The Asymmetric Predictive Power of Indian Ocean Dipole for Subsequent Year's ENSO: Role of Atlantic Ocean as an Intermediary

We examined the Indian Ocean Dipole (IOD)'s predictive ability for El Niño‐Southern Oscillation (ENSO). Our findings indicate that positive IODs have a stronger impact on the subsequent year's ENSO compared to negative IODs. To explain this asymmetry, we proposed the “IOD‐Atlantic‐Pacific” pathway, which involves the Atlantic in boreal winter as an intermediary. The pathway comprises two stages: (a) The asymmetric “IOD‐Atlantic” connection, where positive IODs in autumn trigger winter Atlantic Niño while negative IODs cannot trigger Atlantic Niña. This asymmetry is due to the modulation of the climatological mean state on the IOD‐induced atmospheric anomalies. (b) The “Atlantic‐ENSO” connection, which is symmetric, where winter Atlantic Niño (Niña) promotes the development of La Niña (El Niño), functioning independently of ENSO's self‐oscillation behavior. The Atlantic intermediary mechanism explains the asymmetric connection between IOD and the subsequent year's ENSO and provides the IOD with independent predictive power for ENSO beyond its self‐oscillation.


Plain Language Summary
The Indian Ocean Dipole (IOD) plays a role in predicting the El Niño-Southern Oscillation (ENSO) for the next year, but there is no agreement on how exactly this works and if it is separate from ENSO-cycle itself.This study found that only positive IODs can sufficiently predict the following year's ENSO.This is because, positive IOD can cause a winter Atlantic Niño, which then leads to the development of La Niña in the spring and summer.However, negative IOD cannot cause an Atlantic Niña.Thus, the asymmetry in the influence of IOD on winter Atlantic lead to the asymmetry in the relationship between IOD and the following year's ENSO.The asymmetry in the IOD's influence on the Atlantic is due to how the climatic conditions affect the distribution of IOD-induced atmospheric anomalies.The "IOD-Atlantic-Pacific" pathway gives the IOD its own predictive power for the next year's ENSO.These results do not only advance the understanding of the physical mechanisms of the IOD on the following's ENSO but also provides a foundation for improving ENSO prediction.
• Positive Indian Ocean Dipole (IOD) can effectively predict the subsequent year's El Niño-Southern Oscillation (ENSO), while negative IOD has little predictive power • An Atlantic intermediary mechanism is proposed to link IOD to next year's ENSO, and explain their relationship's asymmetry • IOD's impact on winter Atlantic is asymmetric due to the modulation of climatological convective activity on atmospheric responses to IOD

Supporting Information:
Supporting Information may be found in the online version of this article. 10.1029/2023GL105525 2 of 9 The above controversies motivated this study.Does the IOD have predictive significance for the subsequent year's ENSO beyond the self-oscillation of ENSO?To address this, we take into account the asymmetry in the temporal evolution of ENSO (Larkin & Harrison, 2002;Okumura & Deser, 2010), and examine the impact of IOD on future ENSO prediction separately under different El Niño or La Niña backgrounds.Our study identifies an asymmetry between IOD and the subsequent year's ENSO, and proposed an "Atlantic intermediary" mechanism connecting the two phenomena.This mechanism explains the observed asymmetry and demonstrates that IOD has independent predictive capability for the following year's ENSO.These results are presented in Section 3, following an introduction of data and method in Section 2, and a summary is provided in Section 4.

