The temporal variability of the frequency of short-duration extreme precipitation events in Australia for the period 1910–2006 is examined using the high-quality rainfall dataset identified by the Bureau of Meteorology, Australia, for 189 stations. Extreme events are defined by duration and recurrence interval: 1, 5, 10, and 30 days, and 1, 5, and 20 yr, respectively. The results indicate that temporal variations of the extreme precipitation index (EPI) for various durations and recurrence intervals in the last 100 yr, except for the low frequencies before 1918, have experienced three U-shaped cycles: 1918–53, 1953–74, and 1974–2006. Seasonal results indicate that about two-thirds of 1-day, 1-yr recurrence interval extreme events occur from December to March. Time series of anomalies of the regional EPIs for four regions indicate that northeast Australia and southeast Australia have almost the same temporal variation as the national anomalies, South Australia experienced a negative anomaly of extreme rainfall events in the mid-1950s, and southwest Western Australia (SWWA) experienced relatively small temporal variation. The relationships between extreme rainfall events and the Southern Oscillation index (SOI) and the interdecadal Pacific oscillation (IPO) indicate that extreme rainfall events in Australia have a strong relationship with both, especially during La Niña years and after 1942.
There is an ongoing interest in characterizing possible changes in climatic extremes, since they cause significant damage to agriculture, ecology and infrastructure, disruption to human activities, injury, and loss of life (Coates 1996; Suppiah and Hennessy 1998; Kunkel et al. 1999, 2002, 2003; Joseph et al. 2000; Brunetti et al. 2006; Aryal et al. 2009). Changes in the probability of extreme rainfall also have important implications for engineering, insurance, town planning, and other activities that traditionally assumed climate stationary for the past century (Suppiah and Hennessy 1998). Increases in heavy rainfall may lead to an increase in the frequency of flood events, landslides, soil erosion, accumulation of silt in dams, inundation of lowland areas, and aquifer recharge due to rising water tables. Analysis of observed rainfall records provides insight into the direction and significance of current rainfall trends (Suppiah and Hennessy 1998).
Australia’s precipitation trends have been examined in several studies. Yu and Neil (1991) found no trends in events of more than 40 mm d−1 at 17 stations in southeast Australia from 1889 to 1985. In a subsequent study, Yu and Neil (1993) reported that while annual average rainfall in southwest Western Australia decreased during the period 1911–90, heavy rainfall increased, particularly in spring and summer. Nicholls and Kariko (1993) investigated rainfall data from five representative stations in eastern Australia for 1910–88, finding increased annual rainfall was mainly due to more rain days rather than higher rainfall intensities. Suppiah and Hennessy (1998) analyzed trends in heavy rainfall, total rainfall, and the number of dry days in Australia for the period 1910–90 using daily rainfall records at 125 stations. Heavy rainfall was defined as the 90th and 95th percentiles of daily rainfall for winter and summer half-years, respectively. They found increasing trends in heavy rainfall and total rainfall during the summer half-year, with statistically significant trends for 10%–20% of stations. During the winter half-year, heavy rainfall and total rainfall have also increased, except in far southwest Western Australia and inland Queensland. Changes in the number of dry days were statistically significant at more than 50% of stations. Haylock and Nicholls (2000) analyzed daily rainfall at 91 high-quality stations over eastern and southwestern Australia for 1910–98. They found a decrease in the extreme frequency and extreme intensity in southwest Western Australia and an increase in the extreme percent (the proportion of total rainfall from extreme events) in eastern Australia. They concluded that trends in the extreme intensity and extreme percent are largely dependent on the method used to calculate the index. Groisman et al. (2005) analyzed the changes in intense precipitation (e.g., the frequency of very heavy precipitation; the upper 0.3% of daily precipitation events) for more than half the land area of the globe, and they found that the 52% (100 yr)−1 increase in frequency of very heavy precipitation in southeastern Australia was statistically significant at the 0.05 level, while the 43% (100 yr)−1 decrease in southwestern Australia was statistically significant at the 0.10 level. Westra and Sharma (2006) used a wavelet regression approach to examine the variability of three rainfall characteristics—the total annual rainfall, the annual number of wet days, and the annual maximum daily rainfall—that could be associated with El Niño–Southern Oscillation (ENSO). The results show that no coherent modes of variability could be found for the annual maximum daily rainfall time series, highlighting the greater level of random behavior in the intensity of larger rainfall events compared with the long-term averages. Alexander et al. (2007) reported trends in Australia’s climate means and extremes under a global context and found that most stations had greater absolute trends in extremes than means. They further pointed out that there was some evidence that the trends of the most extreme events of both temperature and precipitation are changing more rapidly in relation to corresponding mean trends than the trends for more moderate