impact of the el nin˜o–southern oscillation, indian ocean...
TRANSCRIPT
Impact of the El Nino–Southern Oscillation, Indian Ocean Dipole, andSouthern Annular Mode on Daily to Subdaily Rainfall Characteristics
in East Australia
ALEXANDER PUI AND ASHISH SHARMA
School of Civil and Environmental Engineering, University of New South Wales, Sydney, Australia
AGUS SANTOSO
Climate Change Research Centre, University of New South Wales, Sydney, Australia
SETH WESTRA
School of Civil, Environmental and Mining Engineering, University of Adelaide, Adelaide, Australia
(Manuscript received 22 July 2011, in final form 10 December 2011)
ABSTRACT
The relationship between seasonal aggregate rainfall and large-scale climate modes, particularly the
El Nino–Southern Oscillation (ENSO), has been the subject of a significant and ongoing research effort.
However, relatively little is known about how the character of individual rainfall events varies as a function of
each of these climate modes. This study investigates the change in rainfall occurrence, intensity, and storm
interevent time at both daily and subdaily time scales in east Australia, as a function of indices for ENSO, the
Indian Ocean dipole (IOD), and the southern annular mode (SAM), with a focus on the cool season months.
Long-record datasets have been used to sample a large variety of climate events for better statistical signif-
icance. Results using both the daily and subdaily rainfall datasets consistently show that it is the occurrence of
rainfall events, rather than the average intensity of rainfall during the events, which is most strongly influ-
enced by each of the climate modes. This is shown to be most likely associated with changes to the time
between wet spells. Furthermore, it is found that despite the recent attention in the research literature on
other climate modes, ENSO remains the leading driver of rainfall variability over east Australia, particularly
farther inland during the winter and spring seasons. These results have important implications for how water
resources are managed, as well as how the implications of large-scale climate modes are included in rainfall
models to best capture interannual and longer-scale variability.
1. Introduction
Understanding rainfall variability at interannual and
longer time scales is an important area of study with
many practical implications for how we manage our
water resources. This variability, which is often linked
to both internally and externally forced fluctuations in
the global sea surface temperature (SST) field (Bigg
et al. 2003; Westra and Sharma, 2010), can often affect
the weather patterns of entire continents via a set of
large-scale pressure and circulation anomalies known
as teleconnections that transmit these anomalies across
large distances (e.g., Barnston and Livezey 1987).
In Australia, much of the early research into climate
variability focused on the El Nino–Southern Oscillation
(ENSO) phenomenon, which has been linked to pre-
cipitation anomalies across Australia, particularly over
the eastern third of the continent (Pittock 1975; McBride
and Nicholls 1983; Drosdowsky 1993; Chiew et al. 1998).
This phenomenon is usually described in terms of cou-
pled oceanic and atmospheric variability centered on
the central and eastern tropical Pacific Ocean, with the
extreme phases of El Nino and La Nina typically re-
sulting in below- and above-average rainfall throughout
large portions of Australia. Importantly, ENSO cycles
between its extreme phases with a period of between
approximately 3 and 7 years (McBride and Nicholls 1983;
Corresponding author address: Ashish Sharma, School of Civil
and Environmental Engineering, University of New South Wales,
Sydney 2032, Australia.
E-mail: [email protected]
MAY 2012 P U I E T A L . 1665
DOI: 10.1175/MWR-D-11-00238.1
� 2012 American Meteorological Society
Drosdowsky 1993), although the predictability of in-
dividual ENSO events generally does not extend much
beyond 1 year (Barnston et al. 1999).
The vast majority of the research described above has
focused either on aggregate rainfall at seasonal time
scales (e.g., McBride and Nicholls 1983; Drosdowsky
1993; Ropelewski and Halpert 1987; Chiew et al. 1998;
Risbey et al. 2009a), or on extremes (e.g., Cayan et al.
1999; Gershunov and Barnett 1998; Liebmann et al. 2001,
Pui et al. 2011). However, use of aggregated rainfall data
can result in significant loss of information about the in-
dividual processes that make up interannual variability.
For example, is an increase in spring average rainfall
during La Nina events due to an increase in the number
of discrete precipitation events occurring, or is it due to
more intense rainfall occurring during each precipitation
event? Such information is not only useful for developing
a better mechanistic understanding on the nature of in-
terannual variability, but also has significant practical
implications in areas such as agriculture (e.g., the number
of wet days per crop growing season, as well as the time
between consecutive wet days) and flood estimation.
ENSO attribution on daily time scales has already
been the subject of several studies, which capture vari-
ous facets of the multiscale rainfall variability issue. For
example, Ropelewski and Bell (2008) focused on shifts
in fine temporal-scale rainfall statistics conditional to
ENSO in the South American continent, whereas Samuel
and Sivapalan (2008) analyzed the variability of inter- and
intra-storm properties at three locations in Australia
(Perth, Newcastle, and Darwin), which experience dif-
ferent climatic regimes. Finally, Nicholls and Kariko
(1993) investigated only daily rainfall characteristics
such as intensities and number of wet days, at five lo-
cations in east Australia that were influenced by the
Southern Oscillation.
In this present study, we will build upon this earlier
research, and address the following questions: 1) to
what extent are rainfall occurrences and the intensity of
rainfall per occurrence affected by these climate modes?
2) Are these changes also reflected when considering
individual storm events using the subdaily rain gauge
dataset? 3) Do these changes fully account for observed
variability at the seasonal and longer time scale? To this
end we will use a much larger dataset of both daily and
subdaily precipitation sequences across east Australia.
This region is selected because of the abundance of data
compared with other locations within Australia, together
with the strong association between rainfall and cli-
mate modes as documented in earlier studies (e.g.,
Nicholls 1997; Chiew et al. 1998).
While ENSO is recognized as the leading source of
climate variability in Australia, other climate modes are
known to have significant impact on regional rainfall
(e.g., Risbey et al. 2009a; Hill et al. 2009). The influence
of various climate modes increases the complexity in
the attribution and thus prediction of seasonal rainfall,
which is further exacerbated by complexities within
ENSO itself (e.g., Westra and Sharma 2006). Thus,
a second goal of this present study is to evaluate the
impact of other major climate modes on daily and sub-
daily rainfall statistics. To this extent, we focus on the
Indian Ocean dipole (IOD), which is of tropical origin,
and the southern annular mode (SAM), of extratropical
origin, both of which have significant impact over re-
gional east Australia during austral winter and spring
(see, e.g., Risbey et al. 2009a). The study presented here
therefore adds to the body of research in this area by
extending the analysis to the finest subdaily time scales,
encompassing ENSO and these large-scale climate
modes, using the long record of daily and subdaily
rainfall datasets as the basis for the analysis.
