Over the past century, particularly after the 1960s, observations of mean maximum temperatures reveal an increasing trend over the southeastern quadrant of the Australian continent. Correlation analysis of seasonally averaged mean maximum temperature anomaly data for the period 1958–2012 is carried out for a representative group of 10 stations in southeast Australia (SEAUS). For the warm season (November–April) there is a positive relationship with the El Niño–Southern Oscillation (ENSO) and the Pacific decadal oscillation (PDO) and an inverse relationship with the Antarctic Oscillation (AAO) for most stations. For the cool season (May–October), most stations exhibit similar relationships with the AAO, positive correlations with the dipole mode index (DMI), and marginal inverse relationships with the Southern Oscillation index (SOI) and the PDO. However, for both seasons, the blocking index (BI, as defined by M. Pook and T. Gibson) in the Tasman Sea (160°E) clearly is the dominant climate mode affecting maximum temperature variability in SEAUS with negative correlations in the range from r = −0.30 to −0.65. These strong negative correlations arise from the usual definition of BI, which is positive when blocking high pressure systems occur over the Tasman Sea (near 45°S, 160°E), favoring the advection of modified cooler, higher-latitude maritime air over SEAUS.
A point-by-point correlation with global sea surface temperatures (SSTs), principal component analysis, and wavelet power spectra support the relationships with ENSO and DMI. Notably, the analysis reveals that the maximum temperature variability of one group of stations is explained primarily by local factors (warmer near-coastal SSTs), rather than teleconnections with large-scale drivers.
Climate variability over Australia is affected by several large-scale drivers, some with larger impacts than others over specific region(s) of the continent. In the last four decades, a plethora of studies have identified the following climate drivers as playing a key role on the global climate over Australia: the Antarctic Oscillation (AAO) or southern annular mode (SAM) (Trenberth 1979; Rogers and van Loon 1982; Karoly 1990; Gong and Wang 1999; Thompson and Solomon 2002; Nan and Li 2003; Benjamin and Fauchereau 2012); the El Niño–Southern Oscillation (ENSO) (Bjerknes 1969; Philander 1990); ENSO-related standard indices such as Niño-1–4 and El Niño Modoki (Ashok et al. 2007; Weng et al. 2007; Ashok and Yamagata 2009); the dipole mode index (DMI) (Saji et al. 1999; Webster et al. 1999); the Madden–Julian oscillation (MJO) (Wheeler and Hendon 2004; Wheeler et al. 2009); and atmospheric blocking (Coughlan 1983; Pook and Gibson 1999; McIntosh et al. 2008; Risbey et al. 2009). The impacts of these climate drivers on rainfall variability have been investigated extensively over the past two decades, owing to observed large changes in rainfall trends, particularly after the 1970s (e.g., Walker 1923; McBride and Nicholls 1983; Ashok et al. 2007; Risbey et al. 2009; Cai et al. 2011a). Those studies augmented our knowledge of some of the physical mechanisms determining the observed long-term rainfall variability. In contrast, to the authors’ knowledge, there have been very few observational studies dedicating their attention on equivalent relationships with seasonal-mean maximum temperature variability (Hendon et al. 2007; Hunter and Binyamin 2012; not to be confused with heat wave events).
Since the 1960s, there has been a positive trend (increase) in mean maximum temperatures over much of Australia, particularly in the southwest and the southeast of the continent (Fig. 1). During the same period, decadal-scale droughts (e.g., Timbal and Murphy 2007; Murphy and Timbal 2008; Timbal 2009, 2010; Verdon-Kidd and Kiem 2009; Timbal and Fawcett 2013) also have affected the southeast and the southwest, both of which have been studied extensively owing to potential long-term freshwater deficits in these populous areas. Although, to the authors' knowledge, there have been very few studies addressing the existence of potential links between rainfall and mean maximum temperature across different regions of Australia, it is likely that both trends (i.e., increased temperatures and reduced rainfall) could be related (Hendon et al. 2007). Consistent with persistent negative trends in rainfall over southeast Australia (SEAUS) and positive trends in mean maximum temperature, this region has also experienced a progressive increase in bushfire activity and property loss (McAneney et al. 2009), particularly in the last decade. Some modeling studies have hypothesized that this observed progressive warming could be attributed to global warming induced by anthropogenic sources (e.g., Karoly and Braganza 2005). This progressive warming and concomitant drying trends would also be consistent with possible increased dustiness over eastern Australia over the next decades (e.g., Lamb et al. 2009).