extreme events. Nine indices, reflecting changes to mean annual/seasonal rainfall and extreme daily rainfall (defined by the 95th and 99th percentiles), have been evaluated for six regions in eastern and southwestern Australia by Gallant et al. (2007). The general conclusion was that the direction of changes in extreme rainfall was consistent with changes in the mean. Alexander and Arblaster (2009) assessed trends in observed and modeled climate extremes over Australia in relation to future projections. They concluded that the majority of nine globally coupled climate models they assessed reproduced the correct sign of trend for precipitation extremes although there was much more variation between the individual model runs, and that Australia showed a shift toward warming of temperature extremes with much longer dry spells interspersed with periods of increased extreme precipitation, irrespective of the scenario used. Beecham and Chowdhury (2010) identified the temporal characteristics and variability of point rainfall measured in Melbourne, Australia. Statistical moments, lag-1 autocorrelation, the Buishand Q test for homogeneity, the Mann–Kendall (MK) test for trend, and wavelet analyses for temporal variability were carried out for rainfall intensities at resolutions of 0.1, 0.5, 1, 3, 6, and 12 h and for the monthly rainfall depths and proportion dry ratios. While no statistically significant trends were found using the MK test, there were indications that trends are more likely as the temporal scale increases.
This study, extending these existing studies, examines temporal changes in the number of extreme rainfall events in Australia that may be linked to flooding. Event durations of 1, 5, 10, and 30 days and recurrence intervals of 1, 5, and 20 yr were analyzed. These thresholds are determined empirically from station climatology using a partial duration series analysis. This particular measure of extreme precipitation events (multiday, exceeding a recurrence interval threshold) was used because previous studies (Changnon and Kunkel 1995; Kunkel et al. 1999, 2002, 2003) had found a good correlation between such events and hydrologic flood events on small- to medium-size rivers in the Midwest of the United States. In the absence of any similar study reported in the literature for Australia, we make the assumption that these durations are also appropriate for a study of Australian extremes.
Extreme precipitation index (EPI), which is defined as the number of extreme precipitation events for a specific duration and recurrence interval, was developed and computed for Australia. Besides the annual and national EPI, we also calculated four seasonal EPIs [autumn, March–May (MAM); winter, June–August (JJA); spring, September–November (SON); and summer, December–February (DJF)] and four regional EPIs [northeast Australia, southeast Australia, South Australia, and southwest Western Australia (SWWA)] to investigate seasonal and spatial differences. The relationship between the number of extreme rainfall events and the total annual rainfall, the number of rainfall days, the Southern Oscillation index (SOI), and the interdecadal Pacific oscillation (IPO) is also investigated.
2. Data and methods
There are 191 long-term rainfall stations in the high-quality rainfall dataset of Lavery et al. (1992). These stations have passed several selection criteria and form the most reliable dataset available for examining rainfall characteristics in Australia (Lavery et al. 1992), and root transformations of daily precipitation amounts for these stations can be fitted pretty well with a truncated normal distribution model (Fu et al. 2010a). The Bureau of Meteorology (BOM), Australia, has filled missing data—if any—to make us have complete records for the entire study period. However, daily precipitation data for two stations—station 024503 (Blanchetown) and station 027003 (Booby Island)—are not available from the National Climate Centre (NCC) of the BOM. The remaining 189 stations, whose data are available, are used in this study. Detailed information for these stations—including station number, World Meteorological Organization (WMO) station number if any, station name, first year of observation, latitude, longitude, and elevation—is listed in Lavery et al. (1992). Our study period is from 1910 to 2006, as records prior to 1910 were not quality checked by Lavery et al. (1992). The high-quality dataset has been extended to 379 stations by Lavery et al. (1997). However, this 379-station dataset was constructed by merging data from two or more nearby stations with incomplete records or short observation periods to produce single rainfall time series. As such, there are high uncertainties in basing trend analysis on this data. Therefore, this updated high-quality dataset is not used in this study. Figure 1 shows the locations of the 189 stations used in this study and four regions used for the regional EPI study.
One critical concern is the accuracy of daily rainfall data, as Viney and Bates (2004) have shown that 102 of 181 stations based on the list of Haylock and Nicholls (2000) contain hidden, untagged accumulations, and the overall prevalence of untagged accumulations in the high-quality dataset is only slightly less than that of tagged accumulations. The results of 1-day extreme rainfall events with a 1-yr recurrence interval show that Monday has the largest number of extreme rainfall days (14.9%) and that Sunday has the second smallest (14.0%). However, as the difference is only about 1%, the effect of untagged accumulations is limited and therefore not addressed in this study.
