Tropical cyclone trends in the Australian region

Authors


Abstract

[1] Tropical cyclone trends in the Australian region are examined using the Bureau of Meteorology best track data. Here the focus is on analyzing differences in trends between the eastern and western subregions of the Australian formation region, under the assumption that any spurious trends in the best track data due to changes in observational practices would be less noticeable in differences between two adjacent portions. Substantial differences in trends are found between the two subregions, with the number, average maximum intensity, and duration at the severe category intensities of tropical cyclones increasing since 1980 in the west but decreasing (in number) or exhibiting no trend (in intensity, severe category duration) in the east. Analyses of trends in atmospheric variables known to be related to tropical cyclone characteristics also indicate substantial differences between the two subregions.

1. Introduction

[2] The study of tropical cyclones (TCs) and their behavioral response to a warming climate are current areas of active research. The global occurrence of several high-impact events during the very active 2004 and 2005 seasons have made understanding and predicting TC behavior due to induced climate change an issue of scientific and social concern [McBride et al., 2006]. Recently, significant warming trends largely attributed to 20th century anthropogenic forcing were detected in tropical ocean basins where TCs are observed to form [Barnett et al., 2005; Santer et al., 2006; Mann and Emanuel, 2006; Trenberth and Shea, 2006; Knutson et al., 2006]. Questions have since been invariably raised as to whether TC characteristics such as frequency, intensity and storm duration are echoing this signal or mainly reflecting known modes of natural variability such as the Atlantic Multidecadal Oscillation [Goldenberg et al., 2001].

[3] Theoretical considerations [Emanuel, 1987; Holland, 1997] and modeling under enhanced greenhouse conditions [Knutson and Tuleya, 1999; Walsh and Ryan, 2000; Knutson and Tuleya, 2004; Walsh et al., 2004; Oouchi et al., 2006; Stowasser et al., 2007] all suggest regional tendencies toward greater TC intensities and the occurrence of more intense TCs that are linked to rising tropical sea-surface temperatures (SST) in a warmer world. Nonetheless, some simulations under such scenarios also indicate that tropospheric stabilization [Shen et al., 2000; Knutson and Tuleya, 2004], due to predicted upper tropospheric warming [Meehl et al., 2007], along with regional projections of vertical wind shear increase (e.g., in the North Atlantic and in the eastern Pacific [Vecchi and Soden, 2007]), may act to negate any possible future rises in overall TC frequency or intensity that might be related to SST changes. While Vecchi and Soden [2007] report increases in the range of −2% to 30% of the mean shear, the overall simulated change obtained for the North Atlantic “shear enhancement region” in the Caribbean was ∼+1.5 m/s [Vecchi and Soden, 2007, Figure 3]. However, the magnitude of this change is still considered to be within the noise of present interannual variability and is small in terms of its potential to physically affect overall TC development and intensification [e.g., Shapiro, 1987].

[4] Recently, there have been several empirical studies examining archived TC best track data for the presence of frequency and intensity trends [Chan and Liu, 2004; Emanuel, 2005; Webster et al., 2005; Klotzbach, 2006; Wu et al., 2006; Holland and Webster, 2008]. While upward trends have been reported in total TC power dissipation [Emanuel, 2005; Sriver and Huber, 2006], and in the global occurrence of Saffir-Simpson category 4–5 (SS45) TCs [Webster et al., 2005, Table 1], there has been considerable skepticism regarding such findings [Landsea, 2005; Chan, 2006; Klotzbach, 2006; Wu et al., 2006; Kossin et al., 2007]. Emanuel [2005] noted increasing trends in potential TC destructiveness, as approximated by a power dissipation index (PDI), for both the North Atlantic (NAT) and western North Pacific (WNP) basins in the latter part of the 20th century. Defining the PDI as the cube of six-hourly maximum winds integrated over storm lifetime and over all TCs, Emanuel [2005] showed strong correlations between the indices and similarly time-filtered basin-wide SST changes, with trends reportedly reflecting increases in storm intensity and longevity. These low-frequency trends have been supported by Sriver and Huber [2006], who used ERA-40 winds to explicitly calculate global TC power dissipation, with results displaying good agreement with the indicial approximations of Emanuel [2005] post-1978. They also showed that the ERA-derived PDI correlated well with annual mean tropical temperatures, but were even better with SST. Using a simpler analysis, Holland and Webster [2008] concluded that there is a strong 100-year trend in hurricane frequencies for the NAT that corresponds well to distinct eastern Atlantic SST increases; however, no observed trend in the proportion of major hurricanes relative to all TCs was found despite an emerging trend in the number of major hurricanes that is linked to the overall rise in total TC numbers.

[5] Likewise, while no discernible regional trends in storm frequency and duration were observed for the period 1970–2004 with the exception of the NAT, Webster et al. [2005] highlighted a near-doubling of the global number and proportion of the more intense SS45 TCs for the period 1990–2004 compared to 1975–1989, with changes occurring in all basins. Using information theory, Hoyos et al. [2006] showed that the tendency toward more SS45 storms globally during the 1970–2004 period, as analyzed by Webster et al. [2005], is directly related to positive long-term tropical SST changes when the trend was isolated from shorter modes of variability in factors known to promote intensification. Interestingly, Webster et al. [2005] found no increases in the observed maximum intensities of the most severe storms globally. Kepert [2006] subsequently noted that this implied a rightward skewing of the frequency distribution of observed intensity, rather than a shift in the global TC intensity distribution itself.

