Interannual variations of the Indonesian throughflow



[1] The relationship of Indonesian throughflow (ITF) transport to the El Niño Southern Oscillation (ENSO) is investigated. Seasonal anomalies of ITF transport are found to be forced by anomalous local, alongshore (South Java) wind as well as by anomalous winds in the equatorial Pacific and Indian Oceans. Ocean stratification, set up by surface freshwater fluxes and the ITF, partitions the anomalous forcing onto two separate layers in the ITF region, a near surface layer (approximately the upper 100 m) and a thermocline layer (approximately 100–500 m). Interannual anomalies in alongshore South Java winds and equatorial Indian Ocean winds cause changes in upper ocean ITF transport, while divergent winds over the equatorial Pacific and Indian Oceans control thermocline layer transport anomalies. Interannual periods when the total depth-integrated transport is most weak occur when there are anomalous westerlies in the equatorial Pacific, anomalous easterlies in the equatorial Indian Ocean (resulting in reduced thermocline level transport), and anomalous westerlies along the coast of southern Java (causing reduced upper ocean transport). Such conditions are found when there are cold sea surface temperature anomalies in the Indonesian Seas as well as the equatorial Indian Ocean, and thus are not strictly related to ENSO variability. Interannual variations in the ITF therefore are controlled by interannual variability associated with ENSO as well as by interannual variability in the Indian Ocean.

1. Introduction

[2] This manuscript investigates the seasonal anomalies of monthly mean Pacific to Indian Ocean exchange in the Indonesian throughflow (ITF). The study is focused on how atmospheric anomalies associated with the El Niño/Southern Oscillation (ENSO), the dominant interannual variability in the tropics, as well as other processes can lead to changes in ITF volume transport.

[3] Early theoretical work by Wyrtki [1987] and Clarke and Liu [1994] hinted at a relationship between ITF volume transport and the ENSO. These studies suggested that ITF transport, normally from the Pacific to the Indian Ocean, would be reduced (enhanced) during El Niño (La Niña) events. The hypotheses were based on the change in Pacific trade winds during ENSO events and the associated responses of Pacific and Indian Ocean sea level and the ITF. For example, during El Niño events, easterly trades relax and the pressure force maintaining the ITF is reduced.

[4] Subsequent work by Meyers [1996] showed that the depth integrated transport from Australia to Java, computed from observations of temperature and salinity along the repeat XBT line IX-1 (from Java to Australia; Figure 1) was indeed enhanced during the La Niña of 1988/1989 and reduced during the El Niños of the early 1990s.

Figure 1.

Map of the ITF region; the IX-1 repeat line, from Australia to Indonesia, is given with the black dots.

[5] Yet, in a more recent study, England and Huang [2005] (henceforth referred to as EH05) found a maximum correlation between ITF transport and ENSO of only−0.35, implying that only 12% of the ITF variability is related to ENSO. However, while the sign of the correlation is consistent with previous studies, a major part of the interannual variance in ITF appears to be independent of ENSO. EH05 used the Simple Ocean Data Assimilation (SODA) model, computed geostrophic, upper ocean transport from model temperature and salinity fields (to mirror the Meyers [1996] observations) and used a multivariate ENSO index (MEI) of Wolter [1987]. The results by EH05 yielded a negative correlation (El Niño events lead to reduced ITF transport) at 8 months lag (ITF lagging MEI).

[6] The lack of direct correspondence of ENSO and ITF transport is reflected in Figure 2, which shows the depth-integrated transport along the World Ocean Circulation Experiment section IX-1 from Java to Australia. Transport was computed from geostrophic estimates based on measured temperature and salinity in the upper 400 db [Meyers, 1996] and in the upper 800 db [Sprintall et al., 2002]. The background shading shows the MEI [Wolter, 1987], similar to that of EH05. As in the works by Meyers [1996] and EH05, there are several years when the ITF transport does not appear to match ENSO variations.

Figure 2.