The remainder of the paper is structured as follows.
In section 2, the data and methodology used to conduct
our analyses in this study are described. Section 3 is re-
served for presenting results on the impact of ENSO on
daily and subdaily rainfall statistics. The effects of the
IOD and the SAM, and their potential influence on the
ENSO–rainfall teleconnection, are explored in section
4, and conclusions are presented in section 5.
2. Data and methodology
a. Rainfall data
Rainfall data for this study comprises daily rainfall
and subdaily (hourly) rainfall records. The daily rain-
fall data is based on a set of high-quality rain gauges
throughout Australia that were identified by Lavery
et al. (1997). For the purpose of this study, only locations
with near-complete records (defined as less than 1%
of days containing missing data) over the period from
1920 and 1999 were used. A total of 88 high-quality
continuous daily rainfall records across Australia met
this criterion, and that the locations of the rainfall stations
provide reasonable spatial coverage for east Australia.
For subdaily rainfall data, a total of 34 rainfall stations
maintained by the Australian Bureau of Meteorology
were used. The number of the stations used is less than
that for the daily rainfall because of missing data at
a number of stations. Our constraint in this case was to
retain the longest possible records while ensuring less
than 15% of the data was missing, with seasons where
data was missing not considered in the analysis. As a
result, these available stations are sparsely distributed
and concentrated over the southeastern region. These
stations have an average record length spanning 54 yr
1666 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
from 1952 to 2005, with each record containing at least
40 yr of data.
b. Climate indices
To examine the effect of the climate modes on the
daily and subdaily rainfall variables, we used a single
index representative of the evolution of each mode. The
Southern Oscillation index (SOI), defined as the anoma-
lous pressure difference between Tahiti and Darwin, is
used as an indicator for ENSO activity. As demonstrated
by Risbey et al. (2009a), rainfall correlations across Aus-
tralia are not sensitive to whether SST-based indices or
SOI are used to represent ENSO variability. This is be-
cause the SOI is highly correlated to the SST-based in-
dices (Barnston et al. 1997). Indeed, our results remain
largely unaltered when the Nino-3.4 index was used. The
dipole mode index (DMI; Saji et al. 1999), which is defined
as the difference in SST anomaly between the western and
eastern tropical Indian Ocean regions, is used to represent
the evolution of the IOD (Rayner et al. 2003). The index of
Fogt et al. (2009) based on the difference between nor-
malized monthly zonal mean sea level pressure at 408 and
658S is used to describe the evolution of the SAM.
c. Multiscale rainfall variables
Rainfall variables derived from the daily and subdaily
data were selected to explain variations occurring at
seasonal time scales. An ENSO event typically initiates
in late austral autumn, peaks in summer, and dissipates
in the autumn of the following year. While the timing of
ENSO impacts on rainfall may vary spatially even in east
Australia, we have primarily focused on variables that
were computed from the winter [June–August (JJA)]
and spring [September–November (SON)] seasons as
they coincide with the maximum impact period for the
majority of stations in east Australia (Chiew et al. 1998).
These seasons also coincide with the period of signifi-
cant correlation between east Australian rainfall and
the IOD, as well as the SAM (e.g., Risbey et al. 2009a;
Hendon et al. 2007), and are thus relevant for assessing
the interaction of impacts among the climate modes.
Here, we derived three variables computed from the
daily rainfall data for each season:
1) Rainfall total (from aggregated daily rainfall),
2) Number of wet days, and
3) Rainfall per wet day.
From the subdaily rainfall data, we derived the following
variables:
4) Number of wet spells,
5) Average wet spell length, and
6) Rainfall intensity per wet spell.
A wet day occurs when daily precipitation is equal to
or exceeds 1 mm as defined by the Australian Bureau
of Meteorology (BOM; available online at http://www.
bom.gov.au/climate/change/about/extremes.shtml). It is
noted that seasonal rainfall total is a function of both
number of wet days and rainfall intensity of each wet day
within the season. The wet days are in turn attributed
by the characteristics of a ‘‘wet spell’’ within each day. A
central assumption pertaining to the subdaily variables
is the definition of a wet spell. Rainfall is often reported
as falling in ‘‘events’’ or ‘‘storms’’ whose beginning and
end are defined by rainless intervals (dry spells) of a
fixed duration minimum interevent time (MIT), with
a wide range of values previously adopted for the MIT
from 15 min to 24 h (Dunkerley 2008). In line with
Tsubo et al. (2005), a 3-h MIT was used here to dis-
tinguish between separate rain events or wet spells to
allow for a compromise between the independence of
rain events and intraevent variability in rain rates. We
considered this MIT to be appropriate in this case
since the focus of this study is not to obtain the most
accurate storm separation threshold, but to perform a
comparative analysis across a vast region with varying
climatic regimes.
d. Impact of climate modes on rainfall variables
To evaluate the impact of the climate modes on the
daily and subdaily rainfall variables, a linear regression
analysis is implemented. The regressed rainfall variables
onto the climate indices are expressed as percentage
change from the corresponding mean values in response
to one standard deviation of the climate index. The im-
pact of interactions between the different climate modes
on rainfall variables is evaluated using multiple linear
regression. Furthermore, to isolate the influence of a cer-
tain climate mode, a partial regression analysis is used
if two climate indices are found to be significantly cor-
related with each other for a particular season. This
widely used technique (see, e.g., Risbey et al. 2009a; Cai
et al. 2011) involves first removing the linear compo-
nent of a given time series (e.g., SOI) from a time series
of interest (e.g., DMI) and a rainfall variable before
performing regression between the two. While many
previous studies involved the use of standard correla-
tion methods, we have elected to apply linear regression
as it enables the ranking of each rainfall variable based
on sensitivity to rainfall through magnitude of percent-
age change. This was used in preference to criteria based
on statistical significance, because practically useful or
important relationships may only be marginally statis-
tically significant, for instance due to a limited sample
size. Conversely, in the case of hydroclimatology re-
search in particular, sensitivities or impacts arising from
MAY 2012 P U I E T A L . 1667
statistically significant relationships can be very small
such that they may be of limited practical value (see
Clarke 2010). Nonetheless, statistical significance is still
reported to provide an indication of a statistically sig-
nificant relationship.