[6] The detection of such intensity trends has also been questioned on the grounds of data quality, consistency and deficiencies within the historical data sets used [Landsea et al., 2006]. Moreover, Klotzbach [2006] showed no trend in net global TC activity and only a small upward trend in SS45 storms when data similar to that used by Webster et al. [2005], but covering only the period 1986–2005, were analyzed. This is despite a warming of about 0.2–0.4°C in tropical SST over that period. Similarly, Wu et al. [2006] revealed downward trends in overall TC destructiveness and in the number of SS45 TCs when two other separate regional best track data sets were examined for the WNP, further disputing the findings of Emanuel [2005] and Webster et al. [2005]. Likewise, Kossin et al. [2007], using degraded-resolution satellite imagery, found no trend in estimates of TC intensity globally since the early 1980s, directly contrary to the conclusions of Webster et al. [2005]. Kossin et al. [2007] showed that while good agreement is seen for intensity trends in the NAT (upward) and the eastern North Pacific (downward), all other basins displayed trend discrepancies between their reanalyzed data and the official best track compilations.

[7] Regardless, these studies have either been largely confined to the Pacific and NAT regions or have investigated global intensity trends overall. In particular, Webster et al. [2005] largely omitted Australian TCs within 115° to 155°E when they examined the south Indian Ocean (SIO) and southwestern Pacific (SWP) basins. While there has been considerable debate regarding the current detectability of intensity trends in the NAT and WNP, relatively little work has been done for the Australian region in terms of investigating and documenting possible trends using archived observational data. Previous work has focused almost exclusively on the relationship between the El Niño–Southern Oscillation phenomenon and seasonal TC frequency and prediction, either basin-wide [Nicholls, 1984, 1992; Evans and Allen, 1992; Nicholls et al., 1998; McDonnell and Holbrook, 2004], or for the Coral Sea/southwest Pacific region [Basher and Zheng, 1995; Grant and Walsh, 2001; Flay and Nott, 2007]. Nicholls et al. [1998] showed a strong downward trend in overall Australian TC numbers for the period 1969–1996, mirroring the Southern Oscillation Index (SOI) trend seen over the same period. However, an abrupt shift to lower frequencies was seen post-1985, suggesting a possible observational bias, which they attributed to an improved ability to properly distinguish other weaker low-pressure systems from TCs. Interestingly, they also found that the severe TCs (<971 hPa, Australian cyclone severity scale; see Table 1) displayed a weak upward trend, contrary to that expected from the downward SOI forcing.

Table 1. Australian Cyclone Severity Scale and Saffir-Simpson Hurricane Scalea
CategoryMin. Central Pressure (hPa)10-Min Avg. MSW (m/s)Peak Gusts (m/s)Equivalent Saffir-Simpson/Australian
  • a

    To aid comparison with the Australian scale (A, top), the Saffir-Simpson scale (SS, bottom) has been expressed in terms of the parameters of the Australian scale (10-min average maximum sustained winds and peak gusts in m/s). Adapted from http://www.australiansevereweather.com.

A1986–99517.5–24.025.0–34.5none
A2971–98524.5–32.535.0–47.0none to Mid-SS1
A3956–97033.0–43.547.0–62.0Mid-SS1 to High-SS2
A4930–95544.0–54.062.5–77.5Low-SS3 to Low-SS4
A5≤929≥55.0≥78.0Mid-SS4 and greater
SS1≥98028.5–37.040.5–53.0Mid-A2 to Mid-A3
SS2965–97937.0–43.053.0–61.0Mid-A3 to High-A3
SS3945–96443.5–50.561.5–72.0High-A3 to Mid-A4
SS4920–94451.0–60.072.5–86.0Mid-A4 to Mid-A5
SS5≤919≥60.5≥86.5Mid-A5 to High-A5

[8] The current paper therefore seeks to compliment the above body of knowledge by analyzing TC best track data in the Australian region for the presence of systematic intensity and duration trends. We ask the following questions: (1) Are TCs in the region, on average, becoming more intense and longer lived? (2) Are the severe TCs, on average, becoming more frequent and spending more time at category 3 intensities or higher? A key feature of the study is that we examine and compare the trends found in both the SIO and SWP portions that constitute the official Australian TC region (90° to 160°E). In this way, differences in trends between the two subregions are less likely to be affected by time-varying data biases than trends in each subregion taken individually. Particular emphasis has also been placed on the severe TCs, systems which reach at least category 3 intensity on the Australian cyclone severity classification scale. This is because seasonal severe TC numbers throughout the analysis period are likely to be more robust, since the threshold of 970 hPa used for severe cyclones equates, in the Australian basin, to TCs with an identifiable eye from satellite imagery [Nicholls et al., 1998].

[9] Despite concerns regarding the homogeneity of these data, substantially different trends are found in severe cyclone statistics for those forming off Western Australia and Northern Territory (collectively referred to as the western sector; see section 2.1) compared with those forming off Queensland and the Gulf of Carpentaria (eastern sector). It is hypothesized that changes in observing practices would be more likely to cause systematic trends across the entire Australian region, rather than observed differences between the western and eastern sectors. Reasons for these disparities are examined, including the substantially different trends in SST in the two regions.