Past seminal studies have shown that, over the Australian continent, two main drivers account for much of the interannual variability in tropospheric temperatures. The first is ENSO (e.g., Angell 1981; Halpert and Ropelewski 1992; Trenberth et al. 2002), particularly in the tropics (e.g., Newell and Weare 1976; Deng et al. 2012). More generally, Trenberth et al. (2002) found that, during 1950–98, ENSO linearly accounts for 0.06°C of global surface temperature increase with this warming lagging SST increases in the Niño-3.4 region by 3 months (consistent with Jones 1989). The second driver is associated with the development of a stationary high pressure cell in the subtropical ridge in the vicinity of eastern Australia. This synoptic pattern has long been recognized as contributing to the weakening of cold fronts and, hence, for inducing large positive temperature anomalies over two-thirds of southern Australia (and thus over SEAUS; Pezza et al. 2012; Sadler et al. 2012). The blocking index as defined by Pook and Gibson (1999), however, is positive when blocking occurs well south of the subtropical ridge in the Tasman Sea (near 45°S, 160°E) and, as will be shown in this study, is associated with negative trends in mean maximum temperature anomalies over SEAUS.
Following the approach of Fierro and Leslie (2013), the present study analyzes mean maximum temperature variability for the 1958–2012 period over SEAUS through wavelet analysis, bootstrap correlations, partial correlations, principal component analysis, and point-by-point correlation analysis of global SSTs. The climate drivers investigated herein are the ENSO-based indices including the Pacific decadal oscillation (PDO) (Mantua et al. 1997; Mantua and Hare 2002; publicly available at http://jisao.washington.edu/pdo/PDO.latest), atmospheric blocking as measured by the blocking index (BI) at 160°E following the definition of Wright (1974) and Pook and Gibson (1999), the SAM index (SAMI) (Nan and Li 2003), and the DMI (Saji et al. 1999).
2. Data and methodology
The monthly mean maximum temperature time series in this study were obtained from the Australian Bureau of Meteorology (BoM) public domain website (http://www.bom.gov.au/climate/data/index.shtml). The updated and homogenized temperature records of carefully selected stations are well suited for monitoring long-term trends and variability (Torok and Nicholls 1996; Della-Marta et al. 2004). The location map of all 10 stations employed in this analysis is presented in Fig. 2. Additional specifics relevant to those stations are provided in Table 1. Those stations were carefully chosen to minimize missing data during the analysis period (1958–2012). Data after April 2012 were not included owing to pending quality-control assessment by BoM for some of the stations. The very few missing data of the final compiled dataset, less than 0.1% of the total data, were simply filled by their respective averages over the whole time period, as in Fierro and Leslie (2013). It is important to highlight the existence of other (perhaps superior) temperature records for climate monitoring over Australia such as the Australian Climate Observations Reference Network Surface Air Temperature dataset (ACORN-SAT) (Trewin 2012). Because this study (like that of Fierro and Leslie 2013) aims at using a handful of representative stations, each characterized by a long record history, instead of a spatial dataset compiled from a dense network of hundreds of stations, the use of the BoM data herein is reasonable. Most importantly, recent comparisons of the ACORN and BoM temperature datasets by the Centre for Australian Weather and Climate Research (CAWCR) showed very similar estimates of Australian temperature changes over the twentieth century with the mean minimum temperature data exhibiting the largest difference (warming): this primarily was since the 1960s (Fawcett et al. 2012).
