Multidecadal fluctuations of winter rainfall over southwest Western Australia simulated in the CSIRO Mark 3 coupled model



[1] Previous studies have suggested that the observed winter rainfall reduction since the late 1960s over southwest Western Australia (SWWA) is consistent with what is expected from greenhouse forcing but the relative importance of potential causes is not conclusive. Here, we investigate the possibility of the rainfall reduction being a part of multidecadal variability using outputs of the CSIRO Mark 3 climate model. We find that multidecadal-long drying trends comparable to the observed exist in an experiment without climate change forcing. The model multidecadal-long rainfall decline manifests as a reduction in high-intensity rainfall events and is accompanied by an upward trend of the southern annular mode (SAM) with an increasing midlatitude mean sea level pressure (MSLP). Thus, multidecadal variability could primarily be responsible for the observed winter rainfall reduction, and could potentially superimpose on a greenhouse-induced drying trend to generate an even greater reduction than what has been observed thus far.

1. Introduction

[2] Since the late 1960s, SWWA has experienced a substantial drying trend with a winter rainfall decrease of some 20% [Indian Ocean Climate Initiative (IOCI), 2002]. The decline is manifest as a reduction in high-intensity rainfall events [Li et al., 2005], and is accompanied by an upward trend of the SAM [Thompson et al., 2000; Marshall, 2003; Marshall et al., 2004] with increasing MSLP in the midlatitudes. The cause of the upward trend of the SAM is a contentious issue. Observational [Thompson and Solomon, 2002] and other modelling studies [Sexton, 2001; Gillett and Thompson, 2003] indicate that it is attributable to ozone depletion over the past decades. Climate models produce an increasing midlatitude MSLP under increasing atmospheric CO2 incorporated in an upward trend of the SAM [Fyfe et al., 1999; Kushner et al., 2001; Cai et al., 2003a]; in association, model rainfall over SWWA decreases in the transient period while CO2 increases [Cai et al., 2003a; Karoly, 2003]. Although such a decrease in midlatitude rainfall is consistent with a reduction in extratropical southern hemisphere cyclones [Fyfe, 2003], the trend rate of the SWWA rainfall reduction in a suite of global climate models [IOCI, 2002] is not as large as that of the observed. This leaves open the possibility that the observed SWWA drying trend is partially an expression of multidecadal variability. Here, we assess this possibility using outputs of the CSIRO Mark 3 coupled model. We examine whether the observed winter rainfall decline is explainable by rainfall variations on multidecadal timescales.

2. Multidecadal Variability of Model Winter Rainfall Over SWWA

[3] The “Mark 3” model [Gordon et al., 2002] runs without flux adjustments and exhibits only a moderate initial climate drift. It simulates a reasonable seasonal rainfall cycle over SWWA with a realistic long term climatological winter rainfall total value of 310 mm compared to the observed of 315 mm (see Cai et al. [2003b] for further details). Here we focus on a control simulation using monthly and daily outputs covering a period of 100 years (years 11–110, excluding the initial years to avoid the drift). The monthly data are utilised to form seasonal anomalies, which are then used to examine multidecadal fluctuations of the SAM and rainfall over SWWA.

[4] Figures 1a and 1b display observed and modeled fluctuations of the winter season (averaged over June, July and August; JJA) rainfall in terms of percentages of the climatological winter mean. Superimposed is a 13-point running mean time series. The observed reduction over the past 30 years is clearly shown. In the model, multidecadal-long drying trends (years 30 to 60; years 75 to years 95) take place at a rate of 20–30% over a period of 20–30 years. Thus, in this model rainfall decreasing trends comparable to the observed can result from multidecadal variability.

Figure 1.

Time series of (a) observed and (b) modeled winter rainfall over SWWA in percentage of model winter (June, July and August) rainfall climatology. A 13-point running mean version is superimposed. (c) First mode from EOF analysis of monthly MSLP anomalies. It accounts for 61% of the total variance. The 13-point running mean version of the associated time series (dashed curve, rescaled and shifted) is plotted in Figure 1b.

[5] To explore the linkage of rainfall variability with MSLP, we applied empirical orthogonal function (EOF) analysis to winter MSLP anomalies in a domain 20°S southward. The first EOF (Figure 1c) is the SAM and accounts for about 61% of the total variance. The 13-year running mean version of its time series, hereafter referred to as the multidecadal SAM index (Figure 1b, dashed curve) shows that the SWWA rainfall fluctuations (grey curve) have corresponding variations in the SAM, a decrease in rainfall being associated with anomalously higher midlatitude MSLP.