The methodology adopted in this study closely follows that of Kunkel et al. (1999, 2003) and uses the same durations and recurrence intervals. Event durations of 1, 5, 10, and 30 days were examined. Three precipitation thresholds were used to screen events defined by recurrence intervals of 1, 5, and 20 yr. The thresholds were determined for each station; for example, for a station with a complete 97-yr observation record, for 5-day events with a 1-yr recurrence interval, the largest ninety-seven 5-day precipitation totals in the period 1910–2006 were identified, with the threshold being the smallest of these 97 values.
The thresholds for 1-day, 1-yr rainfall events range from less than 25 mm in the southwest region of Western Australia, South Australia, and the western region of Victoria to more than 160 mm along the northeast coast of Queensland (Fig. 2). The thresholds for other duration and recurrence interval combinations have similar spatial distributions but different magnitudes and ranges depending on the duration and recurrence intervals. The thresholds for 1-day, 5-yr rainfall events range from 38 to 298 mm (Fig. 3) and those for 1-day, 20-yr rainfall events range from 53 to 449 mm. The average, median, minimum, and maximum values of 189 stations for 1-day, 5-yr events are about 65%–75% larger than those of 1-day, 1-yr events. The thresholds values of 1-day, 20-yr events are about 140%–170% larger than those of 1-day, 1-yr events.
The median and mean thresholds for 5-, 10-, and 30-day events are about 45%–65%, 65%–100%, and 140%–200% larger than those of the corresponding 1-day events (Fig. 3), respectively. The thresholds for 30-day, 20-yr rainfall events range from 120 to 1501 mm, which are 5.4 and 9.0 times of those of 1-day and 1-yr events (Fig. 3).
After the threshold value has been determined for each station, the annual numbers of events for each duration and recurrence interval are identified. The numbers of 1-day events can only be integral values, while those of 5-, 10-, and 30-day events could be decimal value, as a multiday extreme rainfall event can span two different years. There is no overlapping of days for a specific duration and recurrence interval. That is, if 10 consecutive days have been identified as a 10-day extreme “event,” then none of those 10 days can be used for a second event for the same duration and recurrence interval. However, part or all of these 10 days can be used for an extreme event for a different duration and recurrence interval. At durations of 30 days, these are certainly not events in the typical sense but most likely rainfall accumulated over a number of events. However, it is typical to analyze time series of rainfall at a number of very long time scales in the case of methodologies that rely on scaling behavior (Douglas and Barros 2003; Sun and Barros 2010), and consecutive rain events might have hydrological implications. For example, an early rainfall event makes the soil more saturated, which could make a later rainfall event produce more runoff. Therefore, although not physically consistent with the shorter duration events, 30-day events are still treated as single events and compared with them. In contrast, two 1-day events are likely from one single event, but they are treated as two separate events because autocorrelation is not statistically significant at 5% significance limits for all 1-day extreme event time series (statistic testing results are not shown).
To assess national extreme rainfall temporal trends, the numbers of extreme rainfall events for a given duration and recurrence interval across 189 stations are weighted by Thiessen polygon area (Thiessen 1911; Fig. 4) to produce a national EPI for this duration and recurrence interval. These EPIs constitute the basic time series of this study. Spatial correlation may have effects on the amplitude of the EPI associated with the Thiessen weighting procedure. For example, a single storm event occurs across multiple locations at the same time, potentially influencing the amplitude of the variability, as rainfall at a given station would increase the likelihood of rainfall at neighboring stations and thus increase the total number of threshold exceedances in a given year. However, the spatial correlation would not change the timing of the peaks and temporal variation, which are main objectives of this study. In addition, the spatial correlation usually decreases with spatial distance, which implies that a higher spatial correlation coefficient usually has a smaller weight for the Thiessen polygon it represents. Therefore, the effects of spatial correlation on the national EPI are likely to be small.
Four seasonal EPIs and four regional EPIs (mentioned earlier and shown in Fig. 1) are also produced to investigate the seasonal and spatial differences of temporal variability of the number of extreme rainfall during the last 100 years. The regional EPIs are produced by arithmetically averaging all stations in the region, because meteorological stations are relatively evenly distributed in space. These four regions are chosen because they have relatively dense networks of daily rainfall observations.