2. Data and Methods

2.1. Study Domain and Best Track Data

[10] The domain used in the study covers the official TC basin from latitudes 5° to 30°S, with 30°S chosen as the southern limit to remove any extratropical influences. Since the Australian region is geographically bounded by two different oceans, we partitioned the basin into 2 portions along the 135°E meridian: a western sector (90° to 135°E) and an eastern sector (135° to 160°E). Given that TC longitude crossings were found to be at a minimum along this meridian [Kuleshov, 2003], it now represents the widely accepted boundary between the SIO and the SWP (Y. Kuleshov, personal communication, 2006).

[11] In examining the SIO and SWP regions, both Webster et al. [2005] and Klotzbach [2006] used the Neumann data set [Neumann, 1999], with post-2002 data taken from the Joint Typhoon Warning Center (JTWC) best track archive. Here, we analyzed the TC best track data maintained by the National Climate Centre of the Australian Bureau of Meteorology (BoM; http://www.bom.gov.au/climate/how). This data set contains TC records dating from 1906 to season 2004/05 (and updates). The archive consists of systems that either formed or moved into the official Australian region after having achieved a tropical low classification [Hall et al., 2001]. Among other information, the data set includes satellite-derived position fixes (in latitude/longitude coordinates) and corresponding intensity estimates of central pressures and wind speed, although the latter only became widely reported across all the three Tropical Cyclone Warning Centres (TCWC) of Perth, Darwin and Brisbane after 1985.

[12] Because of data deficiencies prior to the 1960s [Holland, 1981], we restricted our analysis to the satellite-covered era spanning cyclone seasons 1969/1970–2004/2005. Specifically, data predating the chosen period contain known errors, omissions and duplications that are currently being addressed as part of an ongoing process of quality checking, with updates included as and when they became available (Y. Kuleshov, personal communication, 2006). Our data was extracted after the June 2006 update.

[13] There are, however, some reservations raised regarding the homogeneity of the data chosen for this study. These relate to the improvements in satellite technology along with the application (and limitations) of the Dvorak analysis techniques [Dvorak, 1975, 1984]. For the selected period, there has been a steady increase in satellite spatial, spectral and temporal resolution over the region and probably methodological changes in the forecaster application of these techniques (J. Kepert, personal communication, 2006), meaning that modern storms were more frequently and better sampled. Our knowledge and skill in detection and analysis have also subsequently improved [Harper and Callaghan, 2006; Harper et al., 2008]. It has therefore been suggested that for intensity estimates, data is considered most reliable from 1980 onward, while a realistic extension to 1970 is possible for TC numbers and positions [Trewin, 2006]. Accordingly, with the possibility that artificial trends in TC behavior may be created by induced sampling biases, observed temporal intensity trends over the whole 1969–2004 period were treated with caution; a trend is considered robust only if it persisted over the last 25 years (i.e., from 1980 to 2004).

2.2. Methodology

[14] TC activity in the two sectors was examined for the following measures of behavior: (1) overall and severe TC frequencies (with the latter type stratified into category 3 (A3) and category 4–5 (A45) cyclones according to the Australian cyclone severity scale); (2) average maximum intensity reached as measured by minimum central pressure (MCP); (3) mean TC lifetime and (4) mean duration at severe category intensities (i.e., the average number of days per severe storm, per year affected by category 3 intensities and above, hereafter designated A3+ and A4+ days). Because dissimilar environmental flow meant the application of different wind-pressure relationships at the various TCWCs (J. Kepert, personal communication, 2006), we chose to analyze intensity by MCP instead to avoid induced biases. We assumed that any changes in forecaster procedures for estimating intensities from satellite imagery were standardized and concurrently implemented across all three TCWCs over the analysis period.

[15] To further maintain consistency, the following criteria were applied: (1) TCs were only included in the overall count if some part of their recorded tracks were within the defined boundaries; (2) TCs which formed west (east) of 135°E were classified as western (eastern) portion systems, with TCs that crossed the defined partition being classified into the portion in which it accrued more storm days; (3) systems were only counted as severe if 970 hPa was attained within the boundaries; this meant the exclusion of some listed systems that were classified as severe in the database but were found to actually reach maximum intensity outside the defined boundaries; and (4) track parts that ventured outside the official region were excluded from the lifetime analysis, thus only the number of days that a system was actually within the domain was aggregated to represent the cumulative TC lifetime within the basin. Any intensification and/or decay would therefore purely reflect the environmental conditions within the region.

[16] Overall, a total of 426 TCs were analyzed for the study, with 186 of them attaining severe TC status. The following results are shown with landfall effects filtered: that is, we only considered TC data for when system center fixes were over water. Five-year running averages were used throughout to smooth out interannual variability and trends were analyzed by fitting linear regressions to the running means. The slopes of these trends are then tested for statistical significance at the 95% confidence level.

3. Results

3.1. Occurrences

[17] Statistics for all and severe TC incidences within the two sectors over the 36-year period are summarized in Table 2 with data binned into 12-year totals to highlight interdecadal differences. Aggregates for each severe intensity category (A3, A45 and SS45) are also shown. From the table, a sustained decline in total eastern TC numbers is seen across the entire 36-year period. Conversely, overall western TC numbers remained relatively constant over the last 24 years. In fact, time series plots of seasonal TC counts smoothed by 5-year running averages (not shown) revealed diverging trends in overall incidences for both subregions after 1985. While the eastern sector displayed a sustained long-term downward trend that had persisted since the 1970s, the western sector showed a statistically significant upswing in average TC frequency for the period covering 1986–1999 (5-year slope: R2 = 0.56, p = 0.012) before another slight downturn is seen at the turn of 21st century. This behavior, notwithstanding the likelihood of misclassification put forward by Nicholls et al. [1998], suggests the presence of a decadal-scale oscillation for total western TC occurrences that explained the more neutral, basin-wide trend seen for the last 20 years (not shown).

