Geophysical Research Letters

Recent unprecedented skewness towards positive Indian Ocean Dipole occurrences and its impact on Australian rainfall



[1] Is the recent high frequency of positive Indian Ocean Dipole (pIOD) events a consequence of global warming? Using available observations and reanalyses, we show that the pIOD occurrences increase from about four per 30 years early in the 20th century to about 10 over the last 30 years; by contrast, the number of negative Indian Ocean Dipole (nIOD) events decreases from about 10 to two over the same periods, respectively. A skewness measure, defined as the difference in occurrences of pIODs and nIODs, illustrates a systematic trend in this parameter commencing early in the 20th century. After 1950, there are more pIODs than nIODs, with consistent mean circulation changes in the pIOD-prevalent seasons. Over southeastern Australia (SEA), these changes potentially account for much of the observed austral winter and spring rainfall reduction since 1950. These features are consistent with projected future climate change and hence with what is expected from global warming.

1. Introduction

[2] Climate models project a future warming pattern in the tropical Indian Ocean (IO) that features a slower warming rate in the eastern IO than in the western IO [Vecchi and Soden, 2007]. In terms of zonal sea surface temperature (SST) gradients such a warming pattern is reminiscent of that during a pIOD event, a phase of variability in which the eastern IO is cooler, and the western IO is warmer, than normal [Saji et al., 1999]. Given that global warming has been occurring, is there a signature in the mean-circulation that supports such a projection? Are there any consistent changes in the IOD properties?

[3] In recent years, there has been an apparent increase in the occurrence of pIOD events. During 2006–2008, the IO experienced an unprecedented occurrence of three consecutive pIOD episodes [Cai et al., 2009]. These recent events appear to support the notion of multi-decadal fluctuations of IOD properties [Saji and Yamagata, 2003a, 2003b; Ashok et al., 2004], and the more recent finding that since 1880 the occurrences of pIOD and nIOD events displayed three distinct regimes [Ihara et al., 2008]: frequent nIODs associated with La Niña events during 1880–1919; weak pIOD events relatively independent of El Niño events during 1920–1949; and strong and frequent occurrences of pIODs associated with El Niño events during 1960–2004. Contrary to this multidecadal-fluctuation interpretation, Abram et al. [2008] suggests that the pIOD intensification may be secular. Here, we introduce a skewness measure, defined as the difference in occurrences of pIOD and nIOD events, to illustrate a systematic trend that commenced early in the 20th century.

[4] Since 1950, rainfall over SEA has decreased significantly. The reduction is greatest in autumn (March, April, May) in terms of either the absolute value or the percentage of climatology. Cai and Cowan [2008] show that a reduction in La Niña events and changing SST gradients in the subtropical IO contributes to the late autumn rainfall decline. As for the winter (June, July, August (JJA)) and spring (September, October, November (SON)) seasons, one would expect the 2006–2008 three-consecutive pIOD events to have contributed to the recent SEA drought. However, Ummenhofer et al. [2009] link the recent drought and many of the large historical droughts with a conspicuous absence of nIOD events. In contrast, Nicholls [2009] suggests that over 70% of the reduction during March–August is attributable to the Southern Annular Mode (SAM) trend. The vastly different interpretations call for further investigations. We determine the extent to which the long-term winter and spring rainfall changes over SEA are attributable to long-term changes in the IOD properties.

2. Reanalysis and Observations

[5] We focus on the JJA and SON seasons, when pIOD events develop and peak respectively. An updated version of the Global Sea Ice and SST reanalysis [Rayner et al., 2003] covering the period from 1880–2007 is used to construct an index of the IOD and mean SST trend patterns, although before 1958 data quality is poor due to a lack of observations [Saji and Yamagata, 2003a]. Reanalyses from the National Center for Environmental Prediction (NCEP) [Kalnay et al., 1996] and the Simple Ocean Data Assimilation with the Parallel Ocean Program (SODA-POP) [Carton and Giese, 2008] are used to examine trends of the ocean and atmospheric circulation fields, like wind stress and the thermocline (20°C isotherm depth, D20). Observed Australian rainfall data since 1950, subjected to extensive quality control from the Australian Bureau of Meteorology Research Centre, are used to examine the IOD-rainfall relationship as well as long-term rainfall changes.

3. Changes in IOD Properties

[6] To investigate a possible IOD change, a time series of an IOD index is constructed following the definition of Saji et al. [1999] using SST anomalies referenced to the climatological mean over the entire period consistent with Ihara et al. [2008]. The indices for JJA (Figure 1a) and SON (Figure 1d) both trend upward. An important feature is that the index spends more time in the positive phase in the post-1950s period than prior to this time. Previous studies using data since 1950 have found an asymmetry between pIODs and nIODs, with a greater amplitude for pIODs than for nIODs [Hong et al., 2008]. This amplitude asymmetry is generated in the Sumatra-Java upwelling zone, where negative SST anomalies often reach a far larger value than the positive SST anomalies. An expression of this IOD amplitude asymmetry is a higher pIOD intensity than that of nIODs. This is evident in Figures 1a and 1d.