Following the methodology of Fierro and Leslie (2013), the data first are correlated with the driver indices listed at the end of the introduction. The ENSO-based indices considered herein comprise the Southern Oscillation index (SOI) (Troup 1965; Trenberth 1984) and of the standard SST-based indices: Niño-1+2 (10°S–0°, 90°–80°W), Niño-3 (5°S–5°N, 150°–90°W), Niño-3.4 (5°S–5°N, 170°–120°W), Niño-4 (5°S–5°N, 160°E–150°W), and El Niño Modoki (Ashok et al. 2007; Weng et al. 2007; Ashok and Yamagata 2009). The updated Modoki index was provided by the Japan Agency for Marine-Earth Science and Technology (http://www.jamstec.go.jp; courtesy of Dr. W. Sasaki). Following Saji et al. (1999), the intensity of the dipole mode index is defined as the difference between the SSTs in the western equatorial Indian Ocean (50°–70°E, 10°S–10°N) and the southeastern equatorial Indian Ocean (90°–110°E, 10°S–0°). The DMI was computed using the areal averaged monthly Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) dataset (http://www.jamstec.go.jp/). With the exception of Modoki, all SST-based ENSO indices are made publicly available by the National Centers for Environmental Prediction (NCEP; http://www.cpc.ncep.noaa.gov/data/indices) and were calculated using the global National Oceanic and Atmospheric Administration (NOAA) extended reconstructed SST (ERSST) version 3b (V3b) dataset (Smith et al. 2008). This study uses SAMI [available online at http://www.lasg.ac.cn/english/], as defined in Nan and Li (2003), which is a modification of the AAO index defined by Gong and Wang (1999). For simplicity, SAMI will simply be referred to as SAM in the remainder of the manuscript. Last, BI is calculated using wind data from the NCEP–National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996). Only the anomalies of all the aforementioned indices were considered herein.
During this analysis, additional drivers other than those listed above were examined but were shown to exert insignificant to marginal influence over SEAUS temperatures and thus were not explored further. Such indices are the 30- and 50-hPa quasi-biennial oscillation indices (QBO) (Lindzen and Holton 1968; publicly available at http://www.cpc.ncep.noaa.gov/data/indices/). It is noted that indices based on upper-level wind records such as BI and QBO are less reliable in the earlier decades (1960s and early 1970s). In addition, it was decided not to assess the role of the MJO (and its eight distinctive phases, Wheeler and Hendon 2004; Wheeler et al. 2009) because this study primarily focuses on large-scale climate drivers beyond seasonal time scales.
The correlations were carried out on detrended, semiannual (6 months) averaged maximum temperature anomalies for the SEAUS warm season [November–April (NOVAPR)] and cold season [May–October (MAYOCT)]. The anomalies are computed relative to the 30-yr period beginning 1 January 1981. (http://www.esrl.noaa.gov/psd/data/gridded/newnormals.html). It is relevant to note that the analysis presented in this work was also carried out using the standard (4) seasonal decomposition and revealed that MAYOCT (NOVAPR) was well representative of June–August (JJA) [December–February (DJF)]. The correlation analysis employs a bootstrap sampling technique with 1000 random samples to determine the statistical significance of the correlations (e.g., Wilks 2006).
Similar to Fierro and Leslie (2013), two correlation methods were investigated owing to the high variability inherent in the temperature data analyzed here. The first technique is the standard Pearson correlation, and the second is the Spearman “rank” method (e.g., Wilks 2006). The results showed only a negligible difference between the correlations obtained from either method and, consequently, only the correlations using the Pearson method (simply referred here to as correlations) are shown.
Past studies documented the existence of an interplay between the ENSO and DMI phases for Australian rainfall (Power et al. 1999; Meyers et al. 2007; Risbey et al. 2009; Cai et al. 2011a,b,c) and, therefore, potentially on mean maximum temperatures. In the present dataset, the correlation between DMI and SST-based ENSO indices ranged between r = 0.35 and 0.45 (SOI: −0.53) for MAYOCT but showed, in contrast, insignificant correlation in NOVAPR (r < 0.12). Consequently, only MAYOCT partial correlations are explored for ENSO and DMI. Additionally, it was found that the PDO correlates moderately (r = 0.4–0.56) with all ENSO-based indices and this during both semiannual periods herein [consistent with Verdon and Franks (2006)]. Hence, partial correlations for ENSO and the PDO were also analyzed.
Given the highly nonlinear nature of the coupled ocean–atmosphere system, correlations do not demonstrate the existence of direct causal links. Nonetheless, correlations often prove to be very useful for indicating patterns, such as dipoles, in these potential links, which might be related to identifiable physical mechanism(s).