The total annual rainfall and number of rainfall days in the last 100 years is determined nationally using the same Thiessen polygon averaging as used for extreme events.
3. Results and discussion
a. Temporal variability in frequency of extreme rainfall events
The composite time series of 1-day, 1-yr events is notable for large interannual- and decadal-scale variability (Fig. 5). Several key climate events of severe droughts or moisture surpluses are evident in this series. For example, droughts during the period 1922–1930 and in 2002 (Horridge et al. 2005) are characterized by a lower number of extreme rainfall events, and 1974, a year of flooding, experienced the highest number of extreme rainfall events during the last 100 years.
It should be noted that Fig. 5 shows overall results for the entire continent and that lower numbers of extreme rainfall events in a specific year could still produce very serious floods at a regional scale. For example, 1986 has a lower number of extreme rainfall events nationally, but Sydney had a serious flood that year (Lynch 1987).
The interdecadal variability of extreme rainfall events in Australia is substantial. This is consistent with previous research indicating significant changes of climate in Australia on decadal-to-multidecadal time scales (Simmonds and Hope 1997; Kiem and Franks 2004). Periods of below-average frequency occurred from the 1920s to the 1940s (Fig. 6). The early and middle 1950s, early-to-middle 1970s, and the 2000s are periods of above-average frequency.
The temporal variability of combinations of other durations and recurrence intervals is highly consistent with the interannual and interdecadal variability of 1-day, 1-yr extreme rainfall (Fig. 6). This is not surprising since there is considerable overlap among the time series for the various durations; that is, within many of the 10-day events, there are 1- and 5-day precipitation totals that exceed the threshold for a 1-, 5- or 20-yr recurrence. Such consistency in temporal variation for different durations and recurrence intervals is also reported in the United States and Canada (Kunkel et al. 1999, 2003).
During the last 100 years, except for the low frequencies before 1918, the temporal variations of extreme rainfall events at various durations and recurrence intervals in Australia show three U-shaped cycles. The first cycle runs from 1918 to 1953 and is asymmetric with 1925–38 being the trough. The second cycle runs from 1953 to 1974. This cycle has the shortest length. However, the high frequencies of extreme rainfall in the mid-1970s are stronger than those in the mid-1950s. In fact, the extreme rainfall events in the mid-1970s have the largest frequencies in the last 100 years. The last cycle runs from 1974 to the early-twenty-first century. The 20-yr recurrence interval EPIs have slightly different spans and amplitudes. However, three U-shaped cycles are also clearly demonstrated (Fig. 6).
Given these three U-shaped cycles, Kendall trend testing does not show significant trends of the numbers of extreme events during the period 1910–2006 for the majority of the stations (Table 1), although some do show a statistically significant trend during the last 100 years. The nonparametric Kendall test was first adopted by Hirsch et al. (1982), modified from the Mann–Kendall test (Kendall 1975; Gan 1998). The advantages of this method include that 1) it can handle nonnormality, censoring, or data reported as values “less than,” missing values, or seasonally; and 2) it has a high asymptotic efficiency (Berryman et al. 1988; Gan 1998; Fu et al. 2004). Overall, there are more stations with negative trends than positive trends independent of whether the trends are statistically significant (Table 1). The number of stations with statistically significant trends declines with increasing recurrence interval. There are only 2–9 stations showing a statistically significant trend for the 5-yr recurrence interval, but 17–28 stations have a statistically significant trend for 1-yr recurrence intervals (Table 1). The time series of 20-yr recurrence intervals consist of only five nonzero values at each station, so their trend analyses have not been implemented in this study. The statistically insignificant trend of the number of extreme events in Australia contrasts with results for the United States and Canada, where the Kendall τ statistic indicated a statistically significant (at the α = 0.05 level) increasing trend (Kunkel et al. 1999, 2002, 2003). This might be because in the latter case, the entire 1895–2000 period has a U-shaped pattern, with rather high frequencies of extreme rainfall events at the end of the nineteenth century and in the early part of the twentieth century for the United States and Canada (Kunkel et al. 1999, 2002, 2003).
Another feature of extreme rainfall events in Australia, in contrast with the United States and Canada (Kunkel et al. 1999, 2003), is the larger variability. The time series of anomalies of the EPI for Australia (7-yr moving average) ranges from −79.7% to 136.4%, whereas the corresponding series for the United States varies from approximately −45% to 50% (Kunkel et al. 2003).