Data and Method
We employed the Extended Reconstructed SST, version 5 (ERSSTv5) (Huang et al., 2017) for SST data, and NCEP/NCAR reanalysis data (Kalnay et al., 1996) for atmosphere including 10-m wind, 200 hPa velocity potential, and 500hpa vertical velocity (omega).We also used the NOAA's Precipitation Reconstruction (PREC) data set (Chen et al., 2002) for precipitation.The SST anomalies (SSTAs) were derived by removing the mean seasonal cycle and trends.The time range for all the data above was set from 1950 to 2022.In addition to the aforementioned observational data, we analyzed a 500-year simulation of the pi-control experiment from CMIP6 (Eyring et al., 2016), employing the CMCC-CM2-SR5 model.Statistical techniques include composite, regression, correlation, partial correlation, and K-means cluster analysis.
ENSO is represented by the Niño3.4index (5°S-5°N, 170°W-120°W).Additionally, two other indices (Cai et al., 2018) are used to represent the eastern Pacific (EP) type and central Pacific (CP) type of ENSO.The IOD is assessed by the differences in SSTAs between the western (10°S-10°N, 50°E-70°E) and eastern (10°S-0°, 90°E-110°E) tropical Indian Ocean (Saji et al., 1999).For the representation of the winter Atlantic Niño, we calculate the average SSTA over the region 15°S-5°N, 30°W-10°E, referred to as the AtlNiño index, which covers the conventional ATL3 region (3°S-3°N, 20°W-0°) but extends further south.The reason for using this index is discussed in Section 3.4.4,along with a comparison between the two Atlantic Niño indices.
We indicate the year number in the lower right corner of months to account for the different years encompassed by the seasons or months.For example, SON 0 represents the autumn from September to November in year 0, while ND 0 J represents the peak season of ENSO lagging one season after SON 0 .Similarly, ND 1 J represents the peak season of ENSO for the subsequent year, occurring 14 months after SON 0 .

Asymmetric Predictive Power of IOD for the Subsequent Year's ENSO
We first examined the correlation map (Figure 1a) of the Indian Ocean SST (SON 0 ) with the subsequent year's Niño3.4index (ND 1 J).It shows a weak negative (positive) IOD pattern that precedes El Niño (La Niña) by 14 months, as reported by Izumo et al. (2010).This correlation map is calculated using all the samples and thus represents a symmetric relationship.We then divided the samples into two groups based on the contemporary ENSO background with IOD (Table S1 in Supporting Information S1), and calculated correlation maps for each group.Interestingly, the El Niño group (Figure 1b) exhibits a more significant IOD pattern than the all-sample case (Figure 1a), whereas the La Niña group (Figure 1c) does not display any IOD pattern at all.The correlation between IOD and the subsequent year's ENSO is −0.48 for the El Niño group, while it is only 0.15 for the La Niña group (Figure 1d).This contrast suggests that the previously reported link between IOD and the subsequent year's ENSO primarily depends on IODs cooccurring with El Niño, rather than those occurring with La Niña.
To investigate whether IOD has independent predictive potential for the subsequent year's ENSO, we calculated the autocorrelation of ENSO separately for El Niño and La Niña in Figure 1e.It indicates a correlation of −0.36 between ENSO and the subsequent year's ENSO, which is notably lower than that between IOD and the following year's ENSO (−0.48, Figure 1d).This contrast in correlation suggests that the IOD likely possesses predictive capabilities for the subsequent year's ENSO, independent of the ENSO self-oscillation.
In Figures 1d and 1e, the grouping is based on the ENSO status cooccurring with IOD.In fact, the asymmetry in Figure 1d lies in the contrasting impacts of positive and negative IODs, because positive (negative) IODs 10.1029/2023GL105525 3 of 9 primarily cooccur with contemporary El Niño (La Niña).Consequently, when the grouping is performed based on the IOD index (Table S2 in Supporting Information S1), this asymmetry can also be observed in Figure 1f (−0.48 vs −0.12), suggesting that a positive IOD is in a much better position than a negative IOD to predict the following year's ENSO.