Figure 1.

Time series of (a) the JJA IOD index (°C) as defined by Saji et al. [1999] (a linear trend line since 1950 is superimposed; the two thin horizontal lines represent 0.75 of the standard deviation), (b) number of pIODs (black) and nIODs (red) in a sliding 30-year period (recorded at the 16th year), and (c) skewness defined as the difference (number of pIODs minus nIODs). (d–f) Same as Figures 1a–1c but for SON.

[7] To investigate whether there is a systematic increase in the pIOD frequency, we define an IOD event as when the index exceeds 0.75 of its long-term standard deviation. Following such a definition, which is similar to that of Ihara et al. [2008] and captures virtually all events they identified, we would count 2002, 2004, 2006, 2007 as pIOD years in the period since 2002. When including 2008, five pIODs occur during 2002–2008. We then proceed to count pIOD and nIOD events in 30-year periods using a sliding-window and record the number at the 16th year of the window (Figures 1b and 1e). The different seasonal count suggests that not all pIODs commencing in JJA persist into SON, and although it is not clear what causes the rise in nIODs around the 1940s, two features stand out. Firstly, there is an increasing frequency of pIODs after the 1950s, consistent with previous studies [Ihara et al., 2008; Abram et al., 2008]. At the start of the 20th century, the number of pIODs in a 30-year period is about four, but during the past 30 years, this number has reached approximately 10 events. Simultaneously, there is a reduction in the number of nIODs, reducing from about 12 at the start of 20th century to two over the past 30 years. Secondly, in the last 30-years, the number of pIODs in JJA and SON are at a record high, whereas the number of nIODs in SON is at a record low (Figure 1e).

[8] Thus, there is apparent skewness toward pIODs. A measure of this skewness is defined as the difference in the number of pIODs and nIODs over the 30-year periods (Figures 1c and 1f). Since the 1960s, the skewness has been rising, surging to an unprecedented level in recent decades. Further, the upward trend appears to commence far earlier (the beginning of the 20th century), strengthening the argument that the change in the IOD properties is systematic.

4. Mean Circulation Changes

[9] Using all monthly SST data, Vecchi and Soden [2007] were unable to find consistent IOD-like warming in various SST products. Because of the lack of oceanic data before the 1950s, we concentrate on the post-1950 period in JJA and SON. The mean SST trends display a slower warming rate in the eastern IO than in the west (Figures 2a and 2d). The attendant changes in zonal wind stress feature easterlies over the eastern IO (Figures 2b and 2e). In association, the thermocline shallows but deepens in the western and the southern tropical IO (Figures 2c and 2f). These trends resemble the anomaly patterns associated with a pIOD event, in which a positive feedback process operates involving anomalous SST, winds, and depth of the thermocline in the east during JJA and SON. This occurs when the mean thermocline in the east is shallow and seasonal upwelling develops along the Sumatra-Java coast [Saji et al., 1999]. The resemblance raises the issue as to whether these mean circulation changes are a manifestation of the changing IOD properties, or vice-versa. We believe that this is not important; what is important is that there are consistent signals in the mean circulation and in the IOD properties.

Figure 2.

JJA trends of (a) SSTs since 1950 using HadISST (°C per 58 years), (b) zonal wind stress (TAUX) since 1950 using NCEP data (N m−2 per 58 years), and (c) thermocline (D20) using SODA-POP over the period 1958–2001 (m per 44 years). (d–f) Same as Figures 2a–2c, but for SON.

[10] A consensus regarding IOD mechanisms is that besides El Niño, other drivers can also induce pIOD events [Shinoda et al., 2004], including the high phase of the SAM [Lau and Nath, 2004] and the onset of monsoon [Fischer et al., 2005]. The increasing number of pIODs are consistent with an upward trend of the SAM [Lau and Nath, 2004], and with more frequent and protracted El Niños associated with a weakening Walker Circulation [Vecchi and Soden, 2007; Ihara et al., 2008], although in recent decades pIOD occurrences appear to be more independent from El Niño [Abram et al., 2008]. Climate models project a further upward trend of the SAM and weakening of the Walker Circulation as global warming continues. Following such a scenario, the pIOD frequency would continue to increase. On the other hand, a pIOD may modify the El Niño-Southern Oscillation amplitude/evolution by inducing wind anomalies associated with the Walker Circulation [Saji and Yamagata, 2003a, 2003b; Li et al., 2003; Kug and Kang, 2006]. Following this scenario, the weaker Walker Circulation may partially result from a stronger westerly wind in the western Pacific in response to the stronger and more frequent pIOD forcing. Our approach here is not to partition the contribution from individual drivers, but to address whether the increasing pIOD frequency is consistent with what is expected from climate change and to examine the impact on Australian rainfall.

