A pattern of recurring below average rainfall has affected southern areas of Australia since the late 1990s. In southeastern Australia (SEA; indicated in Figure 7) rainfall for the period 1997 to 2009 has been 11.4% below the long-term average, making it the driest 13-year period on record by a large margin (previous record low is–7.8% for the 13-year period 1933–1945). Both the duration and intensity of the 1997–2009 rainfall deficit, hereafter referred to as the recent drought event, is without historical precedent in the instrumental record starting from 1900. SEA is defined as continental Australia south of 33.5°S and east of 135.5°E. Periods of rainfall deficiency of similar magnitude were recorded in the past at the time of the formation of the Australian Federation (a decade centred around 1902) and during an extended period encompassing World War II from 1935 to 1945 (hereafter the WWII drought) when the rainfall deficit over these 11 years was–10.6%, nearly as severe a decline as the recent drought but of a shorter duration.
The 1997–2009 drought in SEA is happening during the cooler part of the annual cycle: affecting a continuum of months from March to October with a dominance of March–May (autumn) (Timbal, 2009). This pattern of dry conditions peaking in autumn is reminiscent of a similar decline which occurred in the southwest of western Australia (SWWA) in the 1960s. Rainfall in the two regions is linked on daily to interannual timescales with both affected by the position of the rain-bearing westerly storms (Hope et al., 2009). However, the rainfall deficit in these two regions has been observed to commence at different times, in the 1960s in SWWA and in the 1990s in SEA.
SWWA and SEA are unique in Australia in having cool season rainfall peaks associated with the northward movement of the subtropical ridge (STR). With both regions having severe and sustained rainfall deficits from 1997 to 2009, this raises the possibility that the observed mean sea level pressure (MSLP) increase (Timbal and Hope, 2008) over southern Australia is driving the recent severe downturn in rainfall (Nicholls, 2009). Both regions at this time of the year are located poleward of the STR, a belt of high MSLP corresponding to the descending branch of the Hadley circulation (Peixoto and Oort, 1992) (Figure 1, upper panel). In both regions, rainfall variability is related closely to local variations of MSLP (Allan and Haylock, 1993; Timbal and Murphy, 2007). It is well known that at the latitude of the STR, MSLP is increasing (Solomon et al., 2007). The difference in MSLP between the last 30 years (1980–2009) and the previous 130 years (1850–1979) shows a large band of MSLP increase between 30° and 50°S across the Southern Hemisphere (Figure 2). Details on the MSLP data used are provided in the next section. Arguably, a similar feature can be observed in the Northern Hemisphere but appears confined to the Atlantic and Pacific oceans and is not as annular as in the Southern Hemisphere (Thompson and Solomon, 2002). In the Southern Hemisphere, the MSLP increase is stronger and tends to extend across the continents towards the Tropics, including Australia, as noted by Nicholls (2009).
In the Australian sector of the Southern Hemisphere, the STR has been monitored ever since the existence of the Australian Bureau of Meteorology (BoM) and the observations of MSLP at various locations around Australia. It was often referred to as the L-index (for latitude of the STR) and in general the focus has been on the position of the STR more than its intensity (Kidson, 1925; Deacon, 1953; Das, 1956; Pittock, 1971, 1973 and 1975; Thresher, 2002, 2003; Drosdowsky, 2005; Williams and Stone, 2009). Based on Drosdowsky (2005), the mean position of the STR can vary between 29°S in winter and 40°S in summer (Figure 1, lower panel) and it has long been known that changes in its position are in part responsible for the interannual variability of cool season rainfall in SEA (Pittock, 1975; Drosdowsky, 2005; Williams and Stone, 2009). For this reason, long-term trends of the position of the STR have been studied extensively. In response to previous studies affected by data homogeneities and different definitions of the STR, Drosdowsky (2005) produced a homogenised series starting from 1890 from which no obvious trend in the STR-Position (STR-P) could be inferred.
In this study, the Drosdowsky (2005) STR series is updated to 2009. The data are first used to investigate the relationship between Australian rainfall and the position and intensity of the STR, including the annual cycle of the relationships as well as their stability over time. Secondly, the evolution of the STR during the 20th century is discussed. Third, an attempt is made to relate the anomalies of STR and the rainfall deficiencies in SEA during the two lowest rainfall periods of the historical record (the WWII drought and for the 1997–2009 period). Finally, the results are discussed further in light of an apparent connection between the intensity of the STR and the global warming of the planet.