The Asymmetric Pathway of IOD's Impact on Subsequent Year's ENSO
To understand the asymmetry in Figure 1f, where a strong positive IOD leads to a subsequent La Niña, while a strong negative IOD is not strongly associated with the occurrence of El Niño in the next year, we analyzed strong positive IODs and strong negative IODs separately (defined as absolute values of standardized IOD index exceeding 1.0, see Table S3 in Supporting Information S1). Figure 2 shows that strong positive IODs (Figure 2a-2f) are associated with abnormal westerly winds over the tropical Atlantic during autumn, which weaken the background easterly winds and enhance heat flux, leading to Atlantic warming during winter, known as the Atlantic Niño.In contrast, strong negative IODs (Figures 2g-2l) do not exhibit notable westerly wind anomalies over the Atlantic in autumn, leading to no subsequent Atlantic cooling during winter.A case study (Figure S1 in Supporting Information S1) for 2019 strong positive IOD and 2016 strong negative IOD supports this conclusion.The asymmetry in the winter Atlantic Niño responses to positive and negative IODs is significant (Figures 2p-2r).While the Atlantic Niño is commonly observed in summer, previous studies (Okumura & Xie, 2006) have suggested that in winter, the strengthening of the easterly trade winds in the Atlantic lead to a shallower thermocline and increased upwelling, making SST more sensitive to surface winds and favoring the occurrence of the Atlantic Niño.The co-occurrence of IOD and ENSO in the composite results of Figure 2 raises the question of whether the asymmetric Atlantic anomalies are caused by IOD or ENSO.To address this, we calculated the correlation between IOD (SON 0 ) and Atlantic Niño (D 0 JF) for positive and negative IODs in Figure 2s.The results show a strong association between positive IOD and Atlantic Niño, but no significant link between negative IOD and Atlantic Niña (0.68 vs. −0.16).This "IOD-Atlantic" connection remains robust even after removing the effects of ENSO (Figure 2t), indicating that the asymmetry in the winter Atlantic anomaly is mainly due to IOD rather than ENSO.
As for the impact of IOD on winter Atlantic Niño, a recent study by Zhang and Han (2021) has also reported this connection.Our research further reveals an asymmetry in this impact, which may explain the asymmetry in the influence of IOD on ENSO in the subsequent year.The reasons for this asymmetry will be discussed in Figure 4.However, before delving into this analysis, it is important to continue our discussion on how the IOD induces the development of ENSO in the subsequent year after stimulating the winter Atlantic Niño.
Figure 3 illustrates the impact of winter Atlantic Niño (D 0 JF) on the subsequent year's ENSO development.We use regression analysis to examine the evolution of SST and surface wind anomalies from winter to summer associated with the winter Atlantic Niño, while excluding the influence of winter ENSO (ND 0 J).The results indicate that winter Atlantic Niño can cause abnormal Walker circulation, characterized by anomalous ascendance in the tropical Atlantic and subsidence in the equatorial east Pacific (Figures 3d and 3e).These alterations drive favorable wind anomalies that facilitate ENSO development in spring and summer.
Previous studies mainly recognized the influence of "summer" Atlantic Niño on subsequent ENSO (Dommenget et al., 2006;Wang, 2006).However, a recent study (Hounsou-Gbo et al., 2020) suggests that a "winter" Atlantic Niño also has predictive significance for ENSO, and its mechanism of influence is similar to that of the summer Atlantic Niño (Keenlyside et al., 2013;Polo et al., 2015).Our research (Figure S2 in Supporting Information S1) revealed that this connection between winter Atlantic Niño and ENSO is symmetric regardless the polarity of the Atlantic and is independent of the self-oscillation of ENSO.
Now the pathway linking the IOD and the subsequent year's ENSO has emerged: the autumn IOD triggers the winter Atlantic Niño, which in turn triggers the development of ENSO during spring and summer.Therefore, the Atlantic Ocean acts as an intermediary in mediating the IOD's influence on subsequent year's ENSO.
To further substantiate the Atlantic intermediary mechanism and demonstrate the independent predictive power of IOD for ENSO beyond the self-oscillation of ENSO, we calculated the partial correlation between IOD (SON 0 ) and Niño3.4 (ND 1 J) after removing the effects of the Atlantic intermediary (green curve in Figure 3g).The partial correlation coefficient significantly weakens compared to the unadjusted simple correlation curve (black line in Figure 3g) throughout the entire period.This indicates that winter Atlantic Niño indeed links the IOD and subsequent year's ENSO.Without winter Atlantic Niño, the relationship between these two phenomena diminishes considerably.
In contrast, when we remove the Niño3.4index (ND 0 J) and calculate the sliding partial correlation between IOD (SON 0 ) and ENSO (ND 1 J) (purple line in Figure 3g), the overall partial correlation coefficient remains roughly the same as the unadjusted simple correlation and does not weaken.This suggests that the self-oscillation of ENSO is not the primary mechanism connecting the IOD and the subsequent year's ENSO.Instead, the Atlantic Niño serves as a crucial intermediary connecting the two phenomena, enabling IOD to have an independent predictive capability for the subsequent year's ENSO.