ITF transport was computed from temperature and salinity along the IX-1 repeat XBT line (from Java to Australia). (a) Purple line is from the work by Meyers [1996] who based the calculation on the upper 400 db. (b) Purple line is from the work by Sprintall et al. [2002] who used the upper 800 db and has been fit with an 8-month running mean filter. Negative values indicate an increase in the Pacific to Indian Ocean flow, and the long-term mean has been removed. The shading shows the MEI from the work by Wolter [1987] (scaled by 3 for comparison to the transport).

[7] The apparent discrepancy between ENSO and ITF transport variability may be explained in several ways. First, the observed estimates were based on a geostrophic calculation, and the actual flow may be ageostrophic. However, Ekman transports computed from wind stress along a line from Australia to Indonesia are about 0.5 Sv toward the Indian Ocean and range from 1.5 Sv westward in February to 1.0 Sv eastward in July. Ekman transport computed from the coupled models used in this study give similar magnitudes. In addition, the modeling work of Potemra et al. [2002] showed that flow through the outflow straits was mainly in geostrophic balance, so ageostrophic flow is not a likely cause for the observed interannual changes in ITF transport.

[8] Second, the geostrophic calculation relies on a level of no motion that may not be accurate. This is also unlikely to be a major factor; moorings in specific outflow straits show very weak flows at depth [Molcard et al., 1996, 2001]. The depth of no motion used by Meyers [1996] of 400 db is a little shallow, but subsequent work on the IX-1 data has used a level of 800 db that is more appropriate [Sprintall et al., 2002].

[9] A third reason why ITF transport does not show coherent variability with ENSO is that interannual forcing may act differently on different vertical layers of the ITF, and the result may not have a clear signal in the depth integrated transport value. Potemra et al. [2003] demonstrated that ITF transport occurs in a few (two or three) discrete layers, with the surface layer directly wind forced, and the subsurface layer remotely forced. This possibility is explored in detail in the present study.

[10] Finally, it may be the case that interannual variations in the ITF are caused by something other than, or in combination with, ENSO. Variations in the Indian Ocean that are independent of ENSO, for example, could force seasonal anomalies in the ITF. This was first studied by Murtugudde et al. [1998], who showed that total ITF anomalies correlated to ENSO at only−0.31 (similar to EH05), but the Pacific-forced ITF variations correlated at −0.65, so the non-ENSO interannual variations were coming from the Indian Ocean forcing. Masumoto [2002] used a numerical model forced with observed winds and found interannual variations in ITF transport were correlated with the second empirical orthogonal function of dynamic height and were uncorrelated with the first, suggesting a more complex interaction between ENSO and ITF transport. This is also reflected in the work by Potemra and Schneider [2006], who showed the relationship of low-frequency winds in the Indian Ocean to decadal changes in the ITF transport.

[11] The focus of the present study is twofold, what are the interannual variations in ITF transport and what causes these changes? Our hypothesis is that seasonal variations in wind stress in specific locations force variability in ITF volume transport in two different layers, one above and one within the thermocline. We further hypothesize that local wind anomalies drive upper ocean changes in ITF transport while middepth seasonal anomalies in ITF transport are forced by interannual variations in large-scale wind divergence over the maritime continent. To study this, we used a SODA integration, along with results from two different coupled climate models, Parallel Climate Model (PCM) and Scale Interaction Experiment (SINTEX), described next.

1.1. Model Description

[12] The present study relies on three different numerical models. The first is the Simple Ocean Data Assimilation model (SODA POP v1.4.2; J. A. Carton and GB. Giese, SODA: A reanalysis of ocean climate, submitted to Journal of Geophysical Research, 2005), which was forced by European Centre for Medium-Range Weather Forecasts winds from 1958–2001 (henceforth referred to simply as SODA). The ocean model is based on Parallel Ocean Program physics with an average horizontal resolution of 0.25° × 0.4° and 40 vertical levels. Assimilation is based on optimal interpolation and uses virtually all available hydrographic profile data, as well as ocean station data, moored temperature and salinity time series, surface temperature and salinity observations of various types, and nighttime infrared satellite sea surface temperature (SST) data. The output is in monthly averaged form, for 1958 to 2001, mapped onto a uniform 0.5° × 0.5° grid.