While the linear regression approach is useful, as
previously adopted in various related studies (e.g.,
Nicholls and Kariko 1993; Chiew et al. 1998; Risbey
et al. 2009a), it relies on the assumption of linearity in
the relationships between variables. This assumption
has been found to be reasonable for aggregate sea-
sonal rainfall (e.g., Westra and Sharma 2010). Further-
more, since a comparison of correlations computed based
on the Pearson product moment agrees well with the
nonparametric rank-based correlation performed in
this study, this suggests that our assumption of linearity
is reasonable. As such, for succinctness, much of the
discussions in the next sections will be based with ref-
erence to the La Nina condition, since the linearity en-
sures that changes as a function of El Nino or La Nina are
symmetric.
To specifically examine the nonlinear nature of the
ENSO-driven rainfall impact, a composite analysis was
also conducted. Meyers et al. (2007) conducted a care-
ful study in the classification of years of El Nino and
La Nina using a lagged empirical orthogonal function
analysis. These ENSO years as identified by Meyers
et al. (2007) have been adopted as a basis for our com-
posite analysis. Here, the nonparametric Mann Whitney
U (MWU; Mann and Whitney 1947; Xu et al. 2003) and
Kolmogorov–Smirnov (K–S) tests (Smirnov 1948) are
used. In this case, we use a one-tailed MWU test to infer
whether there is a significant shift in the probability
distribution of each rainfall statistic in a particular di-
rection, at the 95% significance level. The K–S test on
the other hand uses the maximum absolute difference
between the cumulative distribution curves from two
samples to test if the samples come from different
populations. Here, we use the K–S test to ascertain if
rainfall variables from the El Nino and La Nina dis-
tributions are significantly different from each other
at the 95% level. In addition to hypothesis tests, we
also present density estimates of the rainfall variables
such that notable differences, where present, can be
identified via visual inspection. As our study essentially
concerns interannual characteristics of daily/subdaily
rainfall variables arising from the climate modes, linear
trends, although generally small over the entire period
of analysis, were removed from all variables prior to the
above analysis to avoid introducing biases to the corre-
lations between variables. The best-fit estimation of lin-
ear trends from all time series was done via the least
squares method.
3. Daily and subdaily rainfall variability linkedto ENSO
Results from linear regression of the SOI onto the
daily rainfall variables show a significant increase in
winter and spring rainfall totals at nearly all stations
across east Australia as associated with a positive change
in SOI (i.e., a La Nina condition; Figs. 1a,d). This find-
ing is in agreement with previous studies (McBride and
Nicholls 1983; Kane 1997). However, it is particularly
interesting that these changes in rainfall totals, on the
continent scale, are found here to arise more predomi-
nantly from changes in the number of wet days rather
than rainfall intensity per wet day, as revealed in Figs.
1b,c,e,f. Furthermore, regression analysis for the sub-
daily variables reveals that changes in the number of wet
spells are significant across many stations (Figs. 2a,d),
compared to those for the duration and intensity of the
wet spells that, on the other hand, show insignificant
changes (Figs. 2b,c,e,f). This suggests that most of the
changes to the rainfall totals appear to stem from changes
in the number of occurring wet spells.
Note that the direction of changes at nearly all of the
stations is uniform across most of the domain being
analyzed, with the exception of the eastern coastal fringe
where the changes are of a lower magnitude and gen-
erally not statistically significant, although they are of
the same sign (Figs. 1 and 2). This was also found in
earlier work by Timbal (2010), who used this to high-
light the additional difficulty in characterizing variability
across the east Australian coastline. Nevertheless, this
region is only a small part of the study domain, and
given the spatial homogeneity elsewhere, we present
the region wide average percentage change for each
rainfall statistic in Table 1 to aid with interpretation of
the results. The corresponding proportion of number of
stations that are statistically significant is also shown
in Table 1 as an indication of field significance. For
completeness, we also present the results for austral
summer and autumn. With respect to daily rainfall vari-
ables, Table 1 shows a progressively stronger rainfall
impact into the spring season, with a 17% increase in
rainfall per positive standard deviation change in SOI.
It also reveals that changes to the seasonal rainfall total
are in general more apparently driven by changes to the
number of wet days (11% increase) with a much higher
proportion of stations returning statistical significance,
as compared to the less pronounced changes and lower
field significance in the rainfall per wet day statistic.
Table 1 further shows that the only subdaily rainfall
variable to exhibit any significant change is the number
of wet spells across east Australia. It should be noted that
changes to aggregated statistics such as total rainfall
1668 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
amount, number of wet days, and rainfall per wet day
derived using the subdaily rainfall data for the east
Australian region were consistent with the same vari-
ables computed using daily rainfall. This represents a
validation of the quality of the subdaily rainfall data,
and suggests that the lower proportion of statistical sig-
nificance may be explained simply by the shorter record
lengths and increased missing data.
Based on these results, we suggest that the implica-
tions of ENSO on multiscale rainfall change averaged
over the entire east Australian region are in general
driven by changes in the number of wet spells occurring
within a day, rather than their length or intensity. Thus,
these changes in turn explain the changes in the number
of wet days rather than rainfall per wet day. Regional
variations exist whereby either or both the number of
FIG. 1. Changes for daily rainfall variables at individual stations across east Australia corresponding to a 11 standard deviation change
in normalized seasonal SOI. Blue and red circles denote an increase and a decrease in daily rainfall, respectively; while filled circles denote
the statistical significance at the 95% significance level.
MAY 2012 P U I E T A L . 1669
wet days and intensity appear important at a few sta-
tions, possibly due to processes other than ENSO.
Nonetheless, our results appear consistent with those
of Nicholls and Kariko (1993) who found, using only
five daily-rainfall stations, that the Southern Oscilla-
tion primarily influences the number of rain events,
and to a lesser extent their intensity. Examination of
the physical processes behind this differing impact on
the number and intensity of rain events is beyond the
scope of our study. It is likely that, among other things,
stochastic behavior of the air trajectory associated with
rain-producing synoptic systems can play a role, by influ-
encing available moisture relevant for rainfall formation
(Brown et al. 2009).
For the remainder of this section, we further examine
the ENSO-driven rainfall characteristics by focusing on
FIG. 2. Changes for subdaily rainfall variables at individual stations across east Australia corresponding to a 11 standard deviation
change in normalized seasonal SOI. Blue and red circles denote an increase and a decrease in daily rainfall, respectively; while filled circles
denote the statistical significance at the 95% significance level.
1670 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
the composite analysis for the spring season (SON), dur-
ing which the ENSO impact is maximum (see Table 1).