The definition of extreme rainfall events used in this study contrasts with the heavy rainfall events trend analyses of Suppiah and Hennessy (1998), who used the 90th and 95th percentiles of daily rainfall. Therefore, there are some differences between the results and conclusions. For example, they reported increasing trends in heavy rainfall and total rainfall during the summer half-year, with statistically significant trends for 10%–20% of stations; however, our results indicate that there are more decreasing than increasing stations in terms of the number of extreme rainfall events (Table 1). In addition, the study of Suppiah and Hennessy (1998) was mainly centered on trend analyses, whereas this study focused on long-term temporal variations.
b. Seasonal variability in frequency of extreme rainfall events
In Australia, northern regions are affected by summer monsoon rainfall, whereas southern regions are strongly influenced by midlatitude rain-bearing systems in winter (Suppiah and Hennessy 1998). To account for these seasonal and regional differences, the temporal variation of extreme rainfall events for each of four seasons is examined separately.
A monthly analysis (Fig. 7) indicates that about two-thirds of 1-day, 1-yr recurrence interval extreme rainfall events occur from December to March. However, the annual cycle of the extreme rainfall events varies from station to station with some stations winter dominated. For example, 42% of extreme rainfall events at stations 8141 (Willi Gulli North), 9038 (Rottnest Island), 9503 (Boyanup Post Office), and 9519 (Cape Naturaliste) occur in winter (JJA; Fig. 7). All four of these stations are located on the west coast of southern Western Australia, where the majority of annual rainfall occurs in the Southern Hemisphere winter months, with more than 80% falling during April–October.
The seasonal differences are clear in Fig. 8. DJF has the largest number of extreme rainfall events in almost every year. In 1974 and 2000, the high frequencies of extreme rainfall events are consistent with the annual series. MAM has the largest number of extreme rainfall events in certain years. For example, the number of MAM extreme rainfall events is larger than DJF for 1910, 1921, 1932, 1983, 1989, 1990, and 2006. Overall, MAM contributes about 31% of extreme rainfall events for 1-day, 1-yr recurrence intervals in Australia. JJA and SON each contribute about 10% of extreme rainfall events. JJA is the dominant extreme rainfall season for 1923, 1965, and 2005, and SON is never the dominant extreme rainfall season. JJA extreme rainfall events occur mainly in Western Australia.
DJF has the temporal variation that is most similar to the annual time series, that is, three U-shaped cycles, except that it has a relatively short period for the first cycle (Fig. 9). The first runs from 1918 to 1940, then the period 1940–53 is relatively stationary. The period spans for the second and third U-shaped cycles are almost identical to those of the annual series.
In contrast to the three U-shaped cycles of the annual and DJF time series, the MAM time series has five U-shaped cycles in the last 100 years (Figs. 6 and 9). The first cycle runs from 1918 to 1932, the second cycle from 1933 to 1953, the third cycle from 1953 to 1971, the fourth cycle from 1971 to 1991, and the fifth cycle from 1991 to 2003. In some periods, such as the mid-1920s, the mid -950s, and the mid-to-late 1970s, the MAM anomalies index has the same sign as the annual and DJF time series. In other periods, such as the late 1920s to early the 1930s, and the late 1980s to the 1990s, they have opposite signs; that is, the MAM season has a positive anomaly of extreme rainfall events, whereas the annual series has a negative anomaly. Compared with annual and other seasonal time series (Figs. 6 and 9), MAM shows a lower variability during the last 100 years.
The SON time series has four U-shaped cycles (Fig. 9). The timing of the third and fourth cycles is almost the same as those of the second and third cycles for the annual and DJF time series. The difference is that the long time span of the first trough (1918–53) for the annual and DJF time series splits into two for SON: 1917–36 and 1936–53 (Fig. 9). SON has the largest anomalies, about 82% in 1917 and about −60% in 1928. In addition, its anomalies during the periods 1965–66 and 1998–2003 are smaller and larger than the other seasons, respectively. Whereas the mid-1970s have the largest frequencies of extreme rainfall events for annual and other seasonal time series, it is 1917 for the SON season.
As was the case for the annual time series, the JJA anomalies also have three U-shaped cycles (Fig. 9). However, there are two unique features of the JJA extreme rainfall events. First, the time span of the second trough is very short, only about 12 years from 1954 to 1965. It is the shortest span of all the seasonal and annual anomaly cycles. The period from 1965 to 1971 is characterized by a positive anomaly of extreme rainfall events for the JJA season but a trough for other seasons. Second, after 1996, the anomalies of the annual, DJF, and SON time series are positive (i.e., it is a period with a relatively large number of extreme rainfall events). However, the anomalies of JJA and MAM are negative.
c. Temporal variation of annual rainfall and number of rainfall days
The time series of anomalies of total annual rainfall and the number of rainfall days (defined as days with at least 0.1 mm of rainfall) have almost the same temporal variation as the extreme rainfall events, that is, three U-shaped cycles during the last 100 years (Fig. 10). The starting and ending points of each of these three cycles are almost identical for the number of rainfall days, total annual rainfall events and extreme rainfall events (Fig. 10). However, significant differences exist among the temporal variations of these variables.