Table 2. Summary of the Aggregate Occurrences for All and Severe TCs for the Western and Eastern Portions of the Australian Basina
Type of OccurrenceWesternEastern
1969–19801981–19921993–20041969–19801981–19921993–2004
  • a

    Totals for each severe intensity category (A3, A45, and Saffir-Simpson (SS45)) are also shown. The data have been binned into 12-year periods to highlight long-term trends.

All988385665539
Severe (A3 + A45)413947182417
A3 (minor)16121110108
A45 (major)2527368149
Number of SS45 status161225175
% severe → SS4539%31%53%6%29%29%

[18] The statistics also indicate that the western sector was more conducive to greater and more intense severe TC activity throughout the 36-year period. Particularly, the rise in total severe western TC numbers since 1993 was largely driven by an increase in category A45 TCs while the number and percentage, with respect to all TCs, of category A3 cyclones stayed relatively similar. Remarkably, the overall numerical decline in eastern severe TCs for the same period came directly from a reduction in these category A45 storms. Nonetheless, the sector's percentage of such major severe cyclones was comparable to that of the previous period (∼23% to ∼25%), with those reaching SS45 status also remaining proportionately equivalent (∼13%). No change for the eastern half was observed if the proportion of SS45 storms was taken relative to severe TCs only (∼29% in both periods). In stark contrast, more western TCs reached SS45 status for period 1993–2004 (a 208% increase from the previous period, from 12 to 25); this is not seen for the eastern portion in a time when intensity estimates are considered most reliable. Interestingly, sixteen SS45 storms were detected for the western sector in the 1969–1980 period while only one was observed in the eastern half. That said, the possibility of an induced classification bias across both portions due to increased sampling in the 1990s cannot be fully discounted despite the observed differences in occurrence types over the entire 36-year period.

[19] Temporal trends in severe TCs by classification are further highlighted by Figure 1. Although total seasonal severe counts (stacked columns, A3 + A45) in both sectors display interannual variability, distinct upward (A45) and downward (A3) trends are observed for the western half when the number of severe TCs is stratified by intensity category. This behavior is reflective of overall basin-wide trends for the same categories (not shown). Regression analyses of the 5-year running averages revealed highly significant nonzero slopes (p = 0.00022 for A45; p = 0.00016 for A3). However, if a cautionary view of the data is taken and only intensity data from 1980 onward is considered, only the downward regression slope for category A3 cyclones remained statistically significant (p = 0.0115, not shown), while a more neutral trend in category A45 TCs (both in number and proportion) emerged.

Figure 1.

Seasonal severe TC occurrences for each sector when classified by the two severity categories as stacked columns (minor, A3; major, A45). Five-year running averages are also shown (plotted against top x axis) to smooth out interannual variability. Regression slopes (dotted lines) for the running means in the western portion are very significantly nonzero (p = 0.00022 for A45, p = 0.00016 for A3). The abscissa gives the year in which the cyclone season begins (e.g., 1969/1970 is plotted against 1969).

[20] Distinctively, the eastern sector displayed an oscillatory and out-of-phase relationship between the running average occurrences of minor and major severe TCs throughout the entire 36-year period. This relationship is also maintained when corresponding proportions of A3/A45 cyclones relative to all TCs were computed. Correlations between the 5-year running average number and percentage of these minor/major severe TCs are −0.13 and −0.27, respectively, but these are not statistically significant at the 95% confidence level. Interestingly, this result is akin to that reported by Holland and Webster [2008], who noted a similar cyclic behavior in the number and proportions of major and minor hurricanes (Saffir-Simpson scale) relative to all named TCs for the NAT, particularly after 1945. While Holland and Webster [2008] have speculated that such an oscillation between major and minor hurricanes may be linked to the proportion that develops either equatorward or poleward of 25°N, the reasons for our finding for the eastern sector remain questions to be explored in future studies.

3.2. Average Maximum Intensities, Cumulative Lifetimes, and Severe Category Intensity Durations

[21] Figure 2 presents the time series of seasonal average maximum intensity reached (as measured by MCP) for when all and severe TCs are examined in both portions of the basin. Five-year running means are again plotted to smooth out interannual variability. When all TCs are considered over the 36-year period, both sectors show statistically significant, long-term downward trends in average MCP, although it is worth noting that the goodness of fit for the eastern portion is considerably much lower (Western − R2 = 0.64, p < 0.0001; Eastern − R2 = 0.13, p = 0.047). While the western temporal trend was more gradual, the downward eastern trend was modified by sharp increases in seasonal average maximum intensity just prior to 1994. The years 1991–1993 were actually characterized by a short period of heightened intense TC activity in the eastern sector, with seven A45 cyclones spawning (Figure 1) and three of those reaching SS45 status. However, when later, more reliable data from 1980 onward is examined, only the western sector trend remained statistically significant (R2 = 0.33, p = 0.007, not shown). Not surprisingly, the eastern downward trend actually reverses into a slight, though insignificant, upward 25-year trend of decreasing intensity (R2 = 0.09, p = 0.20, not shown), driven by the rising average MCP observed since 1993.