As such, any notable changes as discussed above should
emerge clearly in probability density plots and pass the
MWU and/or K–S statistical tests (see section 2). Prob-
ability density function (PDF) plots of mean-removed
(centered) daily rainfall variables at each station con-
ditional to El Nino and La Nina are presented in Fig. 3,
with the corresponding bold curves indicating the east
Australia–wide averaged response. It can be seen that
the seasonal rainfall total during La Nina years is higher
compared to El Nino years at most stations (Fig. 3a).
These changes can be attributed more apparently to the
number of wet days (Fig. 3b), as noted by the clearer
separation between the two clusters of probability dis-
tributions for the El Nino and La Nina phases, compared
to those for the rainfall per wet day statistic (Fig. 3c). As
a validation statistic, we have introduced a new variable,
the dry spell length defined as number of intervening dry
days between distinct wet days. If there is an increase in
the number of wet days for a fixed time frame (season),
it should logically follow that there will be a correspond-
ing decrease in the length of dry periods, unless these wet
days are consecutive rather than interspersed. Figure 3d
shows that while dry spell lengths are shorter during
La Nina periods, the extent of the separation in the
PDFs between the El Nino and La Nina phases is not
as distinct as that of the number of wet days. This sug-
gests that a fraction of the increase in the number of
wet days at some stations during La Nina is associated
with episodes of consecutive wet days (see also Fig. 5).
The contrasting probability distribution in the num-
ber of wet days is also seen at the subdaily time scale.
As shown in Fig. 4, the subdaily PDF plots show the
clearest separation in the PDFs of number of wet spells,
with an increased and decreased number of wet spells
during the La Nina and El Nino phases, respectively
(Fig. 4a). A curious result is the PDFs of rainfall in-
tensity per wet spell (Fig. 4c), which are not only statis-
tically insignificant between the opposing ENSO phases,
but are also more skewed toward its mean. This result is
consistent with that of Samuel and Sivapalan (2008)
who did not find sufficient evidence (by virtue of the
Student’s t and MWU tests) to conclude that average
storm intensities differed between El Nino and La Nina
phases in both Newcastle and Darwin, both of which
are in east Australia.
It is worth noting that the composite PDFs reveal
La Nina phases as having fatter tails in both the total
rainfall amount and number of wet day statistics than
El Nino phases (Figs. 3a,b). This suggests a certain de-
gree of nonlinearity in the ENSO response, and that
rainfall extremes during La Nina are more intense
and/or frequent. Nonetheless, it should be noted that the
results from linear regression analysis are still in quali-
tative agreement with those of the PDF analysis.
To conclude this section, we provide a continent-wide
perspective of the ENSO influence by presenting the
spatial distribution of the MWU and K–S tests for the
daily rainfall at each station. The MWU tests (at the 95%
significance level) were set out with the null hypothesis
that each rainfall variable for El Nino is not significantly
different from La Nina phases. The corresponding alter-
native hypotheses are as follows:
1) Seasonal rainfall total during El Nino is significantly
lower than during La Nina (one tailed).
2) Number of wet days during El Nino is significantly
lower than during La Nina (one tailed).
3) Rainfall per wet day during El Nino is significantly
different than during La Nina (two tailed).
4) Dry spell lengths during El Nino are significantly
larger than during La Nina (one tailed).
The K–S tests were conducted on the same rainfall
variables to ascertain if their respective distributions
were significantly different at the 95% level. Figure 5
shows the daily rainfall stations returning a p value of
less than 0.05 (i.e., significant; in red) for each rainfall
variable for the MWU (top row) and K–S (bottom row)
tests, as well as statistically insignificant stations (in black).
The MWU test results confirm that the seasonal
rainfall amount at most stations is significantly different
TABLE 1. Average percentage difference corresponding to 11 standard deviation SOI and proportion of stations that are statistically
significant (in parentheses) for each rainfall variable of interest across 88 daily rainfall stations and 34 subdaily rainfall stations in east
Australia for all seasons.
Season
Daily rainfall variables Subdaily rainfall variables
Rainfall
(mm)
No. of
wet days
Rainfall
per wet day
No. of
wet spells
Wet spell length
(h)
Rainfall intensity
per wet spell (mm h21)
DJF 8.9 (0.28) 3.3 (0.17) 5.3 (0.24) 5.6 (0.24) 4.4 (0.15) 0.3 (0.03)
MAM 14.4 (0.51) 8.6 (0.45) 5.9 (0.25) 11.3 (0.38) 2.2 (0.06) 1.8 (0.03)
JJA 18.3 (0.76) 12.5 (0.68) 7.2 (0.31) 10.4 (0.50) 4.9 (0.21) 1.2 (0.03)
SON 16.7 (0.78) 10.8 (0.86) 6.1 (0.33) 12.3 (0.58) 2.5 (0.09) 3.6 (0.12)
MAY 2012 P U I E T A L . 1671
between El Nino and La Nina phases, except at a few
stations along the east coast where local processes in-
cluding orographic effects are more likely to interfere
with ENSO signals. The MWU test for the rest of the
daily rainfall variables shows the number of wet days
exhibiting widespread significance across the region, in
contrast to the rainfall intensity and dry spell length,
which exhibit apparent heterogeneity with a relatively
large proportion of stations showing statistical insig-
nificance. This again confirms the number of wet days
as the determining factor behind seasonal rainfall re-
sponse to ENSO variability across east Australia, con-
sistent with what was revealed from the regression
analysis (Figs. 1d–f).
The K–S tests show that there are significant shifts
in the PDFs of the seasonal rainfall total between the
opposing ENSO phases. These shifts are most pro-
nounced in the inland east region, with most of the
statistically insignificant stations located along the
east coast. The implication of the significance of these
tests for the inland region of east Australia is that
seasonal rainfall response to ENSO does not only in-
volve shifts in the mean, but also the characteristics
of its probability distribution. However, note that the
shifts in the PDFs of the underlying daily rainfall
characteristics paint a more mixed picture than the shifts
in the mean, suggesting that they cannot be solely at-
tributed to number of wet days. Rainfall per wet day as
FIG. 3. Probability density function plots of daily rainfall variables conditioned to El Nino (in red) and La Nina (in
blue) years. Thin lines denote PDFs of mean removed variables for each daily rainfall station, while the thick lines
represent east Australia region averaged values for all stations. Over the period 1920–99, there are 14 and 27 yr
classified as El Nino and La Nina, respectively [see Meyers et al. (2007), their Table 2 for specification of the years].
Thin lines denote PDFs of variable anomalies for each daily rainfall station, while the thick lines represent east
Australia region averaged values for all stations.