The first difference is that the magnitudes of the anomalies of the total annual rainfall and the number of rainfall days are generally smaller than those of extreme rainfall events. The 7-yr moving average anomalies of total annual rainfall and the number of rainfall days vary between −14.7 and 26.2% and −15.4 and 23.5%, respectively, while those of extreme rainfall events range from −33.7 to 51.8% (Fig. 10). The range of anomaly for extreme rainfall events is thus twice those of total annual rainfall and the number of rainfall days (Fig. 10).
The second difference is that the number of rainfall days during the period 1979–98 is larger than the long-term average, while the total annual rainfall and extreme rainfall events are smaller than the long-term average. This period with a higher number of rainfall days leads to a statistically significant (at α = 0.05 level) increasing trend being detected with the Kendall test, whereas there are no statistically significant trends detected for the total annual rainfall or extreme rainfall events. This trend in the number of rainfall days is consistent with the conclusions of Suppiah and Hennessy (1998) and Haylock and Nicholls (2000). Suppiah and Hennessy (1998) state that there has been a reduction in the number of dry days in both summer and winter halves of the year, except in far southwest Western Australia and at a few stations in eastern Australia, where there has been an increase in the number of dry days in the winter half-year. Changes in the number of dry days were statistically significant at more than 50% of stations analyzed by Suppiah and Hennessy (1998). Haylock and Nicholls (2000) detected a significant trend in the number of rain days and confirmed that the total rainfall of Australia is strongly correlated with the frequency and intensity of extreme events.
d. Linkage of extreme rainfall events and climatic variability
The relationships between multidecadal extreme rainfall events and climatic variability, such as the ENSO as measured by the SOI and the IPO index, are investigated here. The SOI is calculated from the monthly or seasonal fluctuations in the air pressure difference between Tahiti and Darwin, Australia (BOM 2009). Sustained negative values of the SOI often indicate El Niño episodes. These negative values are usually accompanied by sustained warming of the central and eastern tropical Pacific Ocean, a decrease in the strength of the Pacific trade winds, and a reduction in rainfall over eastern and northern Australia. Positive values of the SOI, popularly known as La Niña episodes, are associated with stronger Pacific trade winds and warmer sea temperatures to the north of Australia. Waters in the central and eastern tropical Pacific Ocean become cooler during this time. Together, these give rise to an increased probability that eastern and northern Australia will be wetter than normal (BOM 2009).
The IPO is almost the Pacific-wide manifestation of the Pacific decadal oscillation of Mantua et al. (1997), with as much variance in the Southern Hemisphere Pacific down to at least 55°S as in the Northern Hemisphere. The IPO is a multidecadal sea surface temperature pattern quite like that of ENSO but different in several ways (Fu et al. 2009). It shows a marked amount of symmetry about the equator. It was introduced by Power et al. (1999) based on work by Folland et al. (1999). Power et al. (1999) showed that the IPO modulated teleconnections between ENSO and Australian climate.
The results indicate that extreme rainfall events in Australia seem to have a strong relationship with the SOI, especially during La Niña years and after 1942 (Fig. 11). Almost every major La Niña year in Australia results in a relatively high number of extreme rainfall events. This result shows that as well as the known correlation between La Niña and annual rainfall (Abtew et al. 2009; Brown et al. 2009; Fu et al. 2007, 2009, 2010b; Micevski et al. 2006; Misra 2009; Risbey et al. 2009; Ropelewski and Halpert 1987; and many others), there is also a similar relationship with the numbers of extreme rainfall events. However, the major El Niño years do not necessarily produce a trough of extreme rainfall events (Fig. 11a). The phenomenon that the SOI was high when the EPI was at its lowest during the 1920s, and the inconsistency between the SOI and EPI before 1942, needs further investigation.
