Winter rainfall over southwest Western Australia (SWWA) has decreased by 20% since the late 1960s. Why has the reduction occurred in the Southern Hemisphere (SH) winter months but not in summer? To what extent is this reduction attributable to anthropogenic forcing and congruent with the Southern Annular Mode (SAM)? Using reanalysis data and the Intergovernmental Panel on Climate Change 4th Assessment Report (IPCC AR4) 20th century model experiments, we show that a SAM-SWWA relationship exists in winter and not in other seasons. An ensemble result from 71 experiments reveals that anthropogenic forcing contributes to about 50% of the observed rainfall decline. Approximately 70% of the observed trend is congruent with the SAM trend, whereas for the models it is 46%. Our result suggests that other forcing factors must be invoked to fully account for the observed rainfall reduction.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Since the late 1960s, SWWA (west of 118°E, south of 32°S) has experienced a substantial reduction with in winter (June–August, JJA) rainfall of some 20% (See Figure 1a) [Indian Ocean Climate Initiative Panel, 2002] translating to a 40% reduction of inflows to dams. Attribution of the causes for the rainfall reduction is rather difficult. Previous studies have highlighted influences that include a long-term mean sea-level pressure (MSLP) signal [Allan and Haylock, 1993], and decreases in the density of low-pressure systems and sea surface temperatures (SSTs) over the southern Indian Ocean [Smith et al., 2000] as probable causes for such a reduction. Cai et al.  presented modelling evidence linking the SWWA rainfall with the SAM, the dominant mode of the SH extratropical circulation, operating beyond the weather scale [Kidson, 1988; Karoly, 1990; Thompson et al., 2000; Thompson and Solomon, 2002; Hartmann and Lo, 1998]. Their result suggested that an upward trend of the SAM, consistently simulated by many models under increasing CO2 [Fyfe et al., 1999; Kushner et al., 2001], might have contributed to the observed rainfall decline. However, it is not clear why the observed rainfall decline manifests in the winter months when the observed SAM trend is weak [Marshall et al., 2004], whereas there is no trend in summer (December–February, DJF) rainfall (Figure 1b) when the SAM trend is strongest. This is a focus of the present study.
 Further, the role of the SAM in the rainfall reduction remains inconclusive. Based on a reanalysis data set, Meneghini et al.  suggested that although the SAM contributes to inter-annual winter rainfall variability, it might not contribute to the long-term decline, in contrast to what was suggested in the modeling study of Cai et al. . On the other hand, Timbal et al.  analysed simulations of the Parallel Climate Model in conjunction with a statistical downscaling technique, and concluded that the anthropogenic forcing-induced trend in the large-scale circulation probably contributed to the observed rainfall decrease, however, ambiguities remain. Other studies have implicated land-cover change [Pitman et al., 2004] or natural multidecadal fluctuations [Cai et al., 2005]. Until now, a quantitative assessment was not possible because of the small number of models and model experiments available and the fact that there is only one realization of the real world, such that the influence from multidecadal variability cannot be separated from an anthropogenic forcing-induced trend. The unprecedented number of simulations from the late 19th century through to the end of the 20th century, conducted by climate modeling groups worldwide as part of the IPCC AR4, provides an opportunity to overcome this difficulty. This is another focus of the present study.
2. Model and Experiment Details
 Data for the 20th century experiment from 21 IPCC AR4 coupled general circulation models comprising a total of 71 simulations were made available through the U.S Department of Energy's Program for Climate Model Diagnosis and Intercomparison (PCMDI). Model names and details are listed in Table 1 of Cai and Cowan  with references to further documentation. All the models contain greenhouse gas and direct sulfate aerosol forcing, whilst over a half include a stratospheric ozone forcing component. To evaluate the performance of the models and their ensemble, observational data from the National Centers for Environmental Prediction (NCEP) reanalysis is used [Kalnay et al., 1996]. It has been shown that the raw NCEP MSLP produces a SAM trend that is too large [Marshall, 2003]. To correct this over-estimate in the trend we perform a regression of the grid-point NCEP fields onto an observed non-normalized station-based index (between 40°S and 65°S) of the SAM [Marshall, 2003]. We choose the non-normalized option as we are interested in the real value of the trend rather than the spatial coherence. As such, the remainder of the paper we will refer to NCEP as being corrected to station based observations. Observed rainfall data set from the Australian Bureau of Meteorology is used together with the corrected SAM trend to determine the observed relationship between the SAM and SWWA rainfall in terms of the seasonality of circulation variability.
 Empirical orthogonal function (EOF) spatial patterns and time amplitude functions are calculated in the domain of 20°S–70°S to assess the trend of the SAM. We use a covariance EOF, because a correlation EOF emphasizes on coherence, and has the effect of suppressing variance where it is high but enhancing it where variance is low, and does not serve our purpose of calculating the actual trend. The covariance EOF is arranged such that the variance of the spatial pattern of an EOF sums to unity, leaving the trend and variance of an EOF to be recorded in the time series. The time trend of an EOF can therefore be calculated from the corresponding time series. Before application of EOF analysis the climatological mean, which is the average over the period 1950–1999, is removed.
