• A Bayesian Dynamic Linear Model (BDLM), which represents the temporal modulations of the IOD and ENSO impacts, reproduces the timeseries of the eastern and western Indian Ocean upwellings better than a conventional Static Linear regression Model does.

In the Indian Ocean, intense seasonally reversing monsoon wind forcing prevails, more specifically, southeasterly wind dominates in the southern Indian Ocean through the year, and north easterly wind blows along Somalia, Oman in January and reverse in July (see Figure 2 in Han et al., 2017). In response to the seasonally changed monsoon forcing, the alongshore current (Somalia current) also reverse during January and July. In both the eastern Indian Ocean (EIO) and the western Indian Ocean (WIO), seasonal and annual mean upwelling occurs (see Figure 1; Zhang and Han, 2020). In the western basin around 12^oS-2^oS, open-ocean upwelling exists (McCreary et al., 1993; Murtugudde et al., 1999), which is also referred to as the Seychelles–Chagos thermocline ridge (SCTR; Hermes and Reason, 2008; Yokoi et al., 2008; 2009) and in contract, seasonally changed upwelling occurs along the coastal area of EIO only during boreal summer (Figure 1C; Susanto et al., 2001; Schott et al., 2009).

One of the key processes of the upwelling formation is contribution of oceanic Rossby waves, possibly together with local Ekman Pumping (e.g., Tozuka et al., 2010; Trenary and Han, 2012). Previous works have also suggested that interannual anomaly of depth of 20°C isotherm (D20) in WIO upwelling zone is primarily caused by Rossby waves driven by winds in the central and eastern Indian Ocean. A number of studies indicated the importance of Rossby waves over the southern Indian Ocean (e.g., Woodberry et al., 1989; Périgaud and Delecluse, 1992; 1993; Fu and Smith, 1996; Masumoto and Meyers, 1998; Yang et al., 1998; Chambers et al., 1999; Birol and Morrow, 2001; Wang et al., 2001; White, 2001; Jury and Huang, 2004; Baquero-Bernal and Latif, 2005; Rao and Behera, 2005; Zhuang et al., 2013; Zhang and Han, 2020). The Indonesian Throughflow variations in this area is weak (Potemra, 2001) but can also show significant contributions to the southeast Indian Ocean (see also Trenary and Han, 2012; Deepa et al., 2018; Hu et al., 2019).

Since the wind stress forcing at the sea surface plays an important role in the upwelling directly and/or through the Rossby wave propagations, the wind patterns associated with different climate modes, such as the Indian Ocean Dipole (IOD), the El Niño Southern Oscillation (ENSO) and Asian-Indian monsoon, have gathered much attention (e.g., Zhang and Han, 2020; Zhang and Mochizuki, 2022). By trying to estimate the relative importance of the individual climate modes, many studies have suggested that ENSO dominates the wind-driven Rossby waves south of 10^oS whereas the IOD dominates north of 10^oS (Huang and Kinter, 2002; Xie et al., 2002; Rao and Behera, 2005; Yu et al., 2005; Gnanaseelan and Vaid, 2010). On the other hand, Deepa et al. (2018) suggests that larger sea level variability both north and south of 10^oS during pure positive IOD compared to pure El Niño years, and the co-occurrence of IOD and ENSO significantly enhances the variability in magnitude. Murtugudde et al. (2000) also indicated both the local and remote forcings contribute to interannual anomaly of upwelling in the EIO (also see Han and Webster, 2002; Chen et al., 2016). In fact, Susanto et al. (2002) suggested that ENSO plays a major role in interannual variability of coastal upwelling along the Sumatra and Java coast, while other studies suggested that the IOD is more important than ENSO in causing the EIO upwelling (e.g., Shinoda et al., 2004; Yu et al., 2005; Chen et al., 2016). The estimation of ENSO and IOD contributions still represents large uncertainty as above. In addition to the spatial dependency (e.g., EIO or WIO), temporal modulations can also influence on estimations beyond the seasonal to interannual timescales of ENSO and IOD. In particular, we can speculate that longer-term (e.g., decadal) climate variability can modulate the ENSO and IOD contributions by changing background states. When focusing on ENSO contribution on decadal timescales, the impact of the so-called ENSO flavor, namely, the different types of ENSO events such as Central Pacific (CP) El Nino, Eastern Pacific (EP) El Nino, may also represent distinct impacts on the Indian Ocean upwelling.