[13] In addition, two coupled climate models were used. These models, while lacking the spatial resolution of SODA, have been integrated for longer periods and provide internally consistent coupled physics not affected by the internal heat sources introduced by optimal interpolation in SODA. The first model is NCAR’s PCM, and the second is the European Union’s SINTEX model.

[14] The PCM is the result of a joint effort between the Los Alamos National Laboratory (LANL), the Navy Postgraduate School (NPS), the US Army Corps of Engineers’ Cold Regions Research and Engineering Lab, and the National Center for Atmospheric Research (NCAR). This group coupled the NCAR Community Climate Model (CCM-3), the LANL Parallel Ocean Program (POP), and ice model from NPS. This is the so-called version 1 of PCM (PCM1). The model has been applied to ITF studies in the work by Potemra and Schneider [2006] and further details can be found in the works by Washington et al. [2000] and Meehl et al. [2001].

[15] The atmospheric model (CCM-3) is on a 128 × 64 grid, with an approximate spacing of 2.8125° × 2.8125° (the latitudinal spacing is Gaussian). The ocean model, on the basis of POP, has 32 vertical levels and is 384 × 288 in the horizontal. The horizontal grid has an average resolution of equation image and equation image near the equator. It should be noted that this version of the PCM does not include Gent and McWilliams’ [1990] mixing, and thus tracer distribution and pressure gradients may be affected. We analyze 299 years of model integration in this study.

[16] Results from the European Union’s SINTEX model are also used. The atmospheric component of SINTEX is the ECHAM4, and the ocean component is the Océan Parallélisé (OPA version 8.1) from the Laboratoire de Physique de Océans. This model has been used to investigate interannual variability in the both the Pacific and Indian Oceans [Gualdi et al., 2003a, 2003b]. The vertical grid in the ocean model is finer than the PCM. We analyze the results for the upper 22 levels from OPA, approximately the upper 1000 m. The horizontal resolution of OPA is somewhat less than the PCM, roughly equivalent to 2° × 1.5° on average and 2° × 0.5° near the equator.

[17] Each of these models has its limitations, for example, the coupled models have the common split Intertropical Convergence Zone and Pacific cold tongue biases. The SODA model, since it assimilates data at all levels, does not conserve heat. The present study, however, looks at relationships between ENSO, local and remote winds, and the ITF transport, and it is thought that the limitations in these models is not critical. The PCM and SINTEX models have ENSOs with a very regular 24- to 30-month period (from spectral analysis of eastern Equatorial Pacific SST anomalies). The magnitude of the SST anomalies is comparable to observations, but the zonal extend is somewhat exaggerated. Interannual variability in the Indian Ocean is close to observations as well (a more complete analysis is given in the work by Saji et al. [2006]).

[18] Previous studies [Gualdi et al., 2003b; Terray et al., 2005; Cherchi et al., 2006] have shown that the SINTEX model, in general, appears to be able to reproduce the main features of the Indian Summer monsoon precipitation and circulation climatology, as well as the annual timing of the monsoon onset and the interannual variability of its intensity. The model, however, tends to underestimate the monsoon rainfall over India and over the Bay of Bengal. This lack in monsoon precipitation is a well-known bias of ECHAM4, present also when the model is forced with observed SSTs [Roeckner et al., 1996], and that might be due to deficiencies in the parametrization of convection [Terray et al., 2005].

1.2. ITF Mean

[19] The ITF is mainly forced by Pacific and Indian Ocean winds [Wyrtki, 1987; Godfrey, 1989]; thermohaline forcing is thought to be minimal [Shriver and Hurlburt, 1997]. The mean, depth-integrated ITF transport has been estimated to be close to −10 Sv (see the work by Gordon [2005] for a review). The total ITF transport computed from SODA is −12.8 Sv, from PCM, it is−12.0 Sv, and from SINTEX, it is−9.6 Sv; all within current estimates.

[20] These models, like observations, show that most of the ITF source waters originate in the North Pacific. South Pacific water contributes to a lesser degree and at a deeper depth. This may contribute to the vertical stratification of the ITF, and this is discussed in the next section.