3. Seasonal Cycle and the Seasonality of the SAM
 The dynamics of the seasonal SWWA climate cycle is important for understanding the seasonality of the SAM impact. SWWA is a winter rain region; the 1950–1999 climatological winter total rainfall is 335mm, whereas the summer rainfall is just 50mm. This seasonal cycle is determined by the movement of the subtropical high-pressure ridge. Figures 1c and 1d show a latitude-month plot at 120°E of monthly climatological value and standard deviation of MSLP (month 1 corresponding to January). The semi-annual cycle is evident south of 55°S, but not at the mid-latitudes. As the mid-latitude high-pressure belt moves northwards from autumn (March–May, MAM) toward the winter months, and the center of high pressure approaches the SWWA latitudes (32°S–35°S) (straight lines, Figures 1c and 1d), the frequency of synoptical events impinging on this region increases. This is indicated by the stronger variance around the winter months in Figure 1d. This is why most of the rainfall is recorded in the winter season. From spring (September–November, SON) to summer the seasonal movement reverses so that in summer the subtropical ridge is furthest to the south. Thus, there is a sharp seasonal contrast between summer and winter in terms of the mean circulation.
 This basic seasonal cycle is well simulated by the IPCC AR4 models. The intra-seasonal contrast averaged over the 71 experiments is not as strong as in the NCEP reanalysis, giving an all-experiment average of summer rainfall of 49mm, which compares well with the observed, but a winter rainfall average that is too low, at 227mm. Because of this difference, we will compare model and observed rainfall in terms of the percentage of the climatological value.
 In association with the seasonal cycle, the structure of the SAM in each season in terms of MSLP EOF1 is rather different. In winter, the center of maximum mid-latitude positive anomalies is located furthest to the north, closest to the SWWA latitudes (Figure 2a). The anomalies in winter appear to be more concentrated in the Australia-New Zealand longitude band, enabling some influence on rainfall over SWWA. In summer, the center of positive SAM anomalies is furthest to the south (Figure 2b), and the EOF1 incorporates patterns of higher wave number anomalies. Most importantly, there is little anomaly over the SWWA region. In autumn and spring, the center is located at latitudes between the winter and summer seasons (not shown).
 These seasonal contrasts in the SAM characteristic are produced by each of the 21 IPCC AR4 models used in our study. As with the observed, EOF analysis on SLP anomalies (from the mean over the period of 1950–1999) is conducted for summer and winter for each model experiment. Although significant differences exist from one model to another [Miller et al., 2006; Cai and Cowan, 2006], all 21 models produce the feature that maximum mid-latitude anomalies are located closest to SWWA in the winter season. Figures 2c and 2d show the average of all the IPCC model EOF1 patterns for winter and summer. Together with those from every individual experiment (not shown), they illustrate strong inter-model consistency, and further demonstrate the distinctive seasonal contrast in the spatial structure, i.e., the maximum mid-latitude anomalies reside closest to SWWA latitudes in winter.
4. Seasonality of the SAM-SWWA Rainfall Relationship
 As a result of the strong seasonality of the SAM characteristics, the SAM-SWWA rainfall relationship differs significantly from one season to another. Correlation analysis between the detrended time series of the SAM and the SWWA rainfall (in terms of percentage of climatology) shows that there is a relationship in winter, significant at the 95% confidence level with a correlation of −0.30 and a slope of −0.12% with a standard error of 0.05% (unit of EOF1 time series)−1. For the other seasons, the correlation is not significant. During winter, an anomalously high phase of the SAM is associated with a reduction in SWWA rainfall (Figure 3a), whereas in summer there is a weak insignificant positive correlation (Figure 3b).
 The seasonality in the SAM-SWWA rainfall relationship is well simulated by the IPCC AR4 models. Figures 3c and 3d provide similar plots for the IPCC models using all pairs of the EOF1 and rainfall anomaly time series. The result reinforces the seasonal difference in the relationship between the SAM and SWWA rainfall, i.e., the SAM affects SWWA rainfall in winter (Figure 3c) but not in summer (Figure 3d). For the winter season, the correlation coefficient is −0.35 and the slope is −0.16% with a standard error of 0.0009% (unit of EOF1 time series)−1. The lack of a SAM-SWWA rainfall linkage in summer means that a SAM trend exerts little influence on the summer SWWA rainfall trend. On the other hand, the linkage in winter infers that a winter SAM trend affects the SWWA winter rainfall trend. This is in part why rainfall decreases in winter, but not in summer.