A linear model is a simple but effective means to estimate the ENSO and IOD contributions to the Indian Ocean upwelling. By using an advanced Bayesian Dynamic Linear Model (BDLM), Zhang and Han (2020) demonstrated that at interannual timescale, ENSO is more important than the IOD over the SCTR region, but they play comparable roles in the EIO. Furthermore, the impacts of ENSO on EIO upwelling are different between EP and CP events, which is mainly due to the difference of the subsidence of the convection. This implies that the impacts of ENSO and IOD can be modulated on longer timescales corresponding to the background atmosphere and ocean states, while it was out of focus in the previous paper. In this paper, we clarify a major contributor to the decadal modulation of ENSO and IOD impacts on the Indian Ocean upwelling. As a beneficial points of BDLM relative to a simple linear regression estimation, it enables us to demonstrate temporal evolutions of the IOD and ENSO contributions to the ocean upwelling, because the regression coefficients are optimized as time-varying values. We try to give physical interpretation for decadal modulation of the regression coefficients and to understand underlying processes contributing to decadal modulation of the Indian Ocean upwelling. We use satellite data, reanalysis data and the advanced statistical tools (e.g., Bayesian Dynamic Linear Model) for the period of 1958–2016 when long record assimilation data is available.

The paper is organized as follows. A brief description of the data and approach are provided in Section 2. Section 3 describes the observed features of Indian Ocean upwelling associated with IOD and ENSO. Section 4 describes the relationship between the upwelling in the EIO and WIO and the contributions of IOD and ENSO. Section 5 explores the decadal shift of IOD impact on EIO as well as the underlying mechanisms responsible for these changes. The decadal changes of the impacts of IOD and ENSO in relation to the Pacific decadal variability are discussed in Section 6. Concluding remarks and further discussion are provided in Section 7.

Before exploring decadal modulation of ENSO and IOD impacts on the Indian Ocean upwelling, we briefly describe the observed features of Indian Ocean upwelling to show a potential focus in the following analysis. The results indicate that the SSHs in the boxed regions (i.e., EIO and WIO) are key areas to characterize the Indian Ocean upwelling variability, and that the atmospheric and ocean states near the equator should provide us useful information to clarify relating physical processes.

Figure 1A shows the standard deviation of annual mean SSHA based on ORAS4 data during the period of 1958–2016. Clearly, we can see large SSH fluctuations in the WIO (boxed by 50^oE-80^oE, 2^oS-12^oS, about 8 cm) and central Indian Ocean (about 6 cm). Large fluctuation also appears along the coastal area of EIO and the eastern coast of Bay of Bengal (about 6 cm). Figures 1B,C are the composite maps of SSHA during El Nino events and positive IOD events, respectively. During both the El Nino and positive IOD events, southeasterly wind prevails along the southeastern Indian Ocean and easterly wind blows along the equator. There is positive SSHA in EIO and negative SSHA in WIO, where the SSH fluctuations are strongly observed (Figure 1A). Naturally, the ENSO and IOD co-occur events also show a similar spatial pattern (Figure 1D). The satellite observation (Supplementary Figure S1A) and the objective analysis (Supplementary Figure S1C) also show a high variability in WIO and central Indian Ocean as well as along the eastern coast in EIO, consistent to the results of reanalysis data (Supplementary Figure S1B,D). It is known that the weakened convection (e.g., reduced precipitation and enhanced outgoing longwave radiation at the top of the atmosphere) in the eastern basin and the enhanced convection in the western basin is accompanied by easterly wind stress anomalies in the equatorial basin and alongshore winds along Sumatra and Java coast, which will further increase coastal and equatorial upwelling in the EIO and reduce the upwelling in the WIO (Supplementary Figure S2). Meanwhile, the easterly wind anomalies induce negative Ekman Pumping velocity in the eastern basin, and forces positive SSHA which can propagate westward, reducing upwelling in the WIO.

In this section, we verify the timeseries of the EIO and WIO upwellings estimated by SLM and BDLM, which are useful means to discuss decadal modulation of the potential impacts of ENSO and IOD. Figure 2 represents the regressions of the modeled EIO and WIO upwellings onto the annual mean ENSO and IOD indices for the period of 1959–2016. Correlation values suggest that IOD can be slightly more important in EIO (Figures 2A,B), while the relative importance is not clear for the WIO upwelling (Figures 2C,D). Correlation values also indicate superiority of BDLM to SLM in all cases. The errors of the BDLM results are generally smaller than those of the SLM results (Figure 3). In terms of ENSO impact, in the EIO region (the observed standard deviation is 2.6), the Root Mean Squared Error (RMSE) of SLM is 2.3 and the RMSE of BDLM is 1.4. In the WIO region (the observed standard deviation is 3.4), the RMSE of SLM is 3.3 and the RMSE of BDLM is 2.1. Our analysis verifies that the BDLM can represent the temporal modulations of the IOD and ENSO impacts, and well reproduces the timeseries of the EIO and WIO upwellings better than a conventional SLM does. In the following, we try to explain the possible underlying processes that can contribute to improving the ocean upwelling as above, focusing on the regression coefficients mathematically optimized in BDLM estimation.