1672 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
well as the temporal distribution of the wet days can
also play a role.
Similar tests performed on subdaily rainfall variables
reveal that the changes to rainfall per wet day and the
temporal distribution of wet days appear to be solely
attributed to changes in the number of wet spells, with
approximately 71% (24/34) of subdaily rainfall stations
returning a significant result for the MWU test (spatial
plots not shown here). In particular, this result lends
further support to our argument that changes to rain-
fall per wet day is caused by changes to the number of
within-day wet spells rather than changes to rainfall
intensity.
We next turn our attention to the contribution by the
IOD and the SAM, and how they may interplay with
ENSO and/or exacerbate its impacts on east Australian
rainfall documented thus far.
4. Contribution by the IOD and the SAM
a. IOD
The linear regression analysis on daily rainfall for the
IOD presented in Fig. 6 shows a decrease in seasonal
rainfall that is fairly uniform across east Australia as-
sociated with a positive one standard deviation DMI for
the winter and spring seasons. These differences, par-
ticularly during spring, are driven more predominantly
by changes to the number of wet days rather than rain-
fall per wet day (Fig. 6; see also Table 2). Previous studies
have noted the notably significant correlation between
the IOD and ENSO (e.g., Meyers et al. 2007; Risbey et al.
2009a). The linear correlations during JJA and SON
between the SOI and DMI indices over the time period
of our analysis (1920–99) are 20.45 and 20.60, respec-
tively, both of which are statistically significant. The
FIG. 4. Probability density function plots of subdaily rainfall variables conditioned to El Nino (in red) and La Nina
(in blue) years. Thin lines denote PDFs of mean removed variables for each subdaily rainfall station, while the thick
lines represent east Australia region averaged values for all stations.
MAY 2012 P U I E T A L . 1673
higher correlation in spring is expected, as the IOD
matures during this season before it dissipates in sum-
mer when ENSO peaks. Because of such substantial
IOD–ENSO correlations, it is expected that the IOD
impact on the daily rainfall statistics would be due partly
to co-occurrences with ENSO.
The isolated effect of the IOD can be examined by
removing the component from the DMI time series that
is linearly correlated to ENSO, and then regressing
it onto the time series of the rainfall variables in which
the ENSO coherence has also been removed. Our re-
sults are presented in Table 2 and the figures that follow.
From Table 2 and Fig. 7, a partial IOD (i.e., after re-
moving the effect of ENSO from the IOD) has a much
less significant impact on the daily rainfall variables dur-
ing winter to spring. On the other hand, the partial re-
gression on ENSO reveals that the impact of partial
ENSO (after removing the effect of IOD from ENSO)
appears to be less altered from that without the IOD re-
moved, particularly in JJA since their correlations are
weaker (Tables 1 and 2; Figs. 1 and 8). The effect of
IOD on the ENSO impact is more notable in SON.
Again, these effects generally arise from changes in the
number of wet days rather than rainfall per wet day.
These analyses suggest that ENSO appears to have
stronger direct impact on rainfall over east Australia
in winter than in spring by which time the interaction
with IOD becomes important. This can be explained as
follows. Direct impact of ENSO on northern Australian
rainfall is via atmospheric pressure fluctuations as part of
the Southern Oscillation. In addition, anomalous condi-
tions in the Pacific associated with ENSO drive Rossby
wave trains that propagate into the extratropics, influ-
encing rainfall in the higher latitudes of east Australia
(e.g., Hill et al. 2009; Cai et al. 2011). At the same time,
ENSO tends to influence the evolution of the IOD and
modulate convection in the eastern and western Indian
Ocean that in turn drives poleward-propagating baro-
tropic wave trains affecting rainfall across southern
Australia (Cai et al. 2011). Cai et al. (2011) showed that
FIG. 5. (top) MWU and (bottom) K–S test results for daily rainfall variables across the east Australian region. Red (black) filled circles
denote statistically significant (not significant) results at the 95% level for both the MWU and K–S tests.
1674 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
during winter the wave trains associated with the IOD
emanate only from the eastern Indian Ocean, as ENSO
tends to offset convection in the western counterpart
during the developing IOD. In spring, however, both the
western and eastern Indian Ocean convective regions, in
association with stronger ENSO–IOD coherence, con-
tribute to stronger rainfall effects over the southern parts
of the continent (Cai et al. 2011), by conditioning cutoff
low systems that are an important source of rainfall in
the southeastern Australia (Risbey et al. 2009b). In
addition, our results show that ENSO also impacts on
rainfall intensity per wet day in the southern stations
(Fig. 1f), and that this effect becomes apparent once
the IOD effect is removed (Figs. 8d–f).
b. SAM
The SAM impact in winter shows notable spatial var-
iations (Fig. 9a). Phases of positive SAM as associated
FIG. 6. Changes for daily rainfall variables at individual stations across east Australia corresponding to a 11 standard deviation change
in normalized seasonal DMI. Blue and red circles denote an increase and a decrease in daily rainfall, respectively; while filled circles
denote the statistical significance at the 95% significance level.
MAY 2012 P U I E T A L . 1675
with a poleward shift of the subpolar westerlies, corre-
spond to anomalously dry conditions in the southern
parts of the region, including Tasmania. Rainfall in-
creases are, however, seen in the eastern coastal regions,
which become more widespread and apparent in SON,
in association with anomalous easterly flows that in-
tensify in this season (Hendon et al. 2007). As the SAM
was not found to be significantly correlated with either
IOD or ENSO during our period of analysis, partial
regression analysis is not necessary and thus not per-
formed here. Figure 9 shows that the maximum impact
region and season for SAM is the east coast region in
SON. This is consistent with findings in other studies
that found the east coast of Australia experienced the
largest increases during the positive SAM phase, which
may be explained by anomalous easterly winds enhanc-
ing moisture advection from the ocean and hence in-
creased orographic rainfall in this region (Sen Gupta
and England 2007; Gillett et al. 2006; Hendon et al. 2007).
Conversely, the positive SAM also resulted in rainfall
reductions in the southern part of the region especially
during the JJA period (Cai and Cowan 2006; Meneghini
et al. 2007). Previous studies have attributed this de-
crease to the contraction of the westerly wind belt to-
ward the poles (Hendon et al. 2007). However, despite
the mixed response in total rainfall occasioned by SAM,
it transmits its influence in a manner consistent with the
other climate modes above: that changes, where present,
are more pronounced in number of wet days compared
to rainfall per wet day (Fig. 9).