As a second step, the detrended semiannual (6 months) averaged mean maximum temperature anomalies each were correlated with the Smith et al. (2008) ERSST V3b dataset (publicly available at http://www.cdc.noaa.gov/cdc/data.noaa.ersst.html) to establish potential relationships with ocean-driven climate drivers. Owing to the well-established dominant roles of ENSO and DMI in modulating Australian climate, this analysis specifically focuses on the Pacific Ocean (ENSO and PDO) and the Indian Ocean (DMI), potentially to determine further links between those SST fields and mean maximum temperature anomalies. The SST correlation pattern will be evaluated against a principal component analysis conducted on those same SSTs similar to Fierro and Leslie (2013). It is noteworthy that, to better identify potential loading patterns associated with either ENSO or the DMI in the SST data, the annual cycle was excluded from the SST dataset prior to performing the principal component analysis. This accounts for why the variance explained by the leading principal components of the filtered SST data does not exceed 30%.
Finally, similar to Fierro and Leslie (2013) and Speer et al. (2011), a wavelet analysis (Klees and Haagmans 2000; Meyers et al. 1993; Lau and Weng 1995) is carried out on the 6-month (non-detrended) averaged data (not anomalies) for each of the 10 stations shown in Fig. 2. Wavelet power spectra provide information about the time evolution of the amplitude (and hence the periodicity) of a signal existing within a time series. The time series are not detrended prior to the wavelet analysis because a trend is an inherent part of the amplitude of a signal. The global power spectrum, which is equivalent to applying a simple Fourier transform to the time series, also is shown to provide a global view of the dominant amplitudes/wavelength embedded in the time series. As in Fierro and Leslie (2013) and Speer et al. (2011), open source code provided by the Department of Atmospheric and Oceanic Science of the University of Colorado (http://paos.colorado.edu/research/wavelets/) is used in this study. It is relevant to highlight, once more, that this study focuses on longer-term maximum temperature anomalies and not on individual heat wave events. The definition of heat waves remains controversial in the current literature [see the review by Perkins and Alexander (2013)] and heat waves are not part of this study, even though heat waves might contain many of the mean monthly positive maximum temperature anomalies in the time series studied here.
3. Temperature variability and possible teleconnections
Before establishing possible links between mean maximum temperature anomalies and large-scale climate drivers over SEAUS, it is useful to provide an overview of the temperature records of the selected stations (Fig. 2) for both semiannual seasons considered in this study: namely, NOVAPR and MAYOCT. As shown in this section, these were selected because NOVAPR (MAYOCT) is very representative of the warmer (cooler) period over SEAUS. Unless deemed necessary for the sake of clarity, seasonally averaged maximum temperature anomalies will be referred to simply as temperature in the remainder of the manuscript.
The contrast in temperatures between NOVAPR and MAYOCT can be clearly seen in Figs. 3 and 4, respectively, with almost all stations exhibiting an unequivocal increasing trend in temperature for both seasons over the last 60 years. One station (Wilcannia, Fig. 3e) shows no identifiable trend in temperature during NOVAPR. Another station, Nyngan, has only a marginal positive trend in MAYOCT (Fig. 4d). Consistent with Fig. 1, most stations exhibit an ~1°C increase in temperature during the 60-yr period considered here, which corresponds to an increase of ~0.17°C decade−1. The largest temperature increases are observed during MAYOCT for Melbourne and Wagga Wagga, ~1.5° or ~0.25°C decade−1 (Figs. 4i,j). Some of the increases in mean maximum temperature seen for the larger metropolitan areas of Melbourne (Figs. 3j and 4j) and Sydney (Figs. 3g and 4g) could possibly be attributed to the “urban heat island” effect (e.g., Morris et al. 2001).