5. Anthropogenic Forcing and Congruency With the SAM Trend
 To address the contribution of anthropogenic forcing to the winter rainfall reduction and the extent to which it is congruent with the SAM trend, we average all EOF1 time series of winter MSLP and calculate the linear trend. Similarly the winter SAM trend from NCEP is also calculated, which is not statistically significant as already shown by previous studies [Marshall et al., 2004]. To assess the significance of the model SAM trends, we use a multicentury control run from the CSIRO Mk3.0 and GFDL CM2.0 models to estimate the range of the 50-year trends due to natural variability of the coupled climate system. Trend values over 50-year non-overlapping periods from each of the two models are obtained using from the EOF1 time series. Assuming that every model has a similar amplitude in variations of 50-year trends from natural variability, and therefore the same value of standard deviation, the 95% confidence interval is estimated following the approach of Marshall et al. . Similarly, the 95% confidence interval is obtained for SWWA rainfall but in terms of the percentage of climatology.
 The winter SAM trend in the IPCC AR4 models is shown in Figure 4a. Based on a small number of model runs, Marshall et al.  and Arblaster and Meehl  found that in their model, the winter trends are not greater than what might be induced by internal variability. The implication from Figure 4a is that there is a positive trend in the winter SAM that is more than what you would expect from internal variability. Among the 21 models used, a total of 48 out of 71 experiments include an ozone-depletion component. We have similarly calculated the winter SAM trend in each group and find that the trend in each is significant and virtually the same as the trend from the 71 experiments. Thus, this significant trend, averaged over the 71 experiments, is not driven by ozone-depletion.
Figure 4b compares the total winter rainfall trend over 1950–1999 in terms of percentage of climatology for the observed (10%) and the ensemble average (5.2%) (crosses, Figure 4b). Both are statistically significant. It is immediately clear that the model trend is about half of the observed. If we assume that by averaging the large number of experiments the influence from internal variability is by and large removed such that the residual trend is driven solely by anthropogenic forcing, then this forcing accounts for only about half of the observed trend.
 How much of the trend is congruent with the winter SAM trend? This is computed from the SAM-SWWA rainfall relationship (Figures 3a and 3c) by multiplying the regression coefficients (−0.12% and −0.16% (unit of EOF1 time series)−1) with the respective winter SAM trends in Figure 4a (55 units per 50 years for the observed and 14 units per 50 years for the model). This gives 6.7% and 2.4% (circles, Figure 4b) for the observed and modeled, respectively. Thus, 67% (6.7% over 10%) of the observed and 46% (2.4% over 5.2%) of the modeled rainfall reduction is congruent with the SAM trend. Ansell et al.  suggested a link of SWWA rainfall with Indo-Pacific Ocean sea surface temperature (SST), although not as strong as with MSLP. Thus, other factors, such as changing SST gradients could provide the forcing for the rest of the rainfall trend.
 It has been shown that SAM variations on a 50-year timescale have the potential to drive a SWWA rainfall reduction comparable to the observed magnitude [Cai et al., 2005]. The higher level of congruency in the observed perhaps reflects a greater influence by multidecadal variability as the observed has only one realization.
 Since the late 1960s, there has been a substantial reduction in rainfall in the SWWA region. It is not clear why the reduction occurs in the winter months, when the observed SAM trend is weak, but not in the summer months, when the observed SAM trend is strongest. It is also not clear to what extent the reduction is attributable to anthropogenic forcing and is congruent with the SAM. Using IPCC AR4 20th century model experiments and available observations, we show that in winter the mid-latitude center-of-action of the SAM is closest to SWWA latitudes, compared to other seasons. As a result, there is a statistically significant relationship between the SAM and SWWA rainfall in winter, but not in other seasons. The observed winter SAM trend, though not statistically significant, accounts for two thirds of the observed winter rainfall reduction. The ensemble result from 71 experiments reveals that the contribution by anthropogenic forcing accounts for about 50% of the observed rainfall decline; of that, about 50% is congruent with the SAM trend. Ozone-depletion plays no part in the observed reduction.
 Our results suggest that other forcing factors, e.g. multidecadal variability, land use changes, must be invoked to fully account for the observed winter rainfall reduction. Determination of their relative importance is essential for the projection of future water availability in the region. One of the most consistent results from climate models is that as CO2 continues to increase, SWWA rainfall will continue to decrease [Cai et al., 2003]. If multidecadal variability plays a significant part in the observed rainfall reduction, it means that there will be a period in which CO2-induced reduction will be temporarily mitigated by the opposite phase of variability. On the other hand, it also means that CO2 and multidecadal variability could conspire to produce an even greater rate of rainfall reduction in the future.
 This work is part of the CSIRO Water for a Healthy Country flagship and is supported by the IOCI. We thank Ian Smith for his helpful and insightful comments, and the anonymous external reviewers. In addition, we acknowledge the work undertaken by numerous international modelling groups who provided their model experiments for analysis. In particular we recognise the significant work that the PCMDI has achieved in collecting and storing the model data. For more details on model data or documentation, readers are referred to the PCMDI Web site (http://www-pcmdi.llnl.gov/). We also acknowledge the Bureau of Meteorology for the use of their observational rainfall data set.