It should be noted that the way in which the IOD and
the SAM transmit their influence to subdaily rainfall
variables mirrored that of ENSO, with changes, where
present, almost exclusively limited to the occurrence
of subdaily wet spells (results not presented here).
c. Interaction between IOD, SAM, and ENSO
For this section of analysis, we aim to test if inter-
actions between the different climate modes resulted
in enhanced effects on the three daily rainfall variables
described above. A time series of yearly east Australia
normalized rainfall, as well as the SOI, DMI, and the
SAM index is presented in Fig. 10 as a visual aid. We first
tested for changes corresponding to positive one and
negative one standard deviation of SOI and DMI, re-
spectively, using multiple linear regression (i.e., the com-
bination that is expected to result in wetter conditions).
We then paired positive one standard deviation of
SOI and SAM index (again, based on the direction of
change in index that produced coherent impact on
rainfall variables) before including all three climate
indices as predictors into the model. The results are
summarized in Table 3, suggesting that the effect of
stratifying rainfall variables to multiple predictors did
not always translate to an additive effect to rainfall
variables. In particular, the ENSO–IOD pairing resulted
in minimal increase in magnitude of changes compared
to ENSO alone. This result was not surprising consid-
ering that they are correlated. With respect to the SOI–
SAM pairing, rainfall during spring season increased
significantly, with up to 30% change in total seasonal
rainfall amount that can be attributed to 18% increase in
number of wet days, and 12% increase in rainfall per wet
day (Table 3). Note that the effects of the SAM and
ENSO are offsetting in the southern region during win-
ter. However, when all three climate modes were in-
cluded as predictors, no significant new information
could be gleaned suggesting that interactions between
these modes resulted in only minor changes to the dy-
namics of rainfall in east Australia.
5. Conclusions
In this study, we have explored daily to subdaily rain-
fall variability at several sites across east Australia. The
main findings from our analyses can be summarized as
follows:
1) Changes to seasonal rainfall amounts associated with
ENSO variability over east Australia as a whole,
particularly during the cool seasons, can be primarily
attributed to changes in the number of wet days and
TABLE 2. Average percentage difference corresponding to 11 standard deviation of DMI, SAM in isolation, as well as that of partial
SOI and partial DMI, and the proportion of stations that are statistically significant (in parentheses) for each rainfall variable of interest
across 88 daily rainfall stations.
Season Variable 11 std dev DMI 11 std dev SAM 11 std dev partial SOI 11 std dev partial DMI
JJA Rainfall tot 28.7 (0.39) 20.3 (0.34) 17.9 (0.67) 21.8 (0.00)
No. of wet days 24.7 (0.22) 0.2 (0.53) 12.6 (0.56) 0.2 (0.03)
Rainfall per wet day 24.3 (0.18) 20.6 (0.13) 5.8 (0.18) 22.0 (0.03)
SON Rainfall tot 210.9 (0.46) 12.7 (0.53) 17.0 (0.58) 21.9 (0.08)
No. of wet days 27.8 (0.52) 7.7 (0.58) 10.3 (0.48) 22.4 (0.09)
Rainfall per wet day 22.9 (0.18) 5.2 (0.28) 7.1 (0.25) 1.1 (0.07)
1676 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
to a lesser extent, rainfall per wet day. Analysis of
subdaily rainfall data has further established that
changes to rainfall per wet day are in fact caused by
changes to the number of wet spells per day, and not
rainfall intensity per wet spell suggesting that it is the
probability of rainfall occurrence, rather than the
character of rainfall given that it has occurred, which
is the dominant mechanism driving seasonal rainfall
changes.
2) The impacts of ENSO and IOD in combination are
fairly spatially homogenous on the continent scale.
Spatial variations do nonetheless exist, with ENSO
and IOD more strongly impacting the inland east and
southeast regions, respectively. The SAM’s influ-
ence on rainfall is strongest over the east coast region
during SON, but is associated with a more mixed
rainfall response during JJA, with the southern part
of the region experiencing opposite effects to the
FIG. 7. Changes for daily rainfall variables at individual stations across east Australia corresponding to a 1 1 standard deviation change
in normalized seasonal DMI with the SOI signal removed. Blue and red circles denote an increase and a decrease in daily rainfall,
respectively; while filled circles denote the statistical significance at the 95% significance level.
MAY 2012 P U I E T A L . 1677
northern part. Despite having different maximum
impact regions and seasons, the manner in which
IOD and SAM impact on rainfall is somewhat similar
to that of ENSO, in that changes to the number of
wet days is overall more significant than rainfall per
wet day.
3) ENSO has been found to be the primary driver of
climate variability over east Australia during the long
period of analysis (1920–99). While combination of
the modes did occasionally produce additive effects
such as the ENSO–SAM in SON, we found signif-
icant redundancy in the effects of climate modes in
combination perhaps as a result of significant cor-
relation with each other (such as the ENSO–IOD
pairing).
Importantly, we have shown that insight into sub-
daily rainfall characteristics may help explain changes
FIG. 8. Changes for daily rainfall variables at individual stations across east Australia corresponding to a 11 standard deviation change
in normalized seasonal SOI with the DMI signal removed. Blue and red circles denote an increase and a decrease in daily rainfall,
respectively; while filled circles denote the statistical significance at the 95% significance level.
1678 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
occurring at coarser time steps as well as assist with
making some initial inferences into the physical mech-
anisms behind climate-mode-driven rainfall. Because
rainfall intensity per wet spell remains fairly constant
under the influence of climate modes, we suggest that
the nature of the rainfall generating mechanism (i.e.,
stratiform vs convective) does not differ significantly as
a result of climate mode influence. Rather, it is the fre-
quency of rainfall events that are subject to change. It is
interesting that teleconnections appear to influence the
probability rather than the intensity of events when they
occur. The processes by which each of the modes trans-
mits their influence on rainfall characteristics warrant
further investigation.
It should be noted that the influence of climate modes
on rainfall characteristics over east Australia, and the
relationships between the climate modes, can be mod-
ulated by lower-frequency modes such as the IPO (e.g.,
FIG. 9. Changes for daily rainfall variables at individual stations across east Australia corresponding to a 11 standard deviation change
in normalized seasonal SAM. Blue and red circles denote an increase and a decrease in daily rainfall, respectively; while filled circles
denote the statistical significance at the 95% significance level.
MAY 2012 P U I E T A L . 1679
Cai et al. 2010; Kiem et al. 2003), as well as the warming
trend. Such a modulation effect is not investigated in
our study as the long record analyzed here captures both
phases of the IPO, and as such biases due to the records
representing a single phase alone are not expected. Our
analysis is, however, useful for inferring a generalized
role of the climate modes on rainfall characteristics.