As discussed in Section 2, the BDLM can measure the time-varying influence of the climate modes, Figure 4 shows the optimized values of the coefficient b₁ in Eq. 2. It should be noted that all four panels of Figure 4 do not represent same time series. In EIO, in particular, the IOD plays a weaker role before 1980s and a stronger influence afterwards in magnitudes (see Figure 4B). In order to see the related processes that may cause this decadal shift in impacts, we calculated regressions of SSTA, precipitation and wind stress onto IOD index for the period of 1958–1979 and 1980–2016 (Figure 5). Weak and strong easterlies in the equatorial Indian Ocean before and after 1980, respectively, are accompanied by differences in atmospheric circulations over the western Pacific and Indian subcontinent. Even though the zonal SST contrast in the equatorial Indian Ocean is assumed to be unity (i.e., discussing regressions onto IOD index), the relatively stronger easterlies along the coastal area of EIO after 1980s drives a relatively stronger upwelling as shown in Figure 4B. In fact, the observed regressions of the autumnal SST in EIO onto the IOD index particularly focusing on the coastal area (95^oW-110^oW, 10^oS-0^o) are −0.48 and −0.66 during the periods 1958–1979 and 1980–2016, respectively, and the difference of these regression values are significant at a 90% confidence limit in F-test.

These differences are primarily due to the background change of ocean temperature. Figure 6 shows the difference in vertical sections of Indo-Pacific ocean temperature and Walker circulation cell before and after 1980s. After 1980, the strong warming tendency is observed in the surface layer of the whole Indo-Pacific Oceans (Figure 6B) together with the changes in Indo-Pacific Walker circulation (Figure 6A). In particular, this ocean surface warming is accompanied by the strong vertical gradient around the ocean thermocline in the eastern Indian Ocean. As a result, the coastal wind in the eastern Indian Ocean induces the IOD signal more effectively in EIO. The optimized values of the coefficient b₁ represents changes of the efficiency in forming the IOD signal in EIO as an ocean response to the surface wind stress forcing aloft, corresponding to changes of the background states of the Indian ocean. Since the IOD index is defined as zonal contrast of the Indian Ocean SST anomalies, the uniform negative anomaly in the Indian Ocean in Figure 5F suggests that the IOD-related SST anomalies represent the enhancing and reducing tendencies in the EIO and WIO, respectively.

From Figure 4A, we also found that there is decadal modulation in impacts of ENSO on EIO upwelling, namely, stronger impact between 1976 and 2002 and weaker impact before 1976 and after 2002. The ENSO and IOD impacts on WIO also show similar timeseries (see Figures 4C,D). These values of b₁ are highly correlated with the so-called PDO index (The correlation values relative to the PDO index 9-year running mean applied are −0.71, 0.64 and 0.44 in Figures 4A,C,D, respectively). Note that, for convenience, here we refer to the above three periods as the negative, positive and negative PDO periods, respectively. Thus, the ENSO has stronger and weaker impacts on EIO upwelling during the positive and negative PDO periods, respectively. Consistent with this BDLM result, the observed regressions of SST in EIO onto the NINO3.4 index represent the significantly larger value in magnitude in the positive PDO period than in the negative PDO period at a 95% confidence limit. We examine the related processes by discussing the differences of regressions in the positive and negative PDO periods as below (Figure 7). Besides, the composite maps (i.e., the background states) represent the PDO-like patterns (Figure 8). The convective activity in the atmosphere is reduced largely over the western Pacific Ocean as a part of the weak Walker circulation (Figure 8A), corresponding to warm upper ocean in the eastern Pacific Ocean (Figure 8B). In contrast to Figure 6B, the composite maps (i.e., the background states) hardly show significant differences in the upper ocean temperature of the Indian Ocean.

