Multidecadal variability in the transmission of ENSO signals to the Indian Ocean

1. Introduction

[2] Pacific Ocean variability is known to transmit to the Indian Ocean. The consensus is that variations in zonal Pacific equatorial winds force a response primarily along the Western Australia (WA) coast [Clarke, 1991; Meyers, 1996; Masumoto and Meyers, 1998; Potemra, 2001; Wijffels and Meyers, 2004; Cai et al., 2005a, 2005b]. The energy off the WA coast arises from equatorial Rossby waves generated by zonal wind anomalies in the central equatorial Pacific which become coastally trapped waves where the New Guinea coast intersects the Pacific equator. In addition to the equatorial pathway, Cai et al. [2005a] found an off-equatorial pathway: using an updated thermal analysis covering over 20 years since 1980 from the Australian Bureau of Meteorology Research Center (BMRC) [Smith, 1995], they showed that subtropical NP Rossby waves associated with ENSO impinge on the western boundary and move equatorward along the pathway of Kelvin-Munk waves [Godfrey, 1975], and reflect as equatorial Kelvin waves. The reflected Kelvin waves impinge on the Australasian continent and move poleward along the northern WA coast as coastally-trapped waves, radiating Rossby waves into the south Indian Ocean. Cai et al. [2005a] showed that some 55% of the total interannual variance of the WA thermocline is linked to the subtropical NP Rossby waves and participate in the ENSO recharge-discharge [Jin, 1997; Meinen and McPhaden, 2000]. Here we examine the robustness of the amount of energy transmission via each of the pathway in the pre-1980 period and find that there are significant differences to the post-1980 period. We then examine the presence of such multidecadal variability in a coupled climate model.

2. Data and Method

[3] We utilize the newly available reanalysis of SODA-POP (version 1.4.2) (J. A. Carton and S. B. Giese, SODA: A reanalysis of ocean climate, submitted to Monthly Weather Review, 2006). The new model product uses the European Center for Medium Range Forecasts ERA-40 atmospheric reanalysis winds. It has a spatial resolution of 0.5° latitude by 0.5° longitude grid, and covers a period from 1958 to 2001. Both the SODA-POP and BMRC reanalyses independently incorporated Expendable Bathythermograph profiles and time-series from the Tropical Atmosphere Ocean buoy array [McPhaden et al., 1998]. We find that there are remarkable differences in the transmission between the pre- and post-1980 periods. To examine the robustness of such multidecadal variability, we take outputs of a multi-century control experiment with the new CSIRO coupled climate model (version 3.5). The new version simulates a more realistic transmission process, although it still suffers from the common cold tongue bias, i.e., the equatorial Pacific cold tongue extends too far west. The ENSO frequency is reasonably simulated as reported earlier [Cai et al., 2004], but the amplitude is too large. Despite these deficiencies the model produces similar multidecadal variations in the transmission.

3. ENSO Cycle and Transmission Into the Indian Ocean in the Pre- and Post-1980 Period

[4] Figure 1 displays the lag-correlation between Niño3.4 and D20 at various lags for the post-1980 (Figure 1, left) and pre-1980 (Figure 1, right) from SODA-POP. The ENSO cycle is well simulated in both periods but with noticeable differences. The meridional extent is narrower in the pre-1980 period. This difference, and the feature of stronger ENSO since 1980, have been observed by previous studies [Wang, 1995; Wallace et al., 1998; Wang and An, 2001]. A central difference is that little signal is transmitted into the Indian Ocean in the pre-1980 period.

[5] In the post-1980 period (Figure 1, left), 9 months prior to the peak of an El Niño, the pattern in the equatorial Pacific (5°S–5°N) shows a recharged phase, but an off-equatorial upwelling Rossby wave (indicated by negative contours) develops and radiates from the eastern boundary, and is reinforced in the vicinity of (155°W, 17°N). After a strong growth en-route westward, it impinges on the western boundary, moves equatorward and then reflects back as an equatorial Kelvin wave (Lag −3). The reflected Kelvin wave then forces a coastally-trapped wave, which propagates poleward along the WA coast, contributing to a discharged phase of an El Niño at Lag 0. The discharge off the WA coast reaches a maximum approximately three months after an El Niño peaks (Lag +3). Thus some of the signal along the central WA coast can be traced to the subtropical NP. This is the subtropical NP pathway described by Cai et al. [2005a] and SODA-POP simulates this well. The phase speed of the off-equatorial Rossby waves is far faster than that of observed Rossby waves [Chelton and Schlax, 1996; Cipollini et al., 2001]; however, these waves are not free Rossby waves but are strongly controlled by wind anomalies or by the atmosphere-ocean coupling [Cai et al., 2005a].

[6] Rossby waves in the pre-1980 period are closer to the equator, mostly within about 10°S–10°N. As a result, there is little transmission via the subtropical NP pathway. This is further illustrated in Figure 2, which shows lag correlation between D20 at the NP western boundary (Philippine Sea (PS) box, 120°E–125°E, 12.5°N–17.5°N) and D20 everywhere. To compare with Figure 1, the PS D20 is sign-reversed so that a discharge signal is represented by negative correlations. Rossby wave propagation is seen in both periods, but in the post-1980 period (Figure 2, left), there are strong coherence between ENSO and NP Robbsy waves and clear signal transmission to the WA coast; these features are virtually absent in the pre-1980 period (Figure 2, right).