[6] The circulation anomalies associated with SAM (Figure 2) indicate that, at a high-index phase, over the ocean the midlatitude jet shifts southward and over the land westerlies decrease (Figure 2a), and that at this phase weaker midlatitude westerlies are associated with a reduction in the SWWA winter rainfall. Figure 2b suggests that at the high-index phase sea surface temperature (SST) at midlatitudes (particularly over the SWWA latitudes) is generally higher whereas SST at high latitudes is lower. This is in part due to increased cloud cover associated with a lower MSLP [Cai and Watterson, 2002] at high latitudes, and vice versa at low latitudes, and in part due to an anomalous oceanic Ekman drift associated with the anomalously stronger high-latitude westerlies, as will be discussed later. Such an SST anomaly pattern implies, through the thermal wind balance [Stone and Fyfe, 2005], that the meridional temperature gradient has increased to support the southward shift of the midlatitude jet. In association, a southward migration of the position of maximum meridional temperature gradient occurs, which weakens the baroclinic instability to the north, reducing the synoptic disturbances and hence rainfall in the midlatitudes, consistent with findings in other climate models [e.g., Fyfe, 2003].

Figure 2.

(a) Latitudinal profile of zonal wind stress along 120°E for the wet (solid) and dry (dashed) period. (b) Maps of correlation coefficients of the SAM index with gridpoint SST anomalies.

3. Multidecadal Fluctuations of the Properties of Winter SWWA Daily Rainfall

[7] Li et al. [2005] demonstrated that the frequency of the gauged high-intensity SWWA rainfall in the winter season started to decrease at around 1965, a time coinciding with the onset of an upward trend of the SAM with increasing MSLP in the midlatitudes identified by Cai et al. [2003a]. To examine if this linkage exists on multidecadal timescales without climate change forcing, we employ daily data to contrast the statistical properties of the wet (years 28–32, years 71–75) and dry (years 58–62, years 93–97) periods (see the multidecadal SAM index of Figure 1c). Issues to address include whether the statistical properties of SWWA daily rainfall, daily MSLP, and their relationship in the wet and dry are different, and if so, in what fashion.

[8] Time series of daily rainfall and MSLP from a grid point at SWWA are obtained by connecting the daily value for JJA consecutively for the 10 dry and 10 wet years. Gaussian Kernel estimates of the probability density function (PDF) for the two 10-year periods are constructed (Figure 3a). These two PDFs are significantly different at the 1% level based on a Kolmogorov-Smirnoff (KS) test [Press et al., 1992]. Moreover, according to the Wilcoxon-Mann-Whitney rank sum test [Iman, 1994], means of the time series for the wet and dry periods are significantly different (with p-value ≈ 0) for both the rainfall and MSLP, with a higher mean rainfall value but a lower MSLP value shown shifting to the left in Figure 3a (right) in the wet periods, as expected.

Figure 3.

(a) Probability density function of daily SWWA rainfall and MSLP for the 10-year wet and 10-year dry period. (b) Return periods of high-intensity daily rainfall for the 10-year wet and dry period. Circles represent the empirical return period in dry period and pluses in wet period. The black and grey solid curves represent return periods based on the tail estimate, with the 95% confidence interval given as black dash line (dry) and grey dot-dash line (wet). It is evident that the return period in the dry period is significantly longer than that in wet period, implying a reduction of the high-intensity daily SWWA rainfall during the dry period.

[9] For both the wet and dry periods, we construct daily rainfall and MSLP anomalies from their respective 10-year daily climatological values. We find that the standard deviation and the spectral density of the daily rainfall and MSLP anomalies are virtually the same for the wet and dry period, as is the sensitivity of the daily rainfall anomalies to the MSLP anomalies. Thus the difference in the total value between the wet and dry period is due to the difference in the daily climatology between the two periods.

[10] Given that climatological rainfall values include the average of all synoptic events, the difference in the climatological daily value suggests that the property of tail rainfall events may be significantly different. To test this proposition, we apply the analysis detailed by Li et al. [2005], which involves choosing a threshold daily rainfall value, and examining the tail distribution over the threshold, which is 3.5 mm day−1. Return periods of high-intensity daily rainfall for the wet and dry periods are indeed significantly different (Figure 3b). In the wet period, high-intensity daily rainfall events occur more frequently and with a shorter returning period. In fact, this is evident in the rainfall PDF shown in Figure 3a (left) in the range of 3.5–10 mm day−1. Thus, as in the observed, the decrease in rainfall is mainly manifest as a reduction in the high-intensity rainfall event. There are actually more low-intensity rainfall events in the dry period, and in terms of extreme rainfall events, there is little detectable difference (Figure 3).