[21] Godfrey [1989] showed that the mean, barotropic ITF transport can be estimated from the line integral of wind stress along a path from the southern tip of Australia to South America, north along the coast to the equator, east to New Guinea, and south along the west coast of New Guinea and Australia. In a series of papers, Wajsowicz found that this calculation could be expanded to include low-frequency ITF variability and that baroclinic adjustments would occur at the shallow sills of Indonesia [Wajsowicz, 1994, 1995, 1996]. Using the wind stress and ITF computed from the coupled models used in this study, there is no significant correlation between the path integral of wind stress and seasonal anomalies of ITF transport.

1.2.1. Vertical Structure

[22] The ITF and its associated waters form a core of relatively high salinity (34.5 psu) that can be traced in observations zonally between about 8° and 12°S westward from the ITF region [Talley and Sprintall, 2005]. Along the coast of Java and Sumatra, fresh water from surface fluxes in the Indian Ocean form a surface layer and result in a two layer system roughly delineated by 20° and 10°C isotherms (d20 and d10, respectively). The two layers act much like a second baroclinic mode in that the d20 deepens when the d10 shoals.

[23] Almost all of the ITF transport occurs in these two layers. Potemra [2005] showed that 83% of the total ITF transport occurs above 200 m (from an earlier SODA integration). In the present study, the vertical stratification occurs at about 100 m. The mean SODA d20 off Sumatra/Java is 115 m, while the mean d10 is 471 m. The net SODA ITF transport is −12.8 Sv, and −6.6 Sv is carried in the upper 104 m (top nine model levels), and another −5.5 Sv is carried in the 104–522 m range (10 model levels).

[24] For purposes of this analysis, ITF transport computed over the fixed-depth ranges described above gives similar results to transport computed over temperature layers (for example, transport computed above the d20). This is because most transport occurs in the near surface and because the model vertical levels become thicker with depth (and thus the d20 and d10 usually reside within a few depth levels). The mean d20, for example from SODA, computed along the IX-1 line goes from 90 m near Java to 150 m near Australia. The annual amplitude is high near Java (mean annual cycle ranges from 60 to 115 m) and small near Australia (145–155 m). Interannual variability of the d20 is about ±60 m along the IX-1 section.

1.3. ITF Seasonal Cycle

[25] During November through March, the winds along the archipelago are to the east, and surface flow is from the Java Sea (and Pacific) into the Banda Sea. Alongshore winds are not favorable to remove this water, and the resulting Ekman flow causes warm surface waters to accumulate in the Banda Sea and depress the thermocline there [Wyrtki, 1987]. Simultaneously, the large-scale pressure gradient from the Pacific to the Indian Ocean weakens, and the upper ocean ITF transport becomes weak. From May to September during the northern summer monsoon, the winds are more intense and to the northwest over the entire Indonesian region. At this time, the winds tend to remove surface water from the Banda Sea into the Indian Ocean, and the upper ocean ITF transport is strongest.

[26] Twice during the year, during the monsoon transition periods, westerly wind anomalies in the equatorial Indian Ocean force downwelling equatorial Kelvin waves. These downwelling waves then propagate along the southern Sumatra/Java coasts as coastal Kelvin waves. The resulting high sea level and deep upper layer force a reduction in upper ocean ITF transport in May and November.

[27] These features are reproduced in the three models. The coupled models have reduced semiannual variability, most likely because of vertical resolution or poor representation of the monsoons.

2. ITF Transport Anomalies

[28] As a first step, interannual variability in ITF transport was compared with ENSO variability by correlation analysis. All calculations were made with seasonal anomalies, computed by subtracting a mean seasonal cycle from each time series (henceforth referred to as anomalies). Figure 3 shows the correlation of MEI to transport anomalies (in each model level) through a section from Java to Australia using the SODA velocity field.

Figure 3.

The colors show the correlation of SODA transport anomalies (per model level) to MEI as a function of depth. The left side is for correlations with transport anomalies leading MEI, and the right is for ITF lagging MEI. The 95% level is contoured.

