a decline in the prediction skills of the A-AM.
Table 2
Lead-lag correlations between Niño 3.4 and the first and second S-EOF mode principal components. (0) represents the reference year, and (1) represents the following year. The correlation coefficients that are statistically significant at a 95% confidence level are in boldface.
| PC1(0) | PC2(0) |
| 1982–1999 | 2000–2017 | 1982–1999 | 2000–2017 |
\(\:\text{Ni}\stackrel{\text{\sim}}{\text{n}}\text{o}\)-3.4 SSTA(0) | 0.924 | 0.836 | 0.023 | 0.085 |
\(\:\text{Ni}\stackrel{\text{\sim}}{\text{n}}\text{o}\)-3.4 SSTA(1) | -0.026 | -0.103 | 0.827 | 0.331 |
To further compare the lead-lag correlation between the two leading modes of A-AM and ENSO in periods with strong and weak memory, Fig. 9 displays the lead-lag correlation coefficients between four PCs and seasonal (three-month mean) Nino3.4 index for the pre-2000 and post-2000 epochs. In the period 1982–1999, the highest positive correlation coefficient is in D(0)JF(1). The differences between the two epochs in the first mode are minor, both showing a strong positive correlation with Niño 3.4 SSTA. However, the lead-lag correlation coefficients significantly differ for the S-EOF2. During 1982–1999, there is a steady increase in the lead-lag correlations from JJA(-1) to MAM(2), with a maximum correlation of 0.8. This is consistent with the results of Wang et al. (2008). The lowest negative correlation (-0.31) lags El Niño by one year, marking the transition from La Niña to El Niño events. The overall lead-lag correlation coefficients are much higher during the strong memory period (1982–1999) compared to the weak memory period (2000–2017). In the next section, we aim to explain why the association between ENSO and A-AM is stronger during the strong memory epoch from the perspective of MC ocean memory.
b. Influence of ocean memory changes in the MC region
In Section 4, we have identified that the differences in predictability sources of the leading modes of the A-AM primarily arise from variations in the strength of the anomalous anticyclone in WNP during El Niño mature winter to the following spring. The WNPAC, as a crucial subsystem of the A-AM, serves as a critical link between El Niño in the Central-Eastern Pacific and the A-AM (Wang et al. 2000, 2003). Therefore, we hypothesize that the interdecadal variability in the ocean memory of the MC alters the persistence of seasonal evolution of SSTA, destabilizing local air-sea interactions and impacting the forcing of WNPAC anomalies.
The depth of the upper-ocean MLD plays a crucial role in the persistence of SST anomalies on seasonal to interannual time scales (Frankignoul and Hasselmann 1977; de Coёtlogon and Frankignoul 2003). In previous studies, it was found that the upper MLD is thickest in winter and thinnest in summer. Consequently, the strongest ocean memory is in winter and spring (Namias et al. 1988). Thermal anomalies stored in the winter mixed layer weaken during summer and become partially re-entrained into the mixed layer during the following winter. This suggests that the mean seasonal cycle of MLD can induce winter-to-winter memory or persistence of SST anomalies through a "re-emergence" mechanism (Namias and Born 1974; Alexander and Deser 1995; Alexander et al. 2000; Deser et al. 2003). Therefore, to investigate interdecadal variations in the influence of the annual cycle of MC SST persistence on the intensity of the anomalous WNPAC, and to identify the intrinsic memory of the local ocean by isolating the specific effects of local SST anomalies on the subsequent seasonal evolution of SST without the confounding influence of ENSO, we conducted a partial correlation analysis between the regional averaged MC SST index (MC SSTI) in D(0)JF(1) and local SSTA during D(0)JF(1) to JJA(1) after removing the effect of the concurrent Nino3.4 index, as shown in Fig. 10. During the mature phase of ENSO, the MC SST exhibits a consistent cold SSTA mode. The cold SSTA in the period with strong memory is slightly stronger than that in the period with weak memory. Notably, the consistent cold SSTA in the MC region persists during the period with strong memory in MAM(1), whereas it cannot be sustained during the period with weak memory. In the subsequent seasons, the consistent cold SSTA in the MC region remains well-maintained in the period with strong memory, but the intensity of the cold SSTA is weaker compared to the previous period with weak memory. It is worth mentioning that the interdecadal variation of SSTAs in the MC region in MAM(1) is the most prominent among the four seasons. Therefore, in the subsequent discussion, we will focus on the analysis of spring when ENSO signals rapidly decline as a representative season, as the characteristics of other seasons are similar and will not be further elaborated.