We further plot the regression of SSTA, precipitation and wind forcing onto ENSO index during the above defined positive and negative PDO periods (see Figure 7). Even though we always use NINO3.4 index to define the ENSO events, the observed SST anomaly in the eastern Pacific Ocean is slightly shifted eastward in the positive PDO period (Figure 7F). In other words, during the positive and negative PDO periods, the spatial patterns of SST anomaly in a background state are similar to the so-called Eastern Pacific (EP) and Central Pacific (CP) El Nino events, respectively. This is further related to stronger easterly wind along the equatorial Indian Ocean during the positive PDO phase and will drive stronger upwelling in the WIO boxed region. Some studies indicated that EP El Nino events can be more frequently observed in the positive PDO period (Feng et al., 2019), when using two widely used El Nino classification methods: the so-called NINO method (Kug et al., 2009) and the El Nino Modoki index (Ashok et al., 2007). Both the methods define 1 EP (1972/73) and 1CP (1968/69) events during 1958–1975, 4 EP (1976/77, 1982/83, 1986/87, and 1997/98) and 2 CP (1977/78, and 1994/95) events during 1976–2001 and 2 EP (2006/07, and 2014/15) and 3 CP (2002/03, 2004/05, and 2009/10) events during 2002–2016 (e.g., Table 1 in Feng et al., 2019). Note that 2 El Nino events for each of the periods during 1958–1975 and during 1976–2001 are inconsistently classified by the above two methods. This suggests that during the positive PDO period, there are relatively more EP El Nino events, which may lead to stronger equatorial easterly wind in the Indian Ocean, which will further drive stronger upwelling in the WIO. The optimized value of the coefficient b₁ seems to represent changes of ENSO impact primarily due to the distinct ENSO flavor associated with the changes in the background states. The observed regressions of the wintertime SST in central equatorial Pacific (NINO4 region; 160^oE-150^oW, 5^oS-5^oN) onto the NINO index (NINO3.4 index in this study) are 0.57 and 0.71 during the positive and negative PDO periods, respectively. The difference of the regression values of NINO4 SST is significant at a 95% confidence limit, also suggesting the relatively weak contribution of CP El Nino in the positive PDO period.

In this paper, the decadal modulations of ENSO and IOD impacts on the tropical Indian Ocean upwelling are explored by combining the observational data since 1958 and statistical model. The BDLM enables us to demonstrate temporal variations of the contributions of ENSO and IOD by optimizing regression coefficients, thereby showing superiority to the SLM in reproducing the upwelling tendency. We found that the optimized values in the regression coefficients contribute to representing the effects of decadal variations of the background climate states that can modulate the strength of the Indian Ocean upwelling. For example, there is a decadal shift of IOD impact before and after 1980s for the EIO upwelling. This is mainly caused by the shift of wind stress forcing along the eastern coast in magnitudes, corresponding to the changes in sensitivity of the ocean response to atmospheric forcing in relation to the strength of upper ocean stratification (Section 5). The impacts of ENSO on EIO upwelling (and the ENSO and IOD impacts on WIO upwelling) is also modulated on decadal timescales, and the impact is stronger during the period of 1976–2002. Our analysis shows that this is mainly caused by the stronger easterly wind along the equatorial Indian Ocean, dependent on the distinct ENSO flavors dominantly observed in a specific decade (Section 6). Thus, the optimized regression coefficients in BDLM improve the Indian Ocean upwelling particularly on decadal timescales, by reflecting the decadal changes in the Indian Ocean response (i.e., stratification) as in Section 5 and probably in the Pacific Ocean forcing (i.e., ENSO flavor) as in Section 6.

Note here, our results are based on static linear regression model and the Bayesian dynamic linear model, as discussed in Zhang and Han (2020). Thus, the contribution of oceanic waves is not directly estimated but implicitly taken into account in defining regression values in a statistical sense. Since the oceanic wave also plays an important role when moving westward, advanced estimates of its contribution can be a topic for future research. In addition, the upwelling as a response of the ocean to wind stress forcing can represent feedback to wind stress. While, logically speaking, the contribution of this feedback can be grossly represented in our estimation, the detailed coupled processes of atmosphere and ocean are still unclear. Direct verification of coupled climate process modulating the ocean upwelling is also an interesting research questions for future works. Nevertheless, we have built on Zhang and Han (2020), in which they mainly focused on the ocean upwelling on interannual time scales, by showing some important processes that primarily effect on the modulation of the impact of climate modes on decadal timescales. Our results also suggest that the different types of El Nino and the changing atmosphere-ocean state in the climate models towards precise simulations of the impact of climate modes on the Indian Ocean upwelling.