Root Cause of the Asymmetry
Understanding the asymmetry in the impact of IOD on the winter Atlantic is crucial for comprehending the asymmetric relationship between IOD and the subsequent year's ENSO.Figure 4 compares the atmospheric responses to positive and negative IOD. Positive IODs increase convection over the western Indian Ocean and eastern Africa (Figure 4a), leading to westerly wind anomalies over the Atlantic and triggering the Atlantic Niño.In contrast, negative IODs (Figure 4b) exhibit a subsidence center south of the equator in the central Indian Ocean, which does not generate easterly wind anomalies over the Atlantic, thus not triggering the Atlantic Niña. Figure 4d further demonstrates the asymmetric responses of African precipitation to positive and negative IODs.Negative IODs can cause droughts in Africa (Doi et al., 2022), but the negative precipitation anomalies are weaker compared to the flooding caused by positive IODs of the same intensity.
The aforementioned analysis demonstrates that the asymmetric response of the Atlantic Ocean to IOD is due to the asymmetric responses over Africa and the western Indian Ocean.We propose that this asymmetry is attributed to the climatological mean states (Figure 4e).During autumn, the climatological mean state of the western Indian Ocean and eastern Africa is characterized by descending air and minimal precipitation.Consequently, a positive IOD with warm SST in the western Indian Ocean enables the convective threshold to be surpassed, leading to a substantial enhancement in rainfall.On the other hand, during a negative IOD event, the sinking center shifts to the central Indian Ocean south of equator (Figure 4b) instead of being situated over the Africa continent.This shift may be attributed to the strong climatological mean convective activity over the central Indian Ocean south of equator (Figure 4e).Such a configuration restricts the occurrence of anomalous easterly winds over the Atlantic and the formation of Atlantic Niña during negative IODs.

Model Validation
To further substantiate the "IOD-Atlantic Niño-ENSO" connection, we analyzed CMIP6 multi-model pi-control experiments.Despite some models failing to accurately simulate the asymmetry of IOD (McKenna et al., 2020), our argument remain valid in certain models.Figure S3c in Supporting Information S1 illustrates the asymmetric relationship between IOD and the subsequent year's ENSO in the CMCC-CM2-SR5 model.This relationship is due to both the asymmetric influence of the IOD on the winter Atlantic Niño and the symmetric influence of the Atlantic Niño on the subsequent year's ENSO.These results are consistent with observations.Moreover, the model's mechanisms of asymmetry in the influence of IOD also support the observed results, as demonstrated in Figure S4 in Supporting Information S1 (a model analog to Figure 4).