5. Impacts of Changing IOD Properties on Australian Rainfall

[11] A higher pIOD frequency means that pIOD-induced droughts will occur more often in affected regions, such as SEA [Ashok et al., 2003; Cai et al., 2005; Meyers et al., 2007]. Over this region, JJA and SON rainfall accounts for the majority of the annual total rainfall. In six out of the past seven years in (2002–2008), SON rainfall has been anomalously low, while since 1950 rainfall has decreased substantially. Is the rainfall reduction congruent with the changing IOD properties?

[12] Nicholls [2009] proposed a series of useful principles for attributing long-term rainfall changes to climate drivers. One such principle is that a climate driver index containing a local rainfall signal must not be used; for example, a mean sea level pressure (MSLP) based index that has an imprint of the rainfall signal. Because local MSLP variability is often a surrogate for rainfall variations, attributing a rainfall trend to a local rising MSLP would constitute a cyclic argument. Following this principle, the eastern pole of the IOD is deemed to be too close to Australia for the IOD index to be used. However, in this case, the principle is probably too strict, because there is no suggestion that the eastern pole's SST variability is in any way influenced by rainfall signals over SEA. Figure 3, constructed using detrended IOD indices and circulation anomalies, suggests that MSLP anomalies associated with the IOD originate from the eastern pole and propagate to SEA, but not the other way around. The anomalies are relatively weak initially during JJA with correspondingly weak rainfall correlations over Australia (Figures 3a and 3b), but strengthen in the following season (Figures 3c and 3d) to influence SEA rainfall, through a wave train, with clear alternating positive and negative anomalies extending from the eastern IO via SEA into the South Pacific [Saji and Yamagata, 2003b]. Thus, we believe that it is appropriate to use the IOD index for attributing SEA rainfall changes.

Figure 3.

Correlation of a detrended IOD index with (a and c) detrended MSLP and (b and d) Australia rainfall anomalies for JJA and SON, over 1950–2007. Areas confined by dashed white contours in Figures 3a and 3c, and shown in color in Figures 3b and 3d indicate statistical significance at the 95% confidence level.

[13] To this end, we calculate the total trend in the IOD indices since 1950 (red lines, Figures 1a and 1d). We then calculate the sensitivity of rainfall to these indices by regressing detrended anomalies onto the detrended indices. The detrending process ensures that the sensitivities are not due to trends in rainfall and the indices. Multiplying the trend in the indices with the regression coefficients yields rainfall changes congruent with the increasing IOD indices (Figures 4b and 4d). The JJA correlation is weak over SEA, similar to that shown over 1979–1997 [e.g., Ashok et al., 2003]. Because of the overall weak correlation, the IOD-congruent rainfall change in this season is smaller than that in SON. The IOD trend-congruent rainfall reduction in SON is comparable to the observed decrease over much of SEA (Figures 4c and 4d). However, the observed SON rainfall trends since 1950 are not statistically significant using a t-test, and are sensitive to decadal variability. Further, the IOD impact on rainfall trend is quite sensitive to the choice of the commencement decade for the IOD trend calculation. Nonetheless, choosing a later starting decade for the post-1950 IOD trend calculation (e.g. 1960, 1970 and so on) always produces an upward trend in the IOD index, the hence an IOD-congruent winter and spring rainfall reduction (figure not shown).

Figure 4.

(a) Total rainfall trend over Australia (mm) and (b) rainfall changes (mm) congruent with the trend in the IOD index using data over 1950–2007 for JJA. (c and d) Same as Figures 4a and 4b but for SON. Areas shown in color in Figures 4b and 4d are the statistically significant regions from Figures 3b and 3d.

6. Conclusions

[14] We show that since 1950 consistent mean circulation changes have emerged in the tropical IO, trending towards a future warming pattern as projected by climate models, with a slower warming rate in the eastern IO than in the western IO. These changes are accompanied by a steadily increased pIOD frequency, from less than four events per 30 years early in the 20th century to 10 occurrences over the last 30 years, whereas the number of nIOD decreases from about 10 to two per 30 years. Further, there is a trend in skewness toward more frequent pIOD occurrences, which commences early in the 20th century and reaches an unprecedented level in the past 30 years. These changes have significant impacts in regions, such as SEA. There the IOD trend-induced rainfall reduction potentially accounts for much of the observed austral winter and spring rainfall decrease since 1950, although the observed rainfall trends are not statistically significant. Our results suggest that although we can not attribute the trigger of the recent pIODs to climate change, the systematic increase in pIOD occurrences with an associated rainfall reduction over the SEA region is consistent with what is expected from global warming. Because the drivers of these changes are projected to continue, the likely scenario of a further increase in pIOD frequency leading to more frequent Australian droughts should be considered.


[15] This work is supported by the Department of Climate Change, CSIRO Wealth from Ocean Flagship, and the South East Australia Climate Initiative.