Figure 2.

Temporal maximum intensity profiles in both the western and eastern sectors for all and severe TC occurrences. Linear trends have been fitted to the 5-year running averages (dotted lines). Both downward regression slopes for the western half are very significantly nonzero (p < 0.0001, respectively). The intensity regression slope for all eastern TCs over the 36-year period is also statistically significant (p = 0.047). Note that for severe TCs, the running average was adjusted to include only those seasonal values that were present (i.e., in a 5-year contiguous period, if only 4 values are present, then the running average would be the mean of those 4 years).

[22] Differences between the two halves are further accentuated when the average maximum intensities of severe TCs were considered separately. While the eastern sector was observed to exhibit variability of severe intensities on nominal interdecadal-like timescales, its western counterpart showed a pronounced, downward long-term trend in average MCP reached by severe TCs (R2 = 0.71, p < 0.0001). This trend also persisted even for the more reliable data post-1980, with the regression slope remaining statistically different from zero beyond the 99.9% confidence level. Conversely, a statistically significant upward trend in running average MCP was actually observed for the eastern half (R2 = 0.26, p = 0.018, not shown) over the last 25 years, one that is primarily driven by the decrease in average maximum intensity attained for the period 1993–2000. Interestingly, this rise is seen at a time when increasingly positive SST anomalies are observed off the Queensland coast in the band of main severe TC activity; see section 3.3 and Figure 5.

[23] Longer-lived TCs, in particular those that survive longer at the severe category intensities, raise the potential for more storm-related damage through greater wind speeds and increased precipitation. This is especially the case for when storms have near-coastal approaches before making landfall. On average, western sector TCs were found to have longer cumulative TC days (mean: 5.1, std. dev: 1.1) during the chosen analysis period than their eastern counterparts (mean: 4.0, std. dev: 2.4). However, no long-term, 36-year duration trend was seen for the western subregion (Figure 3, top panel), although a slight downward trend is evident from 1990 onward. Despite this, regression analyses of the 5-year running means revealed statistically significant upward trends in both categories of severe intensity days for the western sector. Highly nonzero slopes were not only observed for the entire 36-year period (5yr A3+: R2 = 0.37, p = 0.0002; 5yr A4+: R2 = 0.57, p < 0.0001) but also for the period post-1980 (5yr A3+: R2 = 0.47, p = 0.0006; 5yr A4+: R2 = 0.66, p < 0.0001; not shown), indicating that, on average, intense western TCs have been retaining the more severe intensities for longer in recent times. Like those observed for average maximum intensities, both trends in severe category durations reflected those seen when basin-wide statistics were computed (not shown).

Figure 3.

Temporal profiles of average cumulative (black) and severe category (A3+ and A4+, colored) TC days per season. Thick lines represent the 5-year running means with linear trends fitted for the western sector. The upward regression slopes for both ranges of severe category TC days in the western sector are very significantly nonzero (A3+: p = 0.0002; A4+: p < 0.0001).

[24] In stark contrast, the eastern sector (Figure 3, bottom panel) exhibited temporal variability on interdecadal-like timescales across all duration profiles (i.e., cumulative TC and severe intensity category days). Note that a downward cumulative lifetime trend is noticeable for eastern TCs for the decade covering 1990–2000. This trend reversed the short earlier increase seen up to the early 1980s, driven by the “spike” caused by severe TC Elinor (max. intensity 935 hPa, but one of only 2 eastern storms for the 1982/83 season) which lasted for 21 days owing to an erratic, elliptical track over open water. In any case, the highly significant correlations (beyond the 99% confidence level) found between the temporal profiles of severe intensities and A3+ days (5-year running means: western, r = −0.77; eastern, r = −0.90) do confirm the theoretical relationship that, on average, the more intense storms tend to last longer at their severe intensities from the time the severe category threshold is crossed until maximum storm strength is reached. As such, the disparities in the physical trends could therefore reflect actual differences between specific environmental conditions of the two sectors. Such differences in observed trends are also less likely to be affected by data biases than individual trends in each portion. Data heterogeneities would likely only introduce temporally constant biases that would affect both sectors equally whereas the observed trends are spatially distinct.

3.3. Some Observed Changes in the Large-Scale Tropical Environment

[25] In an effort to understand the disparities seen between the two sectors, we also examined the temporal behavior of three large-scale environmental fields relevant to TC development and intensification over the chosen period. The parameters investigated are SST, mean specific humidity within the 925–500 mb layer (shum925-500) (herein taken to represent the low tropospheric to midtropospheric moisture content), and the environmental vertical wind shear (EVWS) over the Australian TC region. Here, we conventionally defined EVWS to be the magnitude of the vector difference of winds between the Z200 and Z850 pressure levels and given by the following equation:

equation image

[26] To assess the presence of long-term changes, large-scale composite fields for two periods of 18 seasons each (Period 1: 1969/1970–1986/1987; Period 2: 1987/1988–2004/2005) were first constructed using gridded monthly mean (November to April) values from the following sources: (1) NOAA Extended Reconstructed SST version 2 (ERSST.v2) data set of Smith and Reynolds [2004], taken from http://www.cdc.noaa.gov/cdc/data.noaa.ersst.html, and (2) NCEP/NCAR reanalysis data set [Kalnay et al., 1996] with specific humidity (Z925, Z850, Z700, Z600 and Z500) and wind component data (Z200 and Z850) obtained from http://www.cdc.noaa.gov/cdc/data.ncep.reanalysis.derived.pressure.html.

