Revisions to the Australian tropical cyclone best track database

Keywords: tropical cyclone, best track database, historical reanalysis, Dvorak technique, satellite analysis, wind-pressure relationships

The Australian tropical cyclone (TC) best track database (BT) has records since 1909 of varying quality and completeness (BoM 2021; Holland 1981; Trewin 2008). To address inhomogeneities and missing data, the Australian Bureau of Meteorology (BoM) has undertaken a series of revisions since 2005 to develop a more consistent and complete TC dataset in the Australian region (Fig. 1). This has been enabled by the expansion of the TC database format to add fields such as wind radii quadrants, Dvorak parameters and many metadata fields as described in the database specifications document (BoM 2021).

Fig. 1.  The Australian Region of Responsibility showing the historical regional boundaries (BoM).

A significant component of this effort was to investigate the consistency of maximum winds (V_m) estimates and extend the record prior to 1984–85. The Dvorak satellite analysis method (Dvorak 1973, 1975, 1984) was first applied in the Australian region in 1973 and remains the primary tool for determining the TC intensity. Remarkably, there have been only relatively minor refinements to the technique, especially since the enhanced infra-red (EIR) method was introduced in the early 1980s. While the Dvorak Current Intensity (CI) metric has been relatively stable in its derivation, the corresponding set of V_m and central pressure (CP) estimates has been quite varied over the Australian region due to different approaches and use of different wind-pressure relationships (WPRs) being employed across the three historical