a. Correlation analysis
The correlations in Figs. 5–7 are statistically significant when the 2.5%–97.5% percentile interval is strictly below or above the zero line. During NOVAPR, four stations [Tibooburra (TI), Walgett (WA), Nyngan (NY), and Wilcannia (WI)], which geographically are near the center of SEAUS, show statistically significant negative correlations with SAM at values ranging between r = −0.38 and −0.48 (Fig. 5a). This particular group of stations hereafter will be referred to as G1. This inverse relationship with SAM could potentially be explained by the presence of a surface high pressure anomaly near 40°S, following the definition of Nan and Li (2003), that would favor the advection of modified cooler higher-latitude maritime air northward toward SEAUS. The DMI showed overall insignificant correlations with the temperature, the only exception being Melbourne (ME) (Fig. 5b). Statistically significant (inverse) relationships were found between SOI and most stations, especially with G1 (Fig. 5c). In line with this result and consistent with El Niño years being warmer for some stations, the temperature data for some stations (particularly G1) also exhibit significant positive correlations with each of the SST-based ENSO indices (Niño-1+2, Niño-3, Niño-4, and Modoki; Figs. 5d–g) and Niño-3.4 (not shown). The consistency of the magnitude and sign of the correlations with each of the SST-based ENSO indices (detrended 6-month average anomalies) is not surprising, given their mutually high correlations (ranging between 0.6 and 0.95, not shown). The PDO, which also correlates moderately with the ENSO indices (r ~ 0.4–0.5, not shown) shows relatively marginal (positive) correlations with SEAUS temperatures with values ranging between r = 0.25 and 0.3 (Fig. 5h). The climate driver exhibiting by far the most robust relationship with all of the temperature records is the BI (at 160°E) with negative correlations ranging between r = −0.3 and −0.55 (Fig. 5i). Although not documented before for those stations, this result is by itself not surprising to the authors given that (i) BI was shown to be the primary player in modulating SEAUS climate (e.g., Risbey et al. 2009) and (ii) the geographical location of this particular blocking with respect to SEAUS. The definition of blocking originally developed by Wright (1974) and presented in Coughlan (1983) and Pook and Gibson (1999) requires that the high pressure ridge at mean sea level be centered at least 7° latitude south of the mean position of the subtropical ridge for the time of year; In NOVAPR (MAYOCT) this would correspond to latitudes at or higher than ~45°S (~40°S). A blocking high pressure system at this location (40°–45°S, 160°E; i.e., in the Tasman Sea) would produce onshore flow over SEAUS of relatively cold modified maritime air from higher latitudes. This explains why blocking [indicated by a positive BI, as defined by Pook and Gibson (1999)] is, in this case, correlated with negative trends in mean maximum temperature anomalies.
During MAYOCT, SAM maintains a robust inverse relationship with most stations (Fig. 6a). In contrast to NOVAPR, some stations do show a statistically significant relationship with DMI (Fig. 6b). However, the significant correlations with ENSO-based indices obtained earlier in NOVAPR are not present in MAYOCT (cf. Figs. 5c–g with Figs. 6c–g). The correlations with the PDO for MAYOCT are of similar magnitude to those for NOVAPR but of opposite sign (cf. Figs. 5h and 6h). Similar to NOVAPR and perhaps due to a similar physical mechanism than proposed earlier, BI is the index most highly correlated with each of the temperature records with values ranging between r = −0.35 and −0.65 (Fig. 6i).
Because of the existence of a nonlinear interplay between ENSO and DMI (Power et al. 1999; Meyers et al. 2007; Risbey et al. 2009), partial correlations between ENSO-related indices and DMI were computed for both seasons to confirm or refute the earlier relationships drawn from Figs. 5 and 6. Because the PDO also is moderately related to ENSO, its effect on DMI also was considered and briefly analyzed for completeness. It is relevant to stress, however, that owing to some of the fundamental assumptions proper to the partial correlation method (e.g., the effects of the independent variables must be additively and not jointly related) the latter suffers from several limitations. Moreover, the interpretation of the results presumes that the independent variables have causal effects on the dependent variable, which often is not the case.
For MAYOCT, the earlier DMI and ENSO correlations (Fig. 6) are preserved. The relationship with DMI remains when the effect of ENSO is excluded (cf. Fig. 6b with Figs. 7a,c,e), while the lack of a significant relationship with ENSO was not improved when the effect of DMI was removed (cf. Figs. 6c–g with Figs. 7b,d,f). Now, focusing on the partial correlations with ENSO and PDO, the relationship seen earlier in NOVAPR between G1 and ENSO (Figs. 5c–g) was generally degraded (while still remaining statistically significant) when the effect of PDO is excluded (not shown). During MAYOCT, however, the negative correlations obtained earlier with the SOI are larger (more negative by ~0.1) when excluding the influence of the PDO (cf. Figs. 6c and 8b). Similarly, the statistically insignificant positive correlation seen with Niño-4 is increased to marginally significant, ranging between r = 0.25 and 0.35 for all stations except ME and Sydney (SY) (cf. Figs. 6f and 8d). The same behavior is seen with the negative correlation obtained earlier with the PDO, which becomes more negative when ENSO is excluded, particularly when the effect of Niño-4 (cf. Figs. 6h and 8c) and the SOI (cf. Figs. 6h and 8a) is removed, respectively. In summary, the DMI and ENSO correlations are preserved when partially correlated. This, however, is shown not to be the case for the PDO and ENSO.