[27] November to April values were chosen since these months reflect the main period of seasonal TC activity for the entire Australian region. The composite fields were then tested for significant differences in the mean at the 95% confidence level using the Student's t test assuming equal variances. Results are shown in Figures 4a4d, in which notable differences between the two sectors can be seen. In the SST field, a much greater warming has occurred for the western sector as compared to the eastern portion across equivalent latitudes. Near-coastal increases were, however, found to be statistically insignificant (Figure 4a). While this is in contrast to the significant SST rise observed near the coast for the eastern sector equatorward of 18°S, the latter can be explained by the sharp increase seen over the last decade of the analysis; in fact, the temporal trend for the western portion has been gradual and sustained (Figure 5).

Figure 4.

Differences in the mean between the two 18-year periods for (a) SST (degrees Celsius), (b) shum925-500 (g/kg), (c) shum925mb level (g/kg), and (d) EVWS (m/s). Statistically significant changes at the 95% confidence level (i.e., computed t value ≥ critical t value of 2.0322 for 34 degrees of freedom) according to the Student's t test, assuming equal variances, are shaded gray.

Figure 5.

Temporal profiles of SST, shum925-500, and EVWS anomalies in the two bands of main severe TC activity (see text for band definitions) for both the western and eastern sectors. Five-year running average anomalies shown with linear regressions fitted. P values for the regressions in each latitude band are as follows: SST, p < 0.0001 (1W, 2W, 1E), p = 0.021 (2E); shum925-500, p < 0.0001 (1W), p = 0.079 (2W), p = 0.002 (1E), p = 0.102 (2E); EVWS, p = 0.911 (1W), p = 0.164 (2W), p = 0.939 (1E), p = 0.234 (2E).

[28] The western sector has also experienced some significant moistening in the lower troposphere to midtroposphere generally east of 110°E, with values >0.3 g/kg and >0.5 g/kg seen for the Northwest Shelf and Timor Sea regions, respectively (Figure 4b). Interestingly, areas poleward of 12°S and west of 110°E have experienced drying, with a localized, statistically significant decrease observed in the region 16° to 30°S and 90° to 105°E. For the eastern sector, only a localized significant increase (0.4 g/kg contour) is seen for the Gulf of Carpentaria while remarkably, much of the Coral Sea region has shown little or no long-term changes in mean shum925-500. Extensive moistening is also evident in the lower tropospheric boundary layer (represented here by the 925 mb level, i.e., ∼800–850 m above sea level), with statistically significant increases in specific humidity observed over most of the western sector (Figure 4c). The greatest rises (in excess of a +0.9 g/kg difference in the 18-year means) are seen centered within 110°E and 120°E of the Northwest Shelf/Indian Ocean region. Significant increases are also seen for the eastern sector although these are comparatively much smaller in magnitude than those in the western sector over the same latitudes. Analyzing the other tropospheric levels separately, smaller increases (decreases) in moisture content were found generally east (west) of 110°E for levels Z850-Z600 (Z850-Z500), thus explaining the observed spatial pattern of changes in Figure 4b. As height increases, areas of statistically significant increases were observed to gradually retreat further east, while small, yet statistically significant decreases remained roughly confined to latitudes 15° to 25°S, and west of 105°E (not shown). Finally, only minimal changes in mean EVWS are seen in the western region of interest for TC development and intensification, i.e., poleward of 10°S, with next to no change for the eastern portion; the magnitudes observed in the western sector are however too small to be of any physical importance (Figure 4d).

[29] Our analysis of TC records revealed that the bulk of severe TC activity in both sectors occurred within latitudes 10° to 25°S. However, maximum intensities, including those of the higher category A45 and SS45 intensities were mainly observed in the 10° to 20°S latitude band. In this narrower band, two “clusters” of these major intensities were seen approximately around 90° to 100°E (SIO portion) and 115° to 120°E (Northwest Shelf region of Western Australia), while such instances in the eastern sector generally covered a broader spatial extent. We therefore narrowed our examination to this 10-degree latitude belt in further assessing the above environmental parameters for temporal trends. To minimize land effects due to Australia's coastline, we divided this principal band of interest into four smaller 5-degree bands (two per sector): Band 1W (90° to 135°E, 10° to 15°S), Band 2W (90° to 125°E, 15° to 20°S), Band 1E (135° to 160°E, 10° to 15°S) and Band 2E (145° to 160°E, 15° to 20°S).

[30] For each variable, 5-year running averages of the mean seasonal anomaly, spatially averaged for each 5-degree band, are shown in Figure 5. The mean seasonal anomaly was constructed using only November to April values once again, with monthly anomalies defined as the departure of a particular monthly mean value from its corresponding climatology (base 1969–2005).

[31] The behaviors of the temporal profiles in Figure 5 further confirm the observations inferred from Figure 4. Although highly significant upward SST trends are observed across all four bands, regression slopes for the eastern bands are evidently dominated by the steep rise in SST since the 1990s. In fact, the anomaly time series indicate that SSTs in the eastern sector exhibited more interdecadal-like variations despite the warming of the 1990s. Remarkably, the marked increase in eastern sector SSTs over the last decade is observed at a time when the average maximum intensity of TCs is decreasing there, as seen in Figure 3 earlier. Worth noting is the close resemblance of our profiles to that obtained by Hoyos et al. [2006, Figure 1] even though they covered rather different spatial extents in the Southern Hemisphere.