[29] The correlation of ITF to ENSO is maximum at depth, around 200 m, with a positive MEI resulting in weaker ITF a few months later (throughout this paper we define a reduction in ITF transport as a positive anomaly). The region of positive correlation extends to 500 m. The correlation reveals a near surface layer (0 to about 100 m) having a weak, negative correlation to MEI. The depth dependence of the correlation could explain the relatively low correlation of total-depth (or, surface to 400 db) integrated ITF transport with ENSO.

[30] These two layers coincide with the d20 and d10 described above. In addition, seasonal anomalies of d20 are out of phase with the anomalies of d10 (correlation of −0.28) in the eastern equatorial Indian Ocean (Figure 4). For example, in the recent ENSO year 1997/1998, d20 becomes shallower in the equatorial Indian Ocean while d10 becomes deeper. This two-layer structure is maintained along the southern coasts of Sumatra and Java.

Figure 4.

Seasonal anomalies of the 20°C isotherm depth (d20; left panel) and 10°C isotherm depth (d10; right panel) computed from SODA (last 10 years shown; note different color scales on each panel). The horizontal axes are the distance along the equator (starting at 90°E) to Sumatra then south along the coasts of Sumatra and Java. Gaps are due to outcropping (in the case of d20) or intersection with the bottom (in the case of d10).

[31] The d10 and d20 are most likely set by Pacific processes, but they respond to Indian Ocean winds and perhaps variations in the ITF [Potemra and Schneider, 2006]. The variability of these two layers reflects variability in the ITF. For this study, the ITF transport is integrated over two fixed-depth layers defined by the mean d20 and d10. Upper layer transport (ULT) is computed for the surface to 104 m (top nine SODA levels), and middepth layer transport (MLT) is computed from 104 to 522 m (10 SODA levels).

[32] Using transport over the temperature layers gives similar results (Figure 5) as described earlier. The correlation of monthly mean ITF transport above the d20 and transport above 104 m is 0.95 (using SODA results). When the mean seasonal cycle is removed, the correlation is still high (0.89). Transport computed between the d20 and d10 has a similar correlation to transport between 104 and 522 m (0.82 for monthly mean values and 0.82 for monthly anomalies).

Figure 5.

The green lines show monthly mean ITF transport anomalies computed over fixed depth levels (ULT, 0–104 m transport anomalies, and MLT, 104–522 m transport anomalies). The purple lines show the monthly mean ITF transport anomalies above the d20 and between the d20 and d10. The upper panels are for 1955–1980, and the lower panels are for 1980–2005, and positive anomalies indicate a reduction in the Pacific to Indian Ocean flow.

2.1. ULT/MLT Seasonal Anomalies

[33] Monthly mean upper layer ITF transport anomalies (0–104 m; ULT) and middepth transport anomalies (104–522 m; MLT) are given in Figure 6. In some cases, the ULT is out of phase with the MLT, as suggested by the analysis of d20 and d10 (here we are computing transport over a fixed-depth level rather than over a temperature level). The correlation of ULT anomalies to MLT anomalies in SODA is only 0.21, just above the 95% confidence level. On the other hand, the correlation of total depth integrated transport to ULT anomalies is 0.43 and for MLT, it is 0.69.

Figure 6.

The black lines show the ULT (0–104 m transport anomalies), MLT (104–522 m transport anomalies), and total ITF transport anomalies from SODA. The MEI constructed from observations [Wolter, 1987], scaled by 3 to be more clear, is given with the shading. The upper panels are for 1955–1980, and the lower panels are for 1980–2005.

[34] As in Figure 3, the ULT has a negative and very weak correlation to Nino-3 of −0.14, just within the 95% confidence level, with transport leading Nino-3 by 1 month and indicates that increased ULT toward the Indian Ocean is coincident with positive Nino-3 SST anomalies. The correlation of MLT to Nino-3 is higher (0.39) and is in the sense of warm ENSO events lead to a reduction in MLT. The total ITF transport correlation to Nino-3 is 0.32, similar to that of EH05 and Murtugudde et al. [1998].