b. Point-by-point correlations with global SST fields
Figures 5 and 6 reveal that the correlations exhibit a relatively high degree of variability among the 10 stations. While some stations have significant correlations, others remain insignificant (e.g., Fig. 5d). This suggests that the temperature variability for some group(s) of stations could be attributed to other factors not directly related to the large-scale drivers selected here: for instance, local SST effects. To identify some of these potential factors (i.e., not directly related to any of the climate drivers selected here) and, to a certain degree, further validate some of the correlations obtained earlier, particularly with ENSO and DMI, a point-by-point correlation with the ERSST V3b SST dataset was performed for each station (not shown), which then were combined into two distinct groups of stations, for brevity. The first group of stations is G1 and, as discussed earlier, was selected owing to its robust relationship with SST-based ENSO indices in NOVAPR. The second group, G2, comprises the four stations ME, WO, Yongala (YO), and Mildura (MI) (Fig. 2). The choice for G2 was motivated by preliminary SST correlation analysis showing a distinctive SST correlation pattern for these four stations (in NOVAPR) that was not associated with any known SST signatures associated with the SST-based climate drivers in Fig. 5 or Fig. 6. Moreover, it can be seen in Fig. 5 that, in contrast to G1, G2 exhibits overall statistically insignificant correlations with all the SST-based climate drivers (Figs. 5 and 6), further suggesting the existence of potential additional driver(s) for the observed positive trend.
Consistent with G1 exhibiting the most robust relationships with SST-based ENSO indices during NOVAPR (Fig. 5), the point-by-point correlation with the SST exhibits a well-defined horseshoe pattern in the equatorial Pacific (Fig. 9a). The linkage between this particular SST correlation signature and ENSO is well known and has been reported by previous investigators (e.g., Barlow et al. 2001; Ummenhofer at al. 2008, 2009; Krishnamurthy and Kirtman 2009). Figure 9a also confirms the lack of statistically significant relationship of those four stations with DMI during NOVAPR obtained earlier (Fig. 5b). As stated in the foregoing and, in contrast to G1, G2 lacked significant relationships with ENSO and DMI—a result also seen in Fig. 9b. In contrast, moderate correlations are seen over the equatorial Indian Ocean near the Arabian Sea, a pattern that does not correspond to that expected for DMI (Saji et al. 1999). Perhaps the most salient feature of Fig. 9b is a localized area of very high positive correlations (r > 0.6) directly adjacent to the south/southeastern coast of Australia, which strongly suggests that the temperature variability for G2, in contrast to G1, is associated primarily with local SST patterns. This finding is consistent with a recent study by Sadler et al. (2012), who discussed the link between anomalously high positive SST anomalies adjacent to the southern coast of Australia and the occurrence of hot spells in Melbourne (which is included in G2). Figure 9b further suggests that the correlation between high maximum temperature anomalies and high SST anomalies adjacent to the southern coast of Australia extends to stations located in western Victoria and near the coast of South Australia (Fig. 2). This result also highlights the high degree of variability in external factors controlling temperature (and rainfall) over southeast Australia [see, e.g., Fig. 15 in Risbey et al. (2009)].
In agreement with Fig. 6, no significant relationships are found in the Pacific (i.e., ENSO and PDO) for G1 or G2 during MAYOCT (Figs. 9c,d). In contrast, however, and also in concordance with the results presented earlier in Fig. 6b, the largest statistically significant rela-tionships are seen for G2 in the Indian Ocean closer to the Horn of Africa, coupled with insignificant (negative) correlations near Java, a pattern more consistent with a positive phase of the DMI (Fig. 9d). Note that for MAYOCT, when the four stations correlating best with DMI [viz., Wagga Wagga (WW), ME, WA, and YO] are selected instead of G2, the SST correlation pattern in the Indian Ocean becomes even more consistent with a positive DMI phase (not shown).