The effects of the IPO and long-term trends should
nonetheless be investigated in future studies, in particu-
lar, concerning the possibility that the climate modes
may influence finescale rainfall characteristics in a dif-
ferent way in a warming climate.
Finally, we wish to note a number of practical impli-
cations of the results presented here. The overriding
importance of ENSO as the primary driver of rainfall
variability in east Australia was emphasized, with per-
centage changes between around 16%–18% per stan-
dard deviation of the SOI during the winter and spring
seasons, and slightly lower but also reasonably signifi-
cant increases for the remaining seasons; this highlights
that this mode of variability can have profound impli-
cations on flood risk and water resource availability.
This is particularly the case in regions away from the
eastern coastal fringe. The implications on flood risk are
therefore less likely to be due to changes in the intensity
of the flood-producing rainfall event during ENSO, with
a related study recently also showing only a 3.2% and
4.6% change in the intensity of the annual maximum
rainfall event pre standard deviation of the SOI for
subdaily and daily rainfall, respectively (Westra and
Sisson 2011). Rather, the increased number of wet spells
during La Nina periods suggests that it is likely to be the
catchment wetness (also referred to as the ‘‘antecedent
moisture’’) prior to the rainfall-producing flood event
that is most likely to increase, highlighting that the joint
probability of extreme rainfall and antecedent condi-
tions need to be modeled in order to properly capture
the implications of ENSO on flood risk (Kuczera et al.
2006). Analogous implications can be seen for other wa-
ter resources applications, with infiltration properties
and evaporation losses likely to be different for a larger
number of less intense events compared with a smaller
number of more intense events.
In conclusion, while the stochasticity of the climate
system prevents exact sequence prediction of weather
events within a particular season, we have shown that
it is possible to obtain useful information on shifts
in the important rainfall statistics using simple tech-
niques, even if there is still variability that remains
unexplained.
FIG. 10. Time series of normalized yearly east Australia aggregated rainfall, along with
normalized climate indices of SOI (blue), DMI (red), and SAM (green), respectively. The blue
circles are attached to the SOI indices when the magnitude of rainfall anomaly was greater than
one standard deviation. It is useful to note that the El Nino years encompass 1923, 1925, 1930,
1940, 1941, 1957, 1963, 1965, 1972, 1982, 1986, 1987, 1991, and 1997; and La Nina years en-
compass 1922, 1924, 1926, 1928, 1933, 1935, 1938, 1942, 1944, 1945, 1946, 1949, 1950, 1954, 1955,
1961, 1964, 1967, 1970, 1971, 1973, 1978, 1981, 1983, 1988, 1996, and 1998.
TABLE 3. Average percentage difference corresponding to the co-occurrence of different climate modes using multiple linear regression
and proportion of stations that are statistically significant (in parentheses) for each rainfall variable of interest across 88 daily rainfall
stations. Note use of 21 standard deviation DMI to highlight positive change in rainfall amount occasioned by 11 standard deviation SOI.
Season Variable 11 SOI, 21 DMI 11 SOI, 11 SAM 11 SOI, 21 DMI, 11 SAM
JJA Rainfall tot 18.4 (0.74) 20.3 (0.86) 20.3 (0.80)
No. of wet days 11.2 (0.56) 13.0 (0.84) 10.0 (0.83)
Rainfall per wet day 8.4 (0.32) 7.5 (0.42) 6.4 (0.32)
SON Rainfall tot 16.7 (0.72) 29.7 (0.93) 29.2 (0.92)
No. of wet days 10.9 (0.66) 18.4 (0.86) 17.7 (0.83)
Rainfall per wet day 6.0 (0.26) 11.8 (0.49) 10.5 (0.43)
1680 M O N T H L Y W E A T H E R R E V I E W VOLUME 140
Acknowledgments. This study was financially sup-
ported by the Australian Research Council. The rainfall
data used was made available by the Australian Bureau
of Meteorology.
REFERENCES
Barnston, A. G., and R. E. Livezey, 1987: Classification, season-
ality, and persistence of low-frequency atmospheric circula-
tion patterns. Mon. Wea. Rev., 115, 1083–1126.
——, M. Chelliah, and S. B. Goldenberg, 1997: Documentation of
a highly ENSO-related SST region in the equatorial Pacific.
Atmos.–Ocean, 35, 367–383.
——, M. H. Glantz, and Y. X. He, 1999: Predictive skill of statistical
and dynamical climate models in SST forecasts during the
1997/98 El Nino episode and the 1998 La Nina onset. Bull.
Amer. Meteor. Soc., 80, 217–243.
Bigg, G. R., T. D. Jickells, P. S. Liss, and T. J. Osborn, 2003: The
role of the oceans in climate. Int. J. Climatol., 23, 1127–
1159.
Brown, J. N., P. C. McIntosh, M. J. Pook, and J. S. Risbey, 2009:
An investigation of the links between ENSO flavors and
rainfall processes in southeastern Australia. Mon. Wea. Rev.,
137, 3786–3795.
Cai, W., and T. Cowan, 2006: SAM and regional rainfall in IPCC
AR4 models: Can anthropogenic forcing account for south-
west Western Australian winter rainfall reduction? Geophys.
Res. Lett., 33, L24708, doi:10.1029/2006GL028037.
——, P. van Rensch, T. Cowan, and A. Sullivan, 2010: Asymmetry
in ENSO teleconnection with regional rainfall, its multidecadal
variability, and impact. J. Climate, 23, 4944–4955.
——, ——, ——, and H. H. Hendon, 2011: Teleconnection path-
ways of ENSO and the IOD and the mechanisms for impacts
on Australian rainfall. J. Climate, 24, 3910–3924.
Cayan, D. R., K. T. Redmond, and L. G. Riddle, 1999: ENSO and
hydrologic extremes in the western United States. J. Climate,
12, 2881–2893.
Chiew, F. H. S., T. C. Piechota, J. A. Dracup, and T. A. McMahon,
1998: El Nino–Southern Oscillation and Australian rainfall,
streamflow and drought: Links and potential for forecasting.
J. Hydrol., 204, 138–149.
Clarke, B., 2010: Environmental decision-making: Exploring
complexity and practice. Geogr. Res., 48, 448–449.
Drosdowsky, W., 1993: An analysis of Australian seasonal rainfall
anomalies—1950–1987.1. Spatial patterns. Int. J. Climatol., 13,
1–30.