[32] With regards to the layer-averaged specific humidity anomaly profiles, positive regression slopes are only seen for the 10° to 15°S bands of both sectors. The absence of a significant trend for the Band 2 latitudes of the western portion clearly reflects the competing effects of drying (moistening) in areas west (east) of 110°E, inferred from Figure 4. Note however, that the shum925-500 profiles for the eastern sector are dominated by the short period of moisture enhancement during the 1990s. While not specifically shown here, it was found that all anomaly profiles for spatially averaged specific humidity at the 925 mb level (shum925) revealed statistically significant upward trends. Detrended correlations between the 5-year running means of SST and shum925 indicated that these low-level moisture increases are indeed due to actual trends in SST, with all coefficients beyond the 95% confidence level.

[33] Reflecting the marginal difference in means observed in Figure 4d, all average EVWS anomaly profiles exhibited somewhat interdecadal-like variability, particularly in the Band 1 latitudes. While a slight upward trend is seen in Band 2 EWVS across both sectors, no physically significant shear changes are observed in the region of interest between the two 18-year periods overall. A point to note however, is that compared to the other parameters, the spread in the relative magnitudes of mean EVWS anomalies seen between the two 5-degree latitude bands in each sector does suggest that dynamical effects tend to be more spatially localized than the thermodynamic effects.

[34] Though simple, the above analyses have clearly indicated that the western sector has experienced greater changes with respect to SST and low-level moisture content than those experienced by eastern portion. When combined, warmer SSTs and greater layer-averaged humidity certainly have the potential to enhance the regional thermodynamic environment, which can make it more conducive to overall TC development and intensification. However, since TC intensification is also influenced by static stability and other factors such as zonal stretching deformation [Hoyos et al., 2006], unraveling the specific dependence of TC intensity on all these relevant factors will be left to future studies on the Australian TC region.

4. Discussion and Concluding Remarks

[35] A comparison with the results of Kossin et al. [2007] is instructive. For the Southern Hemisphere, their analysis is divided geographically into the SIO (west of 142°E) and South Pacific (SPAC) basins (east of 142°E). While these include larger regions than those in our analysis, there still remain qualitatively similar trends in the differences between the two regions: in the best track data of the SIO region, there is an upward trend in the numbers of intense hurricanes since 1980, while in the SPAC basin the trend is far less pronounced. When the trends are corrected on the basis of their degraded-resolution technique, there is only a slight upward trend in the SIO region and a slight downward trend in the SPAC region. The reduction in the upward trend in the south Indian region after reanalysis is consistent with the results of Harper and Callaghan [2006] and Harper et al. [2008]. A point worth raising however, is that the technique and algorithm developed by Kossin et al. [2007] was exclusively trained for the North Atlantic basin. Considering that comparisons in their paper for the SIO and SPAC regions were made against best track estimates maintained by the JTWC, caution must still be taken when comparing their results with those from the BoM data set presented here. Also, it was further noted that interagency best track data comparisons should be done with utmost care [Kossin et al., 2007].

[36] It is important here to reiterate that the BoM best track data were utilized “as provided” and only subjected to the criteria applied without further modification. Still, the dissimilarity in the trends presented in section 3, in particular those that pertain to the severe TCs, imply that characteristic differences observed between the two sectors may not be completely due to data inconsistencies arising from evolving observing practices and analysis techniques, as such systematic biases would typically manifest themselves in time and not space. Furthermore, we believe that by choosing to examine the differences in trends, this analysis has somewhat removed itself from the debate about trends in absolute terms, thereby increasing its validity.

[37] Nonetheless, the question is inevitably raised as to how credible the observed trends are individually within the two sectors (Figures 1, 2, and 3). Simply put, are these trends real or artificial? Recently, Harper and Callaghan [2006] and Harper et al. [2008] outlined the need for a reanalysis of historical TC intensities for the Australian region in keeping with a similar endeavor currently underway for the NAT. Their papers summarized the results of a recent study commissioned by Woodside Energy Limited (WEL), which consisted of an objective review of west Australian TCs and involved a reanalysis of all available archived meteorological data for the period 1968/1969–2000/2001. Overall, the WEL review did uncover a clear induced bias toward underestimating intensities in the period prior to the early 1980s [Harper et al., 2008]. Interestingly, tendencies to underestimate storm strength in the period after 1985 were also found, such that a downward trend in the average maximum intensity of TCs still remained even after the bias was removed [Harper and Callaghan, 2006, Figure 2]. Also, (B. A. Harper Analysis of tropical cyclone climate change trends for the 10−4 Waves Study, unpublished report, 38 pp., Woodside Energy Ltd., Perth, Australia, 2006) specifically reported a 42% increase in the number of SS45 TCs when decade 1989–1998 (41% of all TCs) was compared to the period 1974–1988 (29% of all TCs). This is a considerable reduction from the modified BoM data (designed during the review to remove possible biases and facilitate comparison), which showed a 203% rise: 27% of all TCs were SS45 during 1989–1998, while only 9% were of that category for 1974–1988. An extended analysis to better match the two 15-year periods in the work of Webster et al. [2005] saw the proportion of SS45 TCs for the period 1989–2003 changed slightly to 43%, amounting to a 49% increase between the two periods (B. A. Harper, personal communication, 2007). Overall, B. A. Harper (unpublished report, 2006) reported that 35% of all TCs in the original BoM data set had their intensities increased at least by one SS category when reanalyzed, while 8.7% underwent a decrease (a net change of 26.2%). Ultimately, he concluded that the WEL reanalysis “does not exhibit any trends that cannot be reasonably explained by expected deficiencies in analysis, natural variability or current multidecadal effects.”