Applicability to Different Flavors of IOD and ENSO
It is widely recognized that the IOD and ENSO exhibit different flavors (Johnson, 2013;Verdon-Kidd, 2018).Consequently, we have examined the Atlantic intermediary mechanism for different types of IOD and ENSO.S4 in Supporting Information S1): a western-type IOD characterized by a stronger western pole and a weaker eastern pole, and an eastern type IOD characterized by a stronger eastern pole and a weaker western pole (Figures S5a and S5b in Supporting Information S1).Interestingly, the asymmetric impact of IOD is primarily observed in the eastern-type IOD, rather than the western-type IOD (Figure S5 in Supporting Information S1).This may be attributed to the fact that the centroid of SSTAs for the western-type IOD is closer to Africa, resulting in relatively symmetrical impacts in surrounding areas.This finding indirectly supports the mechanism proposed in Figure 4. Due to the predominance of the eastern-type of IOD (28 vs 17 events), the impact of IOD exhibits asymmetry when both types of IOD are combined for statistical analysis.From this perspective, the asymmetric influence of IOD on the Atlantic can be partly attributed to the asymmetrical zonal distribution of SSTA exhibited by most IODs.
Unlike the IOD, different types of ENSO do not show notable differences in their association with preceding Atlantic Niño and IOD.Both the EP and CP types of ENSO have symmetric associations with preceding Atlantic Niño (Figures S6a and S6c in Supporting Information S1).The relationship between IOD and the subsequent year's ENSO exhibits asymmetry (Figures S6b and S6d) due to the inclusion of the asymmetric "IOD-Atlantic" connection in this relationship.

Inter-Decadal Stability
Previous studies (e.g., Park et al., 2023) found that the influence of "summer" Atlantic Niño on ENSO has strengthened since the late 1970s (Figure S7 in Supporting Information S1).However, our focus is on the "winter" Atlantic Niño, and its relationship with subsequent ENSO has remained remarkably constant in recent years (dashed line in Figure S7 in Supporting Information S1; purple line in Figure S8 in Supporting Information S1).As for the "IOD-winter Atlantic Niño" relationship, although some fluctuations exist (green line in Figure S8 in Supporting Information S1), statistical tests indicate that these changes are not significant across different periods.Therefore, overall, the predictive power of IOD on the subsequent year's ENSO remains stable across different decades.

Sensitivity to Different Atlantic Niño Indices
This study uses an Atlantic Niño index (AtlNiño) representing a region broader than the traditional ATL3 region, as mentioned in Section 2. This choice is based on the Atlantic SSTAs responding to positive IOD events, as shown in Figures 2d-2f.We also posit that including a wider region allows for a more accurate depiction of the impact of Atlantic Niño. Figure S9 in Supporting Information S1 compares these two indices in the context of the "IOD-Atlantic-ENSO" connection.Both indices show the asymmetric "IOD-Atlantic" connection and the symmetric "Atlantic-ENSO" connection.Therefore, the two indices do not differ fundamentally in elucidating the findings; rather, the difference lies only in their presentation effects: ATL3 exhibits a lower (yet still significant) correlation with both IOD and ENSO compared to the AtlNiño index.

Summary
We have examined the predictive ability of the IOD on the ENSO in the subsequent year through observational and model studies.Our findings reveal an asymmetry in the influence of positive and negative IODs on the development of ENSO.Specifically, positive IODs have a stronger impact on the subsequent year's ENSO than negative IODs.To explain this asymmetry, we propose the "IOD-Atlantic-Pacific" pathway, with the winter Atlantic serving as an intermediary.This mechanism provides the IOD with independent predictive power for ENSO beyond the self-oscillation of ENSO.This pathway consists of two stages: 1.The asymmetric "IOD-Atlantic" connection: Positive IOD events in autumn can trigger anomalous westerly winds over the Atlantic Ocean by exciting anomalous convection over Africa, leading to the Atlantic Niño in winter.This mechanism is asymmetric, as negative IODs do not effectively trigger winter Atlantic Niña.The asymmetry is attributed to the modulation of the climatological mean state on the distribution of atmospheric responses to IOD events.2. The symmetric "Atlantic-ENSO" connection: Winter Atlantic Niño can induce anomalous easterly winds over the central-eastern Pacific through the Walker circulation modulation, promoting the development of Pacific La Niña during the subsequent spring and summer seasons.This connection is symmetric, as winter Atlantic Niña (Niño) can trigger the development of El Niño (Niña) in the following year.Importantly, this connection operates independently of the self-oscillation of ENSO and has remained stable over several decades.
The Atlantic intermediary mechanism differs from the previous direct "IOD-Pacific" pathway via the oceanic channel and the atmospheric bridge, offering an advantage in explaining the asymmetric relationship between IOD and the subsequent year's ENSO.Further research is needed to explore the relative importance of the "IOD-Atlantic-Pacific" mechanism compared to the direct "IOD-Pacific" pathway.