The effects of ENSO on extreme precipitation events seem more significant at interdecadal than at interannual scales, as their relationship was not immediately obvious simply by glancing at Fig. 5; however, it is apparent with Fig. 11a after applying a 5-yr smoothing filter. This conclusion is confirmed with the results of a wavelet analysis (Torrence and Webster 1999; Gobena and Gan 2009; Fig. 12): for the 15–25-yr period, the SOI and EPI are statistically significantly correlated for the entire study time (1910–2006), whether or not a 5-yr smoother was used; for the 4–6-yr period, they are only significantly correlated for certain years, whether or not a 5-yr smoother was applied. The definition of the wavelet coherence of two time series from Torrence and Webster (1999) is adopted in this study:
where 〈·〉 indicates smoothing in both time and scale. The factors s−1 is used to convert to an energy density. Here WnX(s) and WnX(s) are the wavelet transforms of time series of X and Y, which are EPI and SOI in this study, respectively, n is the time index, and s is the scale (Torrence and Webster 1999). “Note that in the numerator, both the real and imaginary parts of the cross-wavelet spectrum are smoothed separately before taking the absolute value, while in the denominator it is the wavelet power spectra (after squaring) that are smoothed” (Torrence and Webster 1999). The statistical significance level (α = 0.05) of the wavelet coherence against background red noise was estimated using Monte Carlo sampling.
The IPO also seems to have a strong relationship with extreme rainfall events in Australia. The EPI is anticorrelated with the IPO (Fig. 11b): every peak of the IPO responds to a trough of the EPI and vice versa, although the magnitudes of two series do not necessarily being proportional.
e. Temporal variation in frequency of extreme rainfall events on regional scales
As Australia has a highly variable climate from region to region, the temporal variations of extreme rainfall events in four regions have been further investigated to explore these regional differences. These four regions include 183 of our 189 stations. The networks of rainfall observations are relatively dense in these regions, and a majority of the Australia population lives in these regions. Their geographic locations and boundaries are shown in Fig. 1.
The threshold values for an extreme rainfall event vary from region to region (Table 2). South Australia has the lowest thresholds for various durations and recurrence intervals. SWWA has the second lowest thresholds. The differences between the minimum and maximum thresholds are also small for these two regions. The ratio of maximum to minimum station thresholds varies from 1.4 to 2.9 and from 1.7 to 3.9 for various combinations of rainfall durations and recurrence intervals, for South Australia and SWWA, respectively (Table 2).
Northeast Australia has the largest thresholds for various rainfall durations and recurrence intervals, especially for the maximum individual station thresholds (Table 2). The minimum station thresholds for northeast Australia are generally 0.9–1.6 times those of SWWA, and 1.2–1.9 times those of South Australia. However, the maximum station thresholds for northeast Australia are generally 2.7–5.5 times those of SWWA and 4.5–8.5 times those of South Australia (Table 2). This is consistent with results from previous studies showing that this region has been the most vulnerable to flooding during the period 1788–96 (Coates 1999). This spatial distribution can also be observed from Fig. 3, which displays 1 of 12 combinations of rainfall duration and recurrence interval used in this study. If other thresholds are plotted, then they have almost identical spatial distributions with large magnitudes and ranges among stations, depending on the duration and recurrence intervals. The ratio of maximum to minimum individual station thresholds for various combinations of rainfall duration and recurrence intervals in northeast Australia varies from 4.6 to 12.4, which is much larger than those in SWWA and South Australia.
The thresholds and ratios of maximum and minimum station thresholds for southeast Australia fall between those of SWWA and South Australia, and northeast Australia (i.e., smaller than those in northeast Australia but larger than those in SWWA and South Australia).
The monthly distributions of extreme rainfall events also vary from region to region (Fig. 13). Northeast Australia has the largest variation: extreme rainfall events in the summer (DJF) comprise about 56% of the annual extreme rainfall events. Both January and February have more than 20% of the annual extreme rainfall events. In contrast, extreme rainfall events in both August and September are less than 1% of the annual extreme rainfall events (Fig. 13). South Australia and southeast Australia have relatively even distributions of monthly extreme rainfall events across the year. The months with the most extreme rainfall events in both regions receive about 15% of the annual extreme rainfall events, and the months with the fewest extreme rainfall events receive about 4%.
SWWA has a relatively small monthly variation of extreme rainfall events. The month with most extreme rainfall events—May—accounts for 16.4% of SWWA’s annual extreme rainfall events and the month with the fewest extreme rainfall events—September—about 3%. This range is smaller than that of northeast Australia but larger than those of South Australia and southeast Australia. However, SWWA receives most of its extreme rainfall events in late autumn (May) and early winter (June; Fig. 13), whereas the other regions receive the majority of their extreme rainfall events in summer (DJF).