To complement this analysis, 1-, 2-, and 3-month lagged correlations using the SST field as the predictor for mean maximum temperature were carried out (not shown). The results revealed that, during NOVAPR, the horseshoe pattern in the equatorial Pacific (indicative of ENSO) persisted, with very similar correlation magnitudes (within 5%). During MAYOCT, the localized area of high positive correlation (r > 0.6) adjacent to the southern coast of Australia remained at 1-month lag but progressively decreased as the lag increased with values between r = 0.5 and 0.6 at 2-month lag and r ≈ 0.5 at 3-month lag. This analysis highlights the shorter time scale nature of the relationship between the SST fields and the mean maximum temperatures of G2 compared to G1.
c. Principal component analysis of global SST fields
To complement the above analysis and highlight the primary climate driver modes in the Indian and Pacific Ocean during the selected semiannual seasons, Fierro and Leslie (2013) conducted a principal component analysis (PCA) on the covariance matrix of the same semiannual SST datasets, which was extended herein for the period 1958–2012 (Fig. 10). For brevity, only the leading PC results are shown. In NOVAPR, the Pacific Ocean exhibited a loading pattern for the two leading principal components (PCs) consistent with an ENSO signature (Fig. 10a), a result in line with Barlow et al. (2001). This signature also was present in MAYOCT, although with lower (by 5.1%) variance explained for the leading PC (Fig. 10b). In contrast with NOVAPR, the Indian Ocean displayed a leading pattern that is more geographically consistent with the DMI signature (Saji et al. 1999) (cf. Figs. 10c and 10d). Consistent with the earlier correlation results, the PCA analysis further suggests that, among the SST-based indices, DMI is expected to exert a more influential role during MAYOCT, while in NOVAPR ENSO is expected to play the more prominent role on the Australian climate.
d. Temperature wavelet analysis
In this section, wavelet power spectra and global spectra for both seasons are computed to discern the presence of significant signal(s) that might be embedded in the time series. In particular, it is important to establish whether any given periodic high powers are consistent with years coincident with ENSO, DMI, and SAM years. Note that such analysis cannot be conducted for the PDO (Fig. 6h) because the 54-yr length of the time period considered in this study (1958–2012) is insufficient to cover several 16–30-yr periods/signals. For this reason, most of the periods exceeding 10 yr are statistically insignificant at the 95% confidence interval (equivalent to the 5% significance level in variances depicted by the cone of influence in Fig. 11). For this analysis, two representative stations (WW and WA) exhibiting statistically significant positive relationships with SST-based ENSO indices in both seasons (Figs. 5 and 6) are shown first. WW and WA also are selected because they correlate well with DMI in MAYOCT (Fig. 6b). For similar reasons, G1 was included owing to its significant relationship with ENSO during NOVAPR. The wavelet analysis for G2, however, is omitted due to its lack of significant relationships with any of the climate drivers of Figs. 5 and 6. Two additional stations (ME and SY) were included for comparison and completeness.
Consistent with their overall relatively significant (positive) ties with ENSO found earlier for NOVAPR (i.e., Figs. 5c–g), the WA and WW respective power spectra are consistently characterized by high power levels in most El Niño years (1969/70, 1972/73, double peak in 1983, and 1997/98), and with a global power spectrum showing a peak at 4 and 3 yr, respectively (Figs. 11a,b). The correlation magnitude and sign of WW and SY with ENSO-related indices is similar (i.e., El Niño), although SY consistently exhibits a less significant relationship than WW with ENSO. This is consistent with their power spectra exhibiting an overall similar profile at a comparable peak period of 4 yr, but remains inconsistent with the overall higher power seen for SY (cf. Figs. 11b and 11c). With the exception of a marginal positive correlation with DMI in NOVAPR, ME is the only station not correlating significantly with all other drivers analyzed here (Fig. 5) but correlates very highly (r > 0.6) with local SST fields (Fig. 9b). The results appear to be consistent with comparatively lower powers seen for ME relative to the other stations (cf. Figs. 11d and 11a–c). The power spectrum for G1 shows a well-defined dual peak at 2 and 4 yr during most years (see above) potentially associated with strong La Niña/El Niño events (Fig. 11e). Supporting this, 1982/83 is characterized by a local maximum in the power spectrum. Despite most temperature stations correlating negatively with SAM in NOVAPR (or positive relationship with negative SAM), none of the power spectra in Figs. 9a–e show a peak of 2 yr in the global power spectrum, which would be consistent with the mean periodicity of SAM (Nan and Li 2003).