3. Interannual Forcing of the ITF

[35] In order to better understand the relationship between ENSO and ITF, results from the PCM and SINTEX coupled models were used. The longer time series (299 and 150 years, respectively) allows for a more robust statistical analysis, and all the previous discussions also apply to both these models. First, transport was computed from each model’s velocity field through a section from Java to Australia. The depth integration was done in two separate layers, similar to the previous discussion. The seasonal transport anomalies in these two layers was then compared to the model wind stress.

3.1. Coupled-Model Results

[36] The correlation of zonal wind stress anomalies to anomalies in ITF transport from the PCM are given in Figure 7. At zero lag, the ULT anomalies (color shading) correlate to zonal wind stress anomalies in the eastern Indian Ocean and within the Indonesian seas. When wind anomalies are westerly in this region, the upper layer ITF becomes weaker. This is consistent with a reduction in the Pacific to Indian Ocean pressure head [e.g., Wyrtki, 1987], as well as in the geostrophic sense of westerly winds causing an increase in model dynamic height on the northern side of the throughflow passage. In addition, westerly winds would generate a northward Ekman flow through the Indonesian seas.

Figure 7.

Correlations of model ITF transport anomalies to zonal wind stress anomalies from the PCM. The correlation with upper layer transport (ULT) is shaded, and the correlation to middepth transport (MLT) is contoured; only values about the 95% confidence level are shown. Positive correlation means anomalously weak ITF transport coincident with easterly zonal wind stress anomalies.

[37] More interesting perhaps is the correlation of the MLT anomalies to the zonal wind stress (contours in Figure 7). The middepth flow appears to be controlled by divergent winds in the equatorial Indian and Pacific Oceans. When winds in the equatorial western and central Pacific are anomalously westerly and zonal winds in the equatorial Indian Ocean are easterly (divergence), MLT is anomalously weak 2–3 months later. This correlation is particularly large for October November December (OND) wind stress anomalies and December January February (DJF) transport anomalies (r = 0.72; not shown).

[38] The relationship between local winds and surface flow, and equatorial Indian/Pacific Ocean winds to middepth flow is the same in the SINTEX integration as with the PCM. Figure 8, made from the SINTEX results, is analogous to Figure 7, and shows that in both models, ULT anomalies are correlated to local wind stress anomalies and MLT anomalies are correlated to zonally divergent winds. Such divergent winds, however, do not always occur during El Niño years; this is discussed more in the final section.

Figure 8.

Correlations of model ITF transport anomalies to zonal wind stress anomalies from SINTEX. The correlation with upper layer transport (ULT) is shaded, and the correlation to middepth transport (MLT) is contoured; only values about the 95% confidence level are shown. Positive correlation means anomalously weak ITF transport coincident with easterly zonal wind stress anomalies.

4. ITF ENSO Composites

[39] The SODA correlations (not shown) are similar to the coupled-model results, albeit with less degrees of freedom. To investigate the ITF transport relationship to ENSO, composites were made using the SODA results (Figure 9). An ENSO year was defined to be when the SST anomalies during December through February in the Nino-3 region exceeded one standard deviation. The mean ULT and MLT seasonal anomalies over the year prior to this through a year following this were then computed for all such events.

Figure 9.

Mean ITF transport anomalies were computed for 1 year prior (year − 1) through 1 year following (year + 1) each El Niño event. An El Niño year was defined when the mean SST anomalies in the Nino-3 region for Dec-Jan-Feb (years 0/1) exceeded 1 standard deviation. The composite ULT is given in green, MLT in purple, and total ITF transport anomalies in black (dashed lines given statistical significance). The composite SST anomalies are shaded in brown.

[40] The result shows clearly the out-of-phase relationship between ULT and MLT during El Niño years. During year 0, just prior to maximum warming of SSTs in the eastern equatorial Pacific, the MLT is reduced by 1 to 2 Sv, while the ULT is enhanced by a lesser amount. More specifically, the ULT becomes stronger in May and November, when the monsoon transition westerly anomalies usually act to reduce the ULT. It appears in SODA that during El Niño years, the May/November reduction in ULT is not as large as during normal years, perhaps because of a decrease in the westerlies. Such a reduction in equatorial westerlies during El Niños was reported in the work by Grodsky et al. [2001].