[7] Does the energy leave the Pacific Ocean via the equatorial pathway in the pre-1980 period? We conduct a lag-correlation analysis of D20 anomalies with time series of D20 anomalies averaged over a central WA box (112°E–120°E, 15°S–22°S) (Figure 3). To compare with Figure 1, the WA D20 is sign-reversed. Maximum discharge off the WA coast appears at Lag 0, and at Lag −3 it corresponds to an El Niño peak (at Lag 0 in Figure 1).

[8] For the post-1980 period (Figure 3, left) there is a strong similarity between Figures 1 and 3, and the WA anomaly is predominantly generated by ENSO processes. In the pre-1980 period (Figure 3, right), the evolution is vastly different. There is little correlation between WA D20 and anomalies elsewhere at most lags, except at Lag −6, when weak but significant correlations exist in the western equatorial Pacific. Corresponding maps of correlation with zonal winds also show a maximum in the western equatorial Pacific at Lag −6, implying that some Pacific signals do propagate through the equatorial pathway. Nevertheless, the overall lack of correlation suggests that in the pre-1980 period the transmission via the equatorial pathway is so weak that it does not manifest above the stochastic noise.

[9] What we have described above is the difference of the statistical properties between the two periods. Within each period, the proportion of energy transmission via each pathway varies significantly from one event to another; for example, in the 1997 event, transmission via the NP pathway is smaller than that via the equatorial pathway. Despite this, it is rather significant that the statistical property of events over one 20-year period is so different from that over another 20-year period, highlighting the existence of an underlying mechanism.

[10] The thermocline in the Pacific Ocean has been changing on decadal timescales, and could have affected the Indian Ocean [McPhaden and Zhang, 2002; Annamalai et al., 2005]. This is supported by our results: the stronger post-1980 ENSO discharge signals contribute to a shallowing thermocline in the southern tropical Indian Ocean (figure not shown) and could affect the development of the Indian Ocean Dipole (IOD), by pre-conditioning a shallower thermocline [Annamalai et al., 2005] as the associated stronger easterly anomalies lift the thermocline. These might explain the better defined IOD pattern during the post-1980 period (Figure 1, bottom left).

4. Dynamics

[11] Pre-1980 ENSO events are weaker and have a narrower meridional extent than the post-1980 ENSOs (Figure 1) [Wang, 1995]. Does such multidecadal variability in the ENSO properties contribute to the difference in the pre- and post-1980 transmission? We take outputs of the new CSIRO multi-century control experiment and examine if similar multidecadal variations exist. Time series of WA D20 and Niño3.4 are constructed from the coupled model outputs and a 20-year sliding window is used to calculate the correlation between them at Lags +3 and +6 (Figure 4, black and blue curves). We calculated more than one lag in case the model transmission signal does not peak at exactly the same time as in SODA-POP. The model transmission undergoes similar multidecadal fluctuations: in some 20-year periods the correlation is not significant, i.e., little is transmitted or generated; in other periods the correlation reaches as high as 0.8. We then calculate a time series of standard deviation of Niño3.4 using a 20-year sliding window (Figure 4, red curve). The amplitude fluctuates significantly, between about 0.6°C and 1.1°C. The central point is that a strong correlation exists between the standard deviation curve (red) and the Lag +3 curve (black) with a correlation of 0.79: a strong transmission is seen when ENSO events are strong, and vice versa.

[12] Maps of correlation (not shown) between Niño3.4 and D20 everywhere and between the WA D20 and D20 everywhere for the strong (centered at year 315) and weak (centered at year 215) transmission periods resemble those of Figures 1 and 3 for the post- and pre-1980 periods, respectively. For the weak transmission period (year 215), there is little involvement of the subtropical NP pathway and the meridional extent of the ENSO anomaly is narrower. These contrasts are also reflected in maps of one-standard deviation anomaly patterns of SST and surface wind associated with ENSO for the strong (Figures 4b and 4d) and weak (Figures 4c and 4e) transmission periods, reminiscent of the difference between the post- and pre-1980 periods in SODA-POP.

[13] During strong-ENSO periods, the tropical Indo-Pacific system is overwhelmed by ENSO signals, therefore the ratio of “ENSO signal to stochastic noise” is greater than that during weak-ENSO periods. To illustrate this, we define “signal” as the standard deviation associated with the Niño3.4, determined from a linear regression onto the Niño3.4 index, and “noise” as the standard deviation of the residual after removing ENSO signals. Maps of such ratios for D20 for SODA-POP and the coupled model are displayed in Figure 5. The ratios are generally much larger for the strong-ENSO periods. The results are therefore consistent with the multidecadal variation of the Indo-Pacific teleconnection depicted in 1-3–4, and provide an explanation as to why in weak-ENSO periods a transmission signal might not manifest itself above stochastic noises.

5. Conclusions

[14] Based on data since 1980, ENSO discharge-recharge signals are believed to transmit into the Indian Ocean arriving mainly at the WA coast via an equatorial pathway, and a subtropical NP pathway, The present study examines the robustness of energy leaving the Pacific via these pathways. Using SODA-POP, we find that in the pre-1980 period, little ENSO signal is transmitted to the Indian Ocean. The lack of transmission results from two interconnected factors: firstly, the NP pathway is not involved because of a narrower meridional extent of ENSO; secondly, the ENSO events are weaker leading to smaller transmission signals via the equatorial pathway that are drowned under stochastic noise. A multi-century coupled climate model experiment reproduces these features, confirming that these are not artefacts of the reanalysis system. The presence of these multidecadal fluctuations in our model without climate change forcing suggests that the stronger discharge in the post-1980 may not be green-house induced. The dynamics that drive the multidecadal fluctuations of ENSO properties need to be investigated, and this will be pursued in another paper.