[11] To explore the linkage between the multidecadal SAM and the midlatitude MSLP annual cycle, we apply EOF analysis to the grid point MSLP annual cycle fields (365 days) referenced to an annual mean climatology field averaged over the 20 years (10 wet years and 10 dry years). The first EOF for both periods shares a pattern (pattern correlation of 0.97), similar to the SAM in Figure 1c. The associated time series reveal that the winter (days 150–240) midlatitude MSLP in the wet period is significantly lower. Thus, the multidecadal fluctuation of the SAM and the SWWA MSLP (Figure 3a (right)) are embedded in this SAM-like pattern of the annual cycle. When the multidecadal SAM is at a high-index phase, the daily winter midlatitude MSLP is anomalously higher, associated with a higher frequency of synoptic events in which MSLP does not go as low as in the low-index phase.

4. Oceanic Anomalies Associated With Variations of the SAM

[12] The oceanic overturning circulation anomalies associated with the SAM contribute to the maintenance of the meridional temperature gradient (Figure 4). Consistent with previous studies [Cai and Watterson, 2002; Hall and Visbeck, 2002], the climatological overturning circulation (Figure 4a) features the well-known cells including the Deacon cell. North and south of the Deacon cell, strong anomalies (Figure 4b) exist with a surface convergence centered at about 40°S, promoting the SST anomalies (Figure 2b). For example, south of the convergence, where cool SST anomalies develop (Figure 1b), the anomalous overturning circulation is advecting cold surface water northward. This reinforces the cold anomalies and hence contributes to the long-lasting SST anomalies and to the maintenance of the anomalous meridional temperature gradient, hence to the length of a dry period.

Figure 4.

(a) Model oceanic overturning circulation for winter season (in Sv) and (b) anomaly pattern associated with the SAM in terms of correlation coefficients.

[13] We have demonstrated that the multidecadal SAM is associated with SWWA rainfall. What causes SAM to fluctuate on multidecadal timescale? One explanation is that MSLP variability associated with the SAM is a direct consequence of the fluctuations in the number of the synoptic low-pressure/high-pressure events, that is, the SAM is really just a measure of the number/intensity of synoptic systems [Fyfe, 2003]. An alternative is that the multidecadal SAM component is generated independently by an oceanic mechanism, and the variation in the high-intensity rainfall events is a consequence of the modulation by it. An examination of the relative importance of these two processes is beyond the scope of the present study. What we have demonstrated is that the oceanic process (Figure 4) reinforces the temperature anomalies (Figure 2b), contributing to the longevity of the anomalies.

5. Conclusions and Discussion

[14] Using outputs of the CSIRO Mark 3 climate model we find that rainfall decreasing trends over SWWA comparable to the observed over the past decades exist as a part of multidecadal fluctuations. The model multidecadal-long drying trends share several observed features including that the reduction is manifest in the form of a smaller number of rainfall events exceeding a threshold value. The reduced frequency of high-intensity rainfall events leads to a lower rainfall climatology, and is accompanied by a high-index phase of the SAM with increasing midlatitude MSLPs. At this phase the meridional temperature gradient at the high latitudes increases, which maintains the associated wind anomalies through thermal wind balance and contributes to a southward shift of the maximum meridional temperature gradient from the SWWA latitude, reducing the baroclinic instability and hence the rainfall there. The associated anomalous oceanic overturning circulation contributes to the development and the longevity of the meridional temperature gradient anomalies, and hence to the multidecadal timescale. On the basis of the present analysis, we conclude that, while we are unable to determine the extent to which the observed SWWA winter rainfall decline is greenhouse driven, multidecadal variability appears to have the potential to generate such a reduction. We note however that there is also evidence that stratospheric ozone depletion could be affecting the SAM, and thus by inference SWWA precipitation. Further, in climate change experiments, greenhouse-induced SWWA drying trends intensify as the atmospheric CO2 increases. The implication is that there is the possibility of a greenhouse-induced drying trend conspiring with a variability-induced multidecadal drying trend to generate a more substantial rainfall decrease than what has been observed thus far. This possibility will be examined in a separate paper.


[15] This work is supported by the Indian Ocean Climate Initiative II (IOCI2) and the CSIRO Water for a Healthy Country Flagship. We thank members of the CSIRO Climate Model and Application Team for developing the model and performing the experiment, Mark Collier and Siobhan O'Farrell for converting the outputs to an easily accessible format, and Tim Cowan for reviewing the paper before submission.