[41] The easterly wind anomalies over the equatorial Indian Ocean act to thin the upper layer (the d20 becomes more shallow), while the lower layer thickens (d10 becomes deeper), as noted in Figure 4. In this way, divergent equatorial winds in the equatorial Pacific/Indian Ocean act to reduce MLT (and enhance the ULT). Westerly winds alongshore South Java can then act to reduce the ULT or at least reduce the coastal waves generated in the equatorial Indian Ocean. Thus, total ITF transport becomes anomalously low when winds in the equatorial Pacific and alongshore South Java are westerly and when winds in the equatorial Indian Ocean are easterly. This does not happen during all warm ENSO events, and thus the ITF-ENSO correlation is not as strong as it otherwise would be.

5. Discussion

[42] As described above, minimum ITF transport is found when both the ULT and MLT are reduced. The former occurs when equatorial Indian Ocean and winds along South Java are westerly. Reduced MLT occurs when winds in the equatorial Pacific are westerly and winds in the equatorial Indian Ocean are easterly (large-scale divergent winds over the maritime continent). Therefore equatorial Indian Ocean winds, through coastal wave processes and baroclinic adjustment, act in opposite ways on ULT and MLT (for example, westerly winds in the equatorial Indian Ocean simultaneously reduce ULT and increase MLT).

[43] Reason et al. [2000] provide seasonal composites of surface winds in the Indian Ocean during El Niño events. During most of the composite El Niño year (December having maximum warming in the eastern Pacific), winds are divergent over the maritime continent. These composites show that during most El Niño events, wind anomalies are favorable to a reduction in MLT, but the effect on ULT can be variable. This is consistent with the results described above (MLT had a higher correlation to ENSO than did ULT).

[44] In the work by Meyers [1996], there are specific years when the ITF transport was low despite no El Niño event (for example, mid-1985, mid-1990, and early 1994), as well as El Niño years when the ITF transport was higher than normal (for example, late-1983). This could be due to wind patterns that do not adhere to the typical ENSO conditions, or it could be that flow at deeper depths (below 500 m) is playing a role.

[45] As an example, Figure 10 shows the ULT, MLT, and total transport during the 5-year period from 1993 to 1997. The MEI was positive during 1993, again positive in early 1994 through mid-1995, and then strongly positive during 1997. Noteworthy during these three time periods is that total ITF transport was enhanced during 1994 (during El Niño conditions) and then reduced during late 1996 (weak La Niña to normal conditions). Further, total ITF transport was only slightly reduced (about 1 Sv) during the extreme El Niño of 1997/1998.

Figure 10.

The upper panel shows the ULT (green), MLT (purple), and total transport (black) anomalies. Negative values indicate an enhanced throughflow. The lower panel shows the zonal wind anomalies averaged over three regions: alongshore South Java (105°–125°E, 12°–10°S; green line), the western equatorial Pacific (140°–160°E, 3°S to 3°N; blue line), and the eastern equatorial Indian Ocean (75°–95°E, 3°S to 3°N; red line). In both panels, the MEI is shaded and scaled to fit.

[46] These discrepancies may be explained in the following way. The enhanced ITF transport in 1994/1995 was due to an increase in MLT followed by a longer period of enhanced ULT (Figure 10a). Positive wind anomalies occurred in the equatorial Indian Ocean (leading to a reduced ULT and enhanced MLT) and in the equatorial Pacific Ocean (leading to an enhanced MLT; Figure 10b). Wind anomalies alongshore South Java were negative, leading to increased ULT. Thus the total transport was increased.

[47] The reduction of ITF transport in late 1996 was caused by positive wind anomalies in the Indian Ocean, which reduced the ULT, and there were no strong wind anomalies along South Java to change the signal in ULT (Figure 10b).