Dunkerley, D., 2008: Identifying individual rain events from
pluviograph records: A review with analysis of data from an
Australian dryland site. Hydrol. Processes, 22, 5024–5036.
Fogt, R. L., J. Perlwitz, A. J. Monaghan, D. H. Bromwich, J. M.
Jones, and G. J. Marshall, 2009: Historical SAM variability.
Part II: Twentieth-century variability and trends from recon-
structions, observations, and the IPCC AR4 models. J. Climate,
22, 5346–5365.
Gershunov, A., and T. P. Barnett, 1998: Interdecadal modulation
of ENSO teleconnections. Bull. Amer. Meteor. Soc., 79, 2715–
2725.
Gillett, N. P., T. D. Kell, and P. D. Jones, 2006: Regional climate
impacts of the Southern Annular Mode. Geophys. Res. Lett.,
33, L23704, doi:10.1029/2006GL027721.
Hendon, H. H., D. W. J. Thompson, and M. C. Wheeler, 2007:
Australian rainfall and surface temperature variations associated
with the Southern Hemisphere annular mode. J. Climate, 20,
2452–2467.
Hill, K. J., A. Santoso, and M. H. England, 2009: Interannual
Tasmanian rainfall variability associated with large-scale cli-
mate modes. J. Climate, 22, 4383–4397.
Kane, R. P., 1997: Relationship of El Nino–Southern Oscillation
and Pacific sea surface temperature with rainfall in various
regions of the globe. Mon. Wea. Rev., 125, 1792–1800.
Kiem, A. S., S. W. Franks, and G. Kuczera, 2003: Multi-decadal
variability of flood risk. Geophys. Res. Lett., 30, 1035,
doi:10.1029/2002GL015992.
Kuczera, G. A., M. F. Lambert, T. M. Heneker, S. A. Jennings,
A. Frost, and P. Coombes, 2006: Joint probability and design
storms at the crossroads. Aust. J. Water Resour., 10 (1), 63–79.
Lavery, B., G. Joung, and N. Nicholls, 1997: An extended high-
quality historical rainfall dataset for Australia. Aust. Meteor.
Mag., 46, 27–38.
Liebmann, B., C. Jones, and L. M. V. de Carvalho, 2001: Inter-
annual variability of daily extreme precipitation events in the
state of Sao Paulo, Brazil. J. Climate, 14, 208–218.
Mann, H. B., and D. R. Whitney, 1947: On a test of whether one of
2 random variables is stochastically larger than the other. Ann.
Math. Stat., 18, 50–60.
McBride, J. L., and N. Nicholls, 1983: Seasonal relationships be-
tween Australian rainfall and the Southern Oscillation. Mon.
Wea. Rev., 111, 1998–2004.
Meneghini, B., I. Simmonds, and I. N. Smith, 2007: Association
between Australian rainfall and the Southern Annular Mode.
Int. J. Climatol., 27, 109–121.
Meyers, G., P. McIntosh, L. Pigot, and M. Pook, 2007: The years of
El Nino, La Nina, and interactions with the tropical Indian
Ocean. J. Climate, 20, 2872–2880.
Nicholls, N., 1997: Developments in climatology in Australia:
1946–1996. Aust. Meteor. Mag., 46, 127–135.
——, and A. Kariko, 1993: East Australian rainfall events: Inter-
annual variations, trends, and relationships with the Southern
Oscillation. J. Climate, 6, 1141–1152.
Pittock, A. B., 1975: Climatic change and patterns of variation in
Australian rainfall. Search, 6, 498–504.
Pui, A., A. Lal, and A. Sharma, 2011: How does the Interdecadal
Pacific Oscillation affect design floods in Australia? Water
Resour. Res., 47, W05554, doi:10.1029/2010WR009420.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V.
Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003:
Global analyses of sea surface temperature, sea ice, and night
marine air temperature since the late nineteenth century.
J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Risbey, J., M. Pook, P. McIntosh, M. C. Wheeler, and H. H.
Hendon, 2009a: On the remote drivers of rainfall variability in
Australia. Mon. Wea. Rev., 137, 3233–3253.
——, ——, ——, C. Ummenhofer, and G. Meyers, 2009b: Char-
acteristics and variability of synoptic features associated with
cool season rainfall in southeastern Australia. Int. J. Climatol.,
29 (11), 1595–1613.
Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional
scale precipitation patterns associated with the El Nino–
Southern Oscillation. Mon. Wea. Rev., 115, 1606–1626.
——, and M. A. Bell, 2008: Shifts in the statistics of daily rainfall in
South America conditional on ENSO phase. J. Climate, 21,
849–865.
Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata,
1999: A dipole mode in the tropical Indian Ocean. Nature,
401, 360–363.
MAY 2012 P U I E T A L . 1681
Samuel, J. M., and M. Sivapalan, 2008: A comparative model-
ing analysis of multiscale temporal variability of rainfall
in Australia. Water Resour. Res., 44, W07401, doi:10.1029/
2007WR006373.
Sen Gupta, A., and M. H. England, 2007: Coupled ocean–atmosphere–
ice response to variations in the Southern Annular Mode.
J. Climate, 20, 3677–3692.
Smirnov, N., 1948: Table for estimating the goodness of fit of em-
pirical distributions. Ann. Math. Stat., 19, 279–279.
Timbal, B., 2010: The climate of the eastern seaboard of Australia:
A challenging entity now and for future projections. IOP
Conf. Ser.: Earth Environ. Sci., 11, 012013, doi:10.1088/1755-
1315/11/1/012013.
Tsubo, M., S. Walker, and M. Hensley, 2005: Quantifying risk for
water harvesting under semi-arid conditions. Part I. Rainfall
intensity generation. Agric. Water Manage., 76, 77–93.
Westra, S., and A. Sharma, 2006: Dominant modes of interannual
variability in Australian rainfall analyzed using wavelets.
J. Geophys. Res., 111, D05102, doi:10.1029/2005JD005996.
——, and ——, 2010: An upper limit to seasonal rainfall pre-
dictability? J. Climate, 23, 3332–3351.
——, and S. A. Sisson, 2011: Detection of nonstationarity in pre-
cipitation extremes using a max-stable process model. J. Hy-
drol., 406, 119–128.
Xu, Z. X., K. Takeuchi, and H. Ishidaira, 2003: Monotonic trend and
step changes in Japanese precipitation. J. Hydrol., 279, 144–150.
1682 M O N T H L Y W E A T H E R R E V I E W VOLUME 140