Figure 1 .
Figure 1.(a)-(c) SST (SON 0 ) correlation maps with subsequent year's Niño3.4(ND 1 J), shown for all samples and separately for El Niño and La Niña years.ENSO classification is based on whether the absolute values of the standardized Niño3.4 (ND 0 J) >0.5.Black contours indicate a significance level of 0.05.Black dots in (b), (c) indicate significant correlation differences between b and c at the significance level of 0.1.(d) Scatter plot of IOD (SON 0 ) and Niño3.4 (ND 1 J) for El Niño and La Niña groups.(e) Scatter plot of Niño3.4 (ND 0 J) and Niño3.4 (ND 1 J) based on the same grouping as (d).Red and blue circles represent standardized Niño3.4 (ND 0 J) >0.5 and <−0.5, respectively.(f) Similar to (d), but grouping is based on standardized IOD (SON 0 ) index >0.5 and <−0.5 (TableS2in Supporting Information S1).In (d)-(f), correlation coefficients are calculated separately for blue and red circles, and an asterisk in the upper right corner indicates a significance level of 0.05.

Figure 2 .
Figure 2. (a)-(l) Composite maps of SST and surface wind (10 m) anomalies during strong positive (a)-(f) and strong negative IODs (g)-(l).Strong IODs are defined as those with a standardized IOD (SON 0 ) index >1.0(Table S3 in Supporting Information S1). Green dots indicate significance at the 0.05 level.(m)-(r) Asymmetry between positive and negative IODs, represented by the difference between the positive IODs and the inversed negative IODs. Green dots indicating significance at the 0.1 level.(s) Scatter plot of IOD (SON 0 ) and Atlantic Niño (D 0 JF) indices.Red and blue circles are classified based on the absolute value of standardized IOD (SON 0 ) >0.5 (TableS2in Supporting Information S1).Correlations for the two groups are denoted with an asterisk in the upper right corner, indicating a significance level of 0.05.(t) Similar to (s), but for the indices after removing the effects of ENSO (ND 0 J).

Figure 4 .
Figure 4. (a) Composite of SON precipitation anomalies (shading; mm/day), 500 hPa vertical velocity anomalies (omega) (contours initiated at ±0.006 Pa/s and proceed at 0.003 Pa/s intervals, with positive values in red and negative values in blue), and 10 m wind anomalies for strong positive IODs.(b) Similar to (a) but for strong negative IODs.Cases were selected based on an absolute standardized IOD index >1.0.(c) Differences between a and inversed b, with significance denoted by purple (precipitation) and orange (vertical velocity) dots at the 0.1 level.(d) Scatter plot of standardized IOD and African precipitation anomaly, averaged over the dashed boxes in (a)-(c).Samples are categorized into blue, gray, and red circles, based on the standardized IOD index threshold of 0.5.Regression lines are fitted to the red and blue circles, respectively.(e) Climatological mean (SON) of precipitation (shading; mm/ day), vertical velocity (contours initiated at ±0.01 Pa/s and proceed at intervals of 0.01 Pa/s, with positive values in red and negative values in blue) and 10 m winds.