[38] The findings of the Woodside review do lessen the claim that a trend in the number and strength of western TCs may be emerging in the record. However, all the intensity comparisons conducted in the review were done with the Saffir-Simpson scale. To assess the validity of the review's overall conclusion to the western sector results presented here, we simply reclassified the WEL data set (B. A. Harper, unpublished report, 2006, Appendix E) according to the Australian cyclone severity scale of Table 1. Comparing with our Figure 1, regression analyses revealed similar statistically significant trends in the average frequency of A45 (upward; p = 0.036) and A3 (downward; p = 0.0001) TCs. In fact, the downward temporal trend in category A3 is very similar to the one found in Figure 1, which suggests that the occurrence trend in that storm class is real. However, the reduced upward trend found in A45 cyclones can indeed be credited to the rise in incidences for that classification prior to 1975 that is also balanced by a sharp increase in occurrences during the 1990s. This was duly reflected in terms of our Figure 2, where we found a smaller linear regression slope when severe TCs in the WEL review sample were analyzed separately. As suggested by B. A. Harper (unpublished report, 2006), the reduced trend can be reasonably explained by the underestimation of maximum storm strengths in the BoM data set particularly for the period prior to 1980. For all TCs, we found that the entire MCP profile was “downward” shifted in the same manner as Figure 2 of Harper and Callaghan [2006]. Nonetheless, both linear regression slopes for average MCP were found to remain statistically significant beyond the 95% level. We do state that while the WEL reviewed data set contains only 183 storms, as opposed to the 266 western sector TCs in our sample, the resemblance of the trends is still worth highlighting.

[39] Combining these with our original results here, one can consequently argue that for the western portion, the intensity trends can be considered real, although exaggerated in the original BoM data due to the time-varying bias identified by the WEL review. Certainly, the original 25-year upward intensity trends post-1980 remain quite robust even after one factors in the overall findings of the review. Therefore, there are reasonable indications that the disparities in the trends are most likely to be true even if one considers that a synonymous time-varying bias may plausibly exist for the eastern subregion. However, this interpretation is based on the earlier stated assumption that methodological changes in forecaster analysis techniques were standardized and concurrently implemented across the board in all three TCWCs. As such, simply extrapolating and accounting for possible similar exaggerations across time in the eastern portion may yet alter the out-of-phase relationship in severe storm type occurrences seen in Figure 1 for example. Nevertheless, this would still be rather speculative at best. Hence, a similar reanalysis for the eastern sector is desirable so that a proper comparison between the two portions can be done. Given the lack of such a review, we cannot categorically conclude that trends in that portion are real without any caveats, nor can we describe them as being fully artificial either, although Holland [1981] did previously indicate that the eastern region may have retained some intensity bias even into the 1970s. We surmise that, by factoring this bias in, the downward average MCP trend (for all TCs) originally seen for the eastern sector in Figure 2 may be removed entirely, while that of the severe TCs may be substantially upward (i.e., decreasing in average intensity).

[40] With regards to the respective duration trends in Figure 3, we consider these to be robust throughout, particularly the upward trends for severe category days in the western sector. Given the strong correlations between severe intensities and days spent at those categories found earlier, we suspect that the overall duration trends shown in Figure 3 remain real even after including the caveats. It is worth noting that the disparities observed between the two sectors, across all ranges, were found despite the potential for misclassification of the strongest storms into weaker categories. Such a possibility was termed as “subclassification” by Fasullo [2006] who demonstrated this to be an induced satellite-sampling-interval bias. Fasullo [2006] argued that the artifact of “subclassification” would manifest itself in prolonged storm durations at the weaker categories, which in turn would induce negative trends and not positive ones over time due to the “disproportionate effect that sampling deficiencies have on the more intense storms.” Furthermore, similar behavioral profiles were seen when landfalling data were included (not shown). The only difference found was in the combined (i.e., basin-wide) average cumulative lifetime of TCs: a minimal, insignificant upward trend was observed when unfiltered data was used, while no trend was seen basin-wide when land effects were removed (not shown).

[41] There still remains the issue of why the trends in the two adjacent Australian formation regions are so different. Although this has not been shown definitively in this paper, it is suggested that these changes are related to similar large trend differences in variables relevant to tropical cyclone intensification, such as SST and low-level relative humidity. Certainly, Emanuel [2007] clearly showed that increases in PDI in the Atlantic since 1980 were due, at least partly, to increased SSTs over that time and an accompanying increase in the maximum potential intensity. It would be instructive to perform a similar analysis on these data to unravel the causes of the clear differences between cyclone trends in the eastern and western portions of the Australian basin.

Acknowledgments

[42] The authors wish to thank Bruce Harper (Systems Engineering Australia) for permission to access and use some of his unpublished results. We are also grateful to Yuriy Kuleshov and Jeff Kepert (both of the Australian Bureau of Meteorology, Centre for Australian Weather and Research) for helpful suggestions on the treatment of the best track data. The authors also wish to thank Greg Holland and an anonymous reviewer for their comments, which helped improve this paper further.