The above result for SAM, however, does not hold during MAYOCT with most stations, hinting at a statistically significant contribution of a 2-yr period signal embedded within the data (Fig. 12). SY, WA, and G1, in particular, produced robust (inverse) relationships with SAM, r ≈ 0.35–0.45 (Fig. 6a); consistent with this, their respective power spectra indicate high power values during negative SAM years, such as 1980/81, 1991–93, 2001, and 2002 (Figs. 12a,c,e). The year 2008, however, also was characterized by a strong negative SAM event but is not characterized by high power in any of the above three spectra. Despite an overall lack of significant relationships between temperature and ENSO during MAYOCT, the power spectra in Fig. 12 indicate some significant contribution of the 3–4-yr signal in the data. Although irregular in nature (Saji et al. 1999, their Fig. 1), it is possible that part of this 3–4-yr signal could be associated with DMI. The positive correlations with DMI in MAYOCT for most stations should yield high powers during positive DMI years (e.g., 1961–63, 1972, 1994, 1997, and 2007). All stations do, indeed, exhibit high powers during most of those years (within the 95% confidence cone; Fig. 12).
4. Discussion and conclusions
Correlations were calculated between season-averaged maximum temperature anomalies over the southeast Australian (SEAUS) region and a range of possible climate driver indices. Correlations with global SST fields, principal component analysis, and wavelet analysis of temperature records suggested possible links between well-known climate drivers and the observed increase in mean maximum temperature anomalies, present in SEAUS observations over the last half century.
The study revealed that during the SEAUS warm season [November–April (NOVAPR)] the temperature variability for most of the stations showed significant linkage with the negative phase of the southern annular mode and the positive phase of ENSO. This strong linkage with SAM held during the cool season [May–October (MAYOCT)] in contrast to ENSO-related indices. MAYOCT mean maximum temperature anomaly variability also showed a robust relationship with the dipole mode index (DMI). Most importantly, during both semiannual seasonal periods, the blocking index (BI) at 160°E was found to exert a dominant influence on the interannual variability of mean maximum temperature anomalies.
The above results with SST-based indices were confirmed further via a point-by-point correlation analysis with global SST fields. The dominant role of the DMI in MAYOCT, and of Modoki/Niño during NOVAPR, was accurately reproduced by identifying typical SST correlation patterns associated with those respective indices. These results remained consistent with a PCA analysis conducted on the same SST datasets. Additionally, this analysis may have elucidated one of the reasons why the correlations at some stations consistently differed with others. It was found that a distinct group (G2) of four stations, which otherwise were largely characterized by insignificant correlations with DMI and ENSO in NOVAPR, exhibited a very robust relationship (r > 0.6) with the SST field directly adjacent to the southern and southeastern coasts of Australia. This result highlights the nonuniform and complex nature of the potential factors that influence temperature variability for nearby stations over one particular region of Australia. While remote large-scale drivers may heavily impact the temperature variability at some stations, local effects primarily control the temperature variability at others.
Additional work is warranted to explain the observed increasing temperature trend in SEAUS and over other regions of the Australian continent. In particular, it would be worthwhile exploring the extent to which temperature variability is influenced by the phase symmetries of some drivers [e.g., DMI and ENSO (Power et al. 1999) or PDO and ENSO (Verdon and Franks 2006)], particularly over southeast and southwest Australia, where the most dramatic changes (increases) in mean maximum temperatures have been observed in the recent decades.
Funding was provided by NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. This work is supported by the NESDIS program, which is under the auspices of the National Oceanic and Atmospheric Administration of the U.S. Department of Commerce under the Grant NOAA-NESDIS-OAR-NA08OAR4320904. Computer resources were provided both by the Oklahoma Supercomputing Center for Education and Research (OSCER) hosted at the University of Oklahoma. The authors also wish to thank Dr. W. Sasaki for providing the updated dataset for the Modoki index, Drs. L. Qi and M. Pook for providing the blocking index data, and Mr. T. Skinner for useful discussions that helped improve the quality of the manuscript.