[48] In 1997, the reduction in total ITF transport was smaller than would otherwise be expected during such an extreme El Niño because the ULT was actually stronger than normal during this time (Figure 10a). Recall that South Java winds are critical for reducing/enhancing the ULT, and in 1997, the winds along South Java were anomalous easterlies (Figure 10b), leading to the enhancement in ULT.

[49] It could be asked if such wind anomalies, i.e., divergent winds over the maritime continent and westerly South Java winds, are associated with signals beside ENSO. In fact, in both the PCM and SINTEX models, divergent wind over the maritime continent are not well correlated to ENSO but rather are correlated to SST anomalies in the Indian Ocean reminiscent of the Indian Ocean Dipole.

[50] Figure 11 shows the correlation of SST anomalies from SODA to the observed MEI (upper panel), to divergent wind stress over the maritime continent (center panel), and to winds along southern Java.

Figure 11.

The upper panel shows correlation of monthly mean SST anomalies from SODA to the observed MEI. The lower panels are (b) the correlation of SST anomalies to divergent winds over the maritime continent and (c) the correlation of SST anomalies to zonal winds along southern Java. Correlations are at zero lag, and only those above 95% confidence level are shown.

[51] The SST pattern associated with ENSO (Figure 11a) is one of warming in the eastern tropical Pacific and throughout the Indian Ocean. Divergent winds over the maritime continent, however, are associated with cold SST anomalies in the east Indian Ocean and throughout the Indonesian Seas (Figure 11b). This broad area of cold SSTs, roughly between 100° and 150°E, should be associated with anomalous atmospheric subsidence (a decrease in the normal convection over the maritime continent). The corresponding increase in convection occurs in the central Pacific and central Indian Ocean, where warm SSTs are correlated to the divergent winds. This scenario is when the MLT becomes anomalously low.

[52] ULT, on the other hand, becomes weak when the winds along shore southern Java are westerly (positive anomalies). Such winds are coincident with cold SSTs in the equatorial Indian Ocean and (weak) warm anomalies in the western Pacific warm pool (Figure 11c).

[53] A possible scenario is as follows: Initially, anomalous easterlies in the Indian Ocean near Sumatra create colder SSTs. A month later, these easterly winds create warm anomalies on the western side of the Indian Ocean. A SST gradient similar to the Indian Ocean dipole is then formed. This enhances zonally divergent winds over the maritime continent. At zero lag, easterly anomalies in the Indian Ocean coincident with westerly anomalies in the Pacific Ocean occur when SSTs are anomalously cold in the eastern Indian Ocean, warm in the western Indian Ocean, and warm in the equatorial western and central Pacific. The important point from this study, however, is that the zonal wind anomalies that are correlated to middepth ITF transport anomalies are not correlated with ENSO.

[54] It should also be pointed out that a large effect in the Indian Ocean is the semiannual Kelvin waves generated by westerly winds during the monsoon transition months. The anomalous forcing of this signal has been shown in this study to have an important effect on the anomalous ITF transport signal. In fact, intraseasonal variability is a major part of the upper ocean transport anomalies (see Figure 2). Low-pass filtering of the ITF could mask some of these important anomalies and be misleading.

[55] Finally, it should be noted that the results presented here focused on the upper 100 m (ULT) and the 100–500 m ITF transport anomalies (MLT). While these two layers account for more than 85% of the total mean ITF transport, monthly anomalies from a deeper layer can be quite significant. This is seen, for example, during early 1993 (Figure 10a) when the total transport was reduced by up to 6 Sv (February 1993), while ULT and MLT account for less than a third of that. The forcing of this deeper ITF layer is might be due to higher baroclinic modes in the Pacific and will be the focus of a future study.


[56] The SODA output was generously provided by Ben Geise. David Pierce and Silvio Gualdi were instrumental in getting the output from the PCM and SINTEX runs, respectively. This research was supported by the Office of Science (BER), US Department of Energy, Grant DE-FG02-04ER63862, and by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) through its sponsorship of the International Pacific Research Center. This is School of Ocean and Earth Science and Technology Contribution number 7035 and International Pacific Research Center Contribution number 429.