2. Data sets
The latest 0.05° (approximately 5 km) gridded rainfall analyses from the BoM (Jones et al., 2009) for the 1900–2009 period are used in this study. The data set includes precipitation, maximum and minimum temperature, and vapour pressure surfaces obtained by interpolating surface stations measurement onto a regular 0.05° grid. A monthly SEA average for rainfall was calculated by subsampling every 1° the 0.05° over continental Australia within a box (display in Figure 7) between coordinates 33.5°S and 135.5°E and coordinates 152.5°E and 39.5°S. This series can be constructed online at http://www.bom.gov.au/cgi-bin/silo/cli_var/area_timeseries.pl.
We use the Drosdowsky (2005) index which measures the position and intensity of the local maxima in monthly surface pressure along zonal profiles for a 5° longitude band around 150°E between 10° and 44°S. This index is simple to calculate, robust and consistent over time (Drosdowsky, 2005). The use of stations also means that it can be calculated as far back as 1890 and with an update to 2009, the index provides 120 years of continuous data.
Two global data sets were also used to provide a global perspective on MSLP and temperature changes. For MSLP, we used the data set from the Hadley Centre of the UK Meteorological Office (HadSLP2) from 1850 to 2009 on a 5° × 5° grid (Allan and Ansell, 2006). For temperature, we used the global average surface temperature (HadCRUT3) prepared jointly by the Climatic Research Unit, University of East Anglia, and the Hadley Centre of the UK Meteorological Office (Brohan et al., 2006).
3. Observed relationship between the STR and Australian rainfall
The annual cycle of the relationship between the STR changes (intensity and position) and rainfall is analysed by computing the correlation coefficients between STR-Intensity (STR-I) or STR-P and the gridded rainfall. The correlation coefficients are calculated using linearly detrended data for 3-month sliding averages from January–February–March (JFM) to December–January–February (DJF) and plotted across the entire Australian continent (Figures 3 and 4). Areas with significant correlations are shaded (blue when the relationship is positive with rainfall, red otherwise) and different shades of colour are used between 0.2 and 0.6 (in 0.1 increments). Positive values for the STR-P indicate a southward shift of the STR.
The negative correlation between the STR (both I and P) and rainfall across SEA has a marked annual cycle. The earliest sign appears in March with the STR-I across southern Victoria, is well established across SEA in autumn (March–April–May, MAM) and peaks in winter (June–July–August, JJA) before receding across spring. During the coldest part of the year, the relationship extends further away from SEA across most of southern and inland Australia, although correlations in excess of–0.3 are mostly limited to SEA, South Australia and the SWWA.
All year around, across southern Australia, the negative relationship with rainfall is always stronger with the intensity than with the position. Areas of significant correlations extend further inland and the largest correlations are reached in SEA. On the contrary, the positive relationship between STR movements and rainfall, visible across eastern Australia in summer [from November–December–January (NDJ) to February–March–April (FMA)], is comparable between STR-I and STR-P.
High resolution gridded rainfall observations provide interesting details in particular related to orographic features such as the Great Dividing Range (GDR) which runs down the eastern side of the Australian continent. The largest negative correlations which are noted along the coastal strip from South Australia to Victoria extend to the GDR across eastern Victoria. On the other side of the range, along the eastern coast, the negative relationship during the autumn–winter–spring period is much weaker than on and west of the range. On the contrary, further north along the coast, the eastern Seaboard has a significant positive relationship (i.e. when the ridge is positioned further south or is stronger, rainfall is increased) which can also be diagnosed using a meridional latitude index (Rakich et al., 2008). The correlations are much higher with the STR-P than with the STR-I, and present all year.
Further south in Tasmania, the negative relationship between the STR-I and rainfall is significant all year implying that a stronger or more southerly STR is associated with suppressed rainfall. It is worth noting that it does appear to peak during the coldest time of the year as in SEA but is equally large in early autumn (FMA) or late spring [October–November–December (OND) and NDJ].
The relationships between rainfall and STR-I across SEA, coastal SA, SWWA and Tasmania are stable (i.e. the rise and fall of the correlation coefficients from one 3-month period to the next are smooth) over an extended part of the annual cycle and locally reach large and very significant values. In contrast, across the tropical half of the Australian continent, correlations are rarely in excess of ± 0.3, spatially patchy and lack stability from one 3-month period to the next.
The strength of the relationships between the STR (I and P) and SEA average rainfall as well as between STR-P and STR-I are summarised in Table I for each individual month (i.e. not using a 3-month sliding mean). Negative correlations between STR series and rainfall, significant at the 95% level using a two-sided Student's t-test (indicated by bold figures), are present from April to November for STR-I and from May to September for STR-P. In addition, the two STR series are significantly correlated from April to December.
Table 1. Monthly correlation coefficients between STR intensity (STR-I) and position (STR-P) and SEA rainfall. Correlations with rainfall are calculated using detrended monthly means from 1900 to 2009, and from 1890 between the two STR series. Statistically significant (at the 95% level) values are in bold
STR-I between SEA-R
STR-P between SEA-R
STR-I between STR-P
Teleconnections within the climate system vary with time; this has been shown for the ENSO variability and Australian rainfall (Nicholls et al, 1996; Power et al., 1999) as well as the relationship between the STR-P and rainfall (Pittock, 1975; Drosdowsky, 2005). Here, the time evolution of the relationships between the STR-I or the STR-P and SEA rainfall is analysed further. Correlation coefficients were computed for 30-year periods (the recommended length by the World Meteorological Organisation to define a climate baseline) between the SEA rainfall monthly series and the two STR monthly values for all the months. Results are shown in Figure 5 for 30-year windows ending in 1929 through to 2009. Time evolution of the correlation coefficients for months for which the correlation coefficients calculated over the 1900–2009 period is significant (as per Table I) is shown with solid line, for the other months dashed lines are used; the annual mean is also shown. In this graph, the correlation coefficient in 2009 was calculated on values from 1980 to 2009. No attempt was made to evaluate the statistical significance of the departure from the long-term correlations for the 30-year coefficients.
In the case of the relationship between STR-I and rainfall, although the correlation coefficients change over time, for the months April to September (when the long-term correlation coefficients are lower than–0.4), they remain in excess of–0.4 in most instances and the fluctuations are generally of small amplitude. Noteworthy are the large changes for the month of May. Correlations were as high as–0.9 during the 30-year period ending 1970 (1940s–1960s) and are now down to–0.4. The very high correlation coefficients in May (the highest of any month from the 1940s to the 1980s) help explain why the STR-I rainfall correlations for the entire period is largest in May (–0.71). The recent value for May (–0.40) is more in line with the view emerging from the 3-month correlations maps: a relationship strengthening during autumn and peaking in the middle of winter; currently, the largest correlation coefficients are indeed noted in August and then July and then June. For the summer months, the correlation coefficients also varied within a limited range. It is interesting to note that while over the entire record only the positive correlation in January is significant, during the most recent period, the correlation coefficients in the rest of summer are equally large. Overall, these evolutions of the correlation coefficients are very likely to be only noise.
In the case of the relationship between the STR-P and rainfall, correlation coefficients for many months exhibit larger swings and are in general noisier than in the case of the STR-I. Noteworthy are the coefficients for the summer months: when computed between 1910 and 1950 (i.e. display on the graph between 1940 and 1950) they were very low, but they are much higher (0.4–0.6 for January and February) from 1950 to 2009. The opposite behaviour is noted in the middle of winter (month of July), with correlations as low as–0.2 in the 1940s decreasing to–0.6 to–0.8 in the more recent period. This is interesting in the context of the long-term trends in position to be discussed next.
Overall, it appears that despite the previously reported changes in the relationship between the STR-P and Australian rainfall (Drosdowsky, 2005) confirmed by results presented here, the highly significant and strong relationship between the STR-I and SEA rainfall during the cold half of the annual cycle (from April to September) is stable over time.
4. Time evolution of the STR
Here we explore the long-term change in the intensity and position of the STR in light of the previously described rainfall relationships. An increase in the intensity of the STR can be inferred from Figures 1 and 2, although a close comparison of the two maps reveals that most of the MSLP increase is located on the southern side of the STR. The time evolution of the STR-I and STR-P are shown by plotting 21-year running means of anomalies of both series for each calendar season and the annual mean (Figure 6). Linear trends from 1900 to 2009 were also computed for each month (Table II); their statistical significance was evaluated by computing the correlation between the time series and the years (bold figures indicate that the 95% significance level is reached).
Table 2. Monthly linear trend for the STR-I (in hPa per century) and the STR-P (in degree South per century) from 1900 to 2009. Statistically significant (at the 95% level) values are in bold
Over the full period of record there is no trend in the position of the STR. The low pass filtering reveals marked multidecadal variation but little trend. Linear trends are not consistent from one month to another. Mostly, southerly trends are noted from April to November but are only significant in April and June (Table II). Trends towards a northerly location of the STR are noted in the other months (significant in February). Although trends are up to 2° in 110 years, only a few are significant because the typical monthly standard deviation is of the order 2–6°.
One can observe that for the summer months, when the STR is typically located south of SEA (Figure 1), the position is trending northward (Table II) hence closer to the southern edge of SEA, and the strength of the positive STR-P relationship is increasing (Figure 5). The time with lowest correlation coefficients (Figure 5) corresponds to a period (from 1910 to 1950) when the STR was anomalously south. Equally, in winter, when the STR is typically located north of SEA (Figure 1), the position is trending southward (Table II) hence closer to the northern edge of SEA and the strength of the negative STR-P relationship is increasing (Figure 5). These observations suggest that the strength of the STR-P and rainfall relationship is not fixed but depends on the position of the STR-P (as can be seen from the annual cycle of the relationship) and the time evolution of the mean position.
In contrast to the STR-P, the intensity of the STR has increased during the 120 years of record, which (Figure 6(b)) is evident for the annual mean, and even more so for winter. Month by month linear trends are consistently positive (Table II). The trends are between 1 and 2.5 hPa for all months from March to October when the relationship between the STR-I and SEA rainfall is strong. The magnitudes of these trends are comparable to the monthly standard deviation (between 1.5 and 2.5 hPa) and hence are statistically significant from March to October (with the exception of May).
5. Role of the STR changes during the dry decades
With this new evidence about changes in the STR, it is worth investigating the role that the STR may have played in the recent SEA rainfall decline. This is particularly relevant since there is a striking similarity between the area which shows the strongest and longest lasting relationship between the STR-I and rainfall (Figures 3 and 4) and the area of rainfall decline (Figure 7(b)).
Rainfall deciles from the two drier periods on record are shown in Figure 7: for the 11-year period 1935–1945 (a) and panel for the 13-year period from 1997 to 2009 (b), relative to the 1900–2009 long-term climatology. The recent drought is localised to southern Australia and SE Queensland while the rest of the Australian continent experienced average to above average rainfall. This contrasts with the WWII drought when about three quarters of the Australian continent was recording below average rainfall. It is also interesting to note that recently the area with the most severe rainfall decline (lowest 13-year rainfall deficiency on record) is located in the southern part of SEA, in the state of Victoria: along the southern ocean coast and in the vicinity of a significant orographic feature, the GDR. In comparison, at the time of the WWII drought, the areas within SEA with lowest rainfall on record were located further north, across northwest Victoria and New South Wales. Overall, as pointed out in Timbal (2009), the area of recent rainfall deficit has a lot of similarity with the maps showing the relationship between rainfall and the STR-I (e.g. the MAM map in Figure 3). This is particularly interesting since it is now well established that the largest part of the rainfall deficit is due to a reduction of the autumn rainfall (Murphy and Timbal, 2008; Timbal, 2009).
The annual cycle of the STR-P (x-axis) and STR-I (y-axis) averaged across the two periods of low rainfall (1935–1945 and 1997–2009) is compared with the long-term (1890–2009) climatology (Figure 8). Both periods display anomalies towards higher intensity (right on the graph) and more southerly (down on the graph) monthly means. For the 1997 to 2009 period, it is particularly marked in autumn to early winter, while for the WWII drought, the largest change is from June to October. Some very large anomalies are also noticeable during the warm part of the year (e.g. the large anomalies for the month of November for the recent period) but based on the maps of the relationship with rainfall (Figures 3 and 4) this is less likely to have an impact on the rainfall than the anomalies between April and October.
In order to quantify how much the rainfall anomalies can potentially be explained by the monthly anomalies in intensity and position of the STR, the linear relationships between the STR-P and STR-I and rainfall were used. We computed STR-related monthly rainfall anomalies for the two dry periods in SEA and compared them with observed values, using the following multiple linear regression models:
with a, b and c, three parameters fitted to the observed values for each month. This model was used three times:
1.with b = 0, to evaluate the role of the STR intensification alone;
2.with a = 0, to evaluate the role of the shift south of the STR alone; and
3.with neither a or b equal to 0, to evaluate the joint effects of the STR intensification and shift south while taking into account the interdependence between these two variables.
Month by month anomalies constructed from STR-I alone are compared with the observed rainfall anomalies for the recent and WWII droughts (Figure 9). For the recent period, the largest reconstructed anomalies were obtained with the STR-I alone, with rainfall anomalies being equal to 62% of the observed decline. This decreases to 59% when both STR-I and STR-P are considered. This is despite the fact that the rainfall anomalies computed from the STR-P anomalies alone are equivalent to 19% of the observed decline and that a higher percentage of observed rainfall variance is being reproduced when both STR-I and STR-P are used (42.4% averaged across the 12 months) compared to STR-I alone (37.2%). The magnitude of the STR-I reconstructed rainfall anomalies increases from 62 to 80% when the rainfall average across SEA is limited to the area with the strongest relationship with the STR. This area was in the southwest of eastern Australia (SWEA) defined as the area where the correlation between the STR-I and rainfall in May–June–July (MJJ) explains more than 20% of the observed rainfall interannual variability (Timbal et al., 2010).
Although there is general agreement on the time of the year when reconstructed and observed monthly deficits are of the same sign, the match between the magnitudes is not good. The reconstructed anomalies are smaller than those observed in autumn, in particular in March when no rainfall decline was anticipated based on the lack of correlations with the STR-I for that month (Table I). The months of July and August are when the reconstructed and observed rainfall anomalies are the most similar.
During the WWII drought (Figure 9(a)), the STR rainfall anomalies are smaller than for the recent drought (Figure 9(b)), and much less than the observed anomalies in most months. Reconstructed rainfall anomalies are equal to 26% of the observed decline with both STR-I and STR-P, 31% with STR-I alone and 4% with STR-P alone. Reconstructed anomalies are comparable to those observed in winter but grossly underestimate autumn and spring months.
In most months and in both periods, despite the sizeable effect of STR-P on rainfall when considered in isolation, it does not appear to have an additional effect when combined with STR-I, in increasing the size of the reconstructed rainfall decline. The intensification of the STR is sufficient. This is in part due to the significant correlation (between 0.39 and 0.64) between STR-I and STR-P from April to November. However, the possibility that the shift in position has an additive but nonlinear effect compared to the strengthening of the ridge cannot be ruled out completely based on this analysis, as the statistics used here are simple and rely on linearity assumptions. More advanced nonlinear techniques would be required to investigate this further. Moreover, the magnitude of the reconstructed rainfall anomalies (two thirds for the recent period and a third for the WWII drought, compared to the observed one) has the potential to be an underestimation of the true effect of the STR changes on the rainfall due to the simplicity of the linear model used.
6. Relationship between the STR and global temperature
The long-term evolution of the STR-I has similarities with the time evolution of the global average surface temperature (Figure 10(a)). For both variables, 21-year running anomalies were calculated, centred on the 1961–1990 World Meteorological Organisation (WMO) reference period. On the graph, different y-axes (on the right for the STR and on the left for global temperature) are used so that the vertical spans of the two curves are comparable.
The long-term correlation of the highly smoothed variables is noticeable. The previous high values of the STR correspond to the 1940s (at about the time of the WWII drought) when the global temperature reaches a maximum before flattening until the 1960s and then rising again in the 1970s as does the intensity of the STR. Although the correlation between the two raw series is relatively high (0.5, Table III), most of it is due to the long-term trends. Once data are detrended, the interannual relationship is weak (0.23) but significant at the 99% level. The relationship with the position of the ridge is lower [0.19 raw (95% level) and 0.12 detrended (90% level)]. Therefore, the apparent relationship on multidecadal time scales between the intensity of the STR and global warming does not appear to arise from a strong linkage at the interannual timescale.
Table 3. Correlation coefficients between the global mean temperature and the STR intensity and position from 1900 to 2009; using raw annual means (first line) and detrended data (second line, the correlated variables were also detrended in that case). Statistically significant (at the 99% level) values are bold
Detrended global temperature
A similar comparison can be made using 21-year running means of SEA rainfall and the detrended global temperature (Figure 10(b), Nota Bene (NB): SEA rainfall sign has been reversed). The global temperature was detrended using the linear trend from 1900 to 2009 and SEA rainfall anomalies were computed from the 1900 to 2009 climatology. An additional element to show that it is more the detrended global warming than the absolute value which matters for SEA rainfall is that the correlation between SEA rainfall and the global temperature is larger when data are detrended than when they are not. With detrended global warming, the correlation is weak but significant at the 99% level.
7. Discussion and conclusions
Rainfall deficiencies across southern Australia and in particular SEA have occurred from 1997 to 2009 in conjunction with above average rainfall over much of the continent further north. This is in contrast with the WWII drought during which the continent as a whole experienced below average rainfall. The spatial pattern of the recent drought has a very strong similarity with the signature of the STR-I correlation with rainfall on interannual timescales. This was not the case in the WWII drought.
The spatial signature of the 1997–2009 drought in SEA (largest decline observed along the southern coast and near significant orography) and its temporal signature (mostly during autumn–winter–spring) further indicate that the rainfall decline is linked to a weakening of the dominant westerly atmospheric flow as one can expect in response to a strengthening of the belt of high pressure.
For the 1997–2009 drought, a rainfall decline equivalent to 62% of the observed decline in SEA rainfall can be reconstructed based on the strengthening of the STR. The annual cycle of the STR-reconstructed SEA rainfall deficit also has a reasonable similarity with that observed. It is also interesting to note, that earlier in the century, during the WWII drought, a rainfall deficit up to 31% of the observed deficit can be inferred from the strengthening of the ridge, implying that factors unrelated to the STR played a more significant role.
Although the southward shift of the ridge has traditionally been looked at as a possible explanation for SEA rainfall variability (Drosdowsky, 2005 and references within), our results suggest that the intensification is a more important factor. The shift in STR-P is not very marked compared to the observed decadal variability, and linear trends are significant for only a few months. In addition, using simple linear statistics, although the southerly shift in the STR in itself can explain some of the observed rainfall decline, much of this explanation arises because of the covariability of the STR-P with the STR-I.
As noted earlier, the STR is the surface signature of the descending branch of the Hadley cell, one of the most significant planetary circulations which contribute to the meridional circulation of energy from the tropics to the higher latitudes (Peixoto and Oort, 1992). Recent studies indicate that, as a whole, the Hadley cell is undergoing significant change, with a poleward expansion which is consistent with the enhanced greenhouse effect (Seidel et al., 2007). In particular, the width of the circulation is expanding. There are numerous lines of evidence to support this, but the magnitude of it is uncertain, estimates range from 2° to 8° in the last 30–50 years (Seidel et al., 2007). These numbers are considerably larger than those projected from climate models (Johanson and Fu, 2009). The intensity of the Hadley cell may also be changing, but the results from different studies are inconsistent (Mitas and Clement, 2005; 2006).
From a mid and upper tropospheric perspective, the Hadley circulation is clearly expanding but its changes in intensity are unclear. From the surface signature of the descending branch, it is clear that the STR as measured by Drosdowsky (2005) is intensifying, while it is unclear that it is significantly shifting south. These findings are apparently contradictory, but it is beyond the scope of this study to reconcile them. A better understanding of the relationship between observed changes in the Hadley cell and the STR is required as well as their relationship with the global mean temperature of the planet. Results shown here suggest that this relationship is important and will contribute to the long-term rainfall variability and the projections of future rainfall in SEA at least for the cooler part of the year (from April to October) when the observed relationship is strong and stable over times. However, while the STR-I appears to be statistically related to global warming, SEA rainfall appears to be more closely related to the detrended global warming.
This work was supported by the South Eastern Australian Climate Initiative (SEACI). Morwenna Griffiths (BoM) generated plots for Figures 1–4. Plots in Figure 7 were produced by the National Climate Centre of the Australian Bureau of Meteorology.