Our analysis suggests that the maintenance of the anomalous WNPAC is influenced by different modes of SST anomalies in the MC region during two periods. To examine this hypothesized mechanism, we conducted partial regression of the 500-hPa geopotential height and wind anomalies in MAM(1) with the MC SSTI during two distinct epochs (Fig. 11). Our analysis reveals that in the strong memory epoch, an anomalous anticyclone centered on the Philippine Sea is accompanied by the cold SSTA in the MC region, indicating that the Matsuno/Gill-type response, resulting from the abnormal atmospheric thermal forcing caused by the cold SSTA, strengthens the anticyclonic circulation. However, during the period of weak memory, the cold SSTA in the MC region during D(0)JF(1) cannot be sustained until the following MAM(1), leading to an insignificant anticyclone anomaly. This suggests that the passive response of the atmosphere to SST is weakened. To further elucidate the impact of the MC SSTA on the enhanced anomalous anticyclonic circulation, we performed partial regression analysis using the MC SSTI, to examine the velocity potential and divergent wind at both 850-hPa and 250-hPa levels (Fig. 12). Our finding indicates that the cold SSTA in the MC region is accompanied by convergence in the upper troposphere and divergence in the lower troposphere near the Philippine Sea. During the period of weak memory, when the cold SSTA weakens, the intensity of upper tropospheric convergence and lower tropospheric divergence also weakens, thereby diminishing its role in maintaining the WNPAC in MAM(1). Likewise, the spatial pattern of the partial regression field for precipitation anomalies during MAM(1) aligns with the partial correlation field of SSTA.
To further validate the impact of the interdecadal variation of the MC SSTA pattern on the anomalous WNPAC, the LBM experiment is conducted to analyze its underlying mechanisms. Heating rates are estimated here by the precipitation anomalies with forcing prescribed over the area 122.5° -147.5°E, 2.5°S-17.5°N. The independent effect of MC SSTA on the intensity of WNPAC during years with strong and weak memory is compared. Figure 13 displays the steady response of the 500-hPa geopotential height field and winds. It is evident that the 31–45 days average results in the LBM experiment closely resemble the observed results. During periods of strong memory, a stronger anticyclonic circulation anomaly is observed near the Philippine Sea in the 500-hPa wind field, with significant positive geopotential height anomalies. This feature is greatly diminished and less pronounced during the later period. Additionally, it is found that the cold MC SSTA caused upper tropospheric convergence and lower tropospheric divergence in the Philippine Sea to its northwest through a Gill-type response (Fig. 14a-c) (Gill 1980). Compared to the atmospheric circulation pattern during periods of weak memory (Fig. 14d-e), this effect is more pronounced during periods with significant ocean memory.
Therefore, since the 1980s, under the influence of memory changes, there have been interdecadal variations in the seasonal evolution of the MC SSTA and its driving effects on anomalous WNPAC. The physical mechanism underlying these variations is illustrated in Fig. 15. During periods of strong memory, the cold SSTA in the MC region exhibits strong persistence from the decaying phase of El Niño, allowing abnormal atmospheric signals to be maintained for a longer duration. The convection restrained by the local cooling SSTA is robust, facilitating a stronger coupling between the anomalous WNPAC and the cold MC SSTA through a positive wind–evaporation–SST (WES) feedback (Wang et al. 2000) mechanism. Through the Gill-type response, the MC SSTA exerts a more prominent role in promoting the maintenance of the anomalous WNPAC, thereby reinforcing the A-AM system. On the other hand, the sustained development of anomalous WNPAC also helps to strengthen the connection between ENSO and A-AM systems, facilitating the transmission of SST signals in the central and eastern equatorial Pacific, thereby enhancing the predictive skills of the A-AM system. However, during periods of weak memory, the cold SSTA in the MC region during the mature phase of ENSO cannot be well sustained in the subsequent seasons due to decreased SST persistence, leading to a less significant independent effect on the WNP anomalous anticyclone. Consequently, the weakened intensity of the A-AM system and the diminished coupling relationship with ENSO result in lower prediction skills compared to periods of strong memory.