Time series of regional EPI anomalies for the four regions indicate that northeast Australia and southeast Australia have almost the same temporal variation as the national anomalies (Fig. 14). The three U-shaped cycles are clear, and the starting and ending years for each cycle are almost identical. However, there are two major differences between these regional and national EPIs. First, both northeast Australia and southeast Australia show high frequencies in the mid-1950s. These anomalies are almost as high as those of the high-frequency period in the mid-1970s (Fig. 14). The national EPI anomaly is less significant during this period (Figs. 5 and 6). Second, the regional anomalies at the end of the twentieth century and the beginning of the twenty-first century are negative for both northeast Australia and southeast Australia (Fig. 14), whereas the national EPI anomaly is positive (Fig. 6).
The three U-shaped cycles for South Australia are also clear in Fig. 14. However, one significant difference between the South Australian anomalies and the national and other regional anomalies is that when northeast Australia and southeast Australia, as well as the nation, experienced a high frequency of extreme rainfall events in the mid-1950s, South Australia experienced a trough of extreme rainfall events (Fig. 14). The end point for the first trough has advanced to 1948 accordingly.
Except for a high frequency of extreme rainfall events from 1916 to 1920, SWWA experienced small temporal variation compared with national and other regional anomalies (Fig. 14). The three U-shaped cycles seen in the other regions and nation do not appear in the SWWA time series. However, the winter (JJA) extreme rainfall event anomalies indicate a significant decreasing trend since the 1970s: there are four regular cycles from 1910 to 1970 and each cycle is about 15 years, and this periodic phenomenon has disappeared since 1970s (Fig. 15). This is temporally consistent with the decline of the inflow into reservoirs of Perth’s water supply by about 40%–65% since the 1970s (IOCI 2002; Bates et al. 2008), for an annual rainfall decrease only 10%–15%. This conclusion is also consistent with a previous finding that the winter daily rainfall extremes in SWWA are lower after 1965 (Li et al. 2005).
The definition of events of extreme rainfall used in this study is based on duration and recurrence intervals proposed by Kunkel et al. (1999, 2003). The high-quality dataset identified by Lavery et al. (1992) is used to study the long-term variation of extreme rainfall events in Australia. The results indicate that the time series of the number of extreme rainfall events for various rainfall durations (1, 5, 10, and 30 days) and recurrence intervals (1, 5, and 20 yr) show notable interannual and interdecadal variability. During the last 100 years, except for the low frequencies before 1918, the temporal variations of extreme rainfall events in Australia are summarized by three U-shaped cycles: 1918–53, 1953–74, and 1974–2006.
Seasonal results indicate that about two-thirds of 1-day, 1-yr recurrence interval extreme events occur from December to March (Fig. 7). Time series of anomalies of the regional EPIs for four regions indicate that northeast Australia and southeast Australia have almost the same temporal variation as the national anomalies, while South Australia experienced a negative anomaly of extreme rainfall events in the mid-1950s when other regions had a positive anomaly, and SWWA experienced a relatively small temporal variation. However, the SWWA winter (JJA) anomalies of extreme rainfall events indicate that there has been a statistically significant decreasing trend since the 1970s.
The relationships between extreme rainfall events and the SOI and the IPO index indicate that extreme rainfall events in Australia have a strong relationship with both, especially during La Niña years and after 1942. The EPI U-shaped cycles occur consistently anticorrelated with the IPO index.
The present analysis weights the statistics for all stations by Thiessen polygon areas, including the large desert regions in the center of the Australia, where limited observations are available. Thus, our results contain a relatively greater contribution from individual desert region stations and fewer contributions from wetter and dense observation stations, such as the stations in the eastern coast regions.
However, one major uncertainty is the effect of future climatic change. If climate change shifts regional and national climate into regimes outside the observed record, then this analysis needs to be updated to account for this change. This work will be extended in the future to assess the effects of climatic change on regional and national extreme rainfall frequency and intensity.
This research was partly funded by the Australian Greenhouse Office Climate Change Science Program, the Indian Ocean Climate Initiative (IOCI), the South Eastern Australia Climate Initiative (SEACI), the “Hundred Talents Program” of Chinese Academy of Sciences, and eWater CRC projects. We wish to thank Prof. Ana P. Barros (chief editor) and three anonymous reviewers for their invaluable comments and constructive suggestions used to improve the quality of the manuscript.
Corresponding author address: Guobin Fu, CSIRO Land and Water, Private Bag 5, Wembley, WA 6913, Australia. Email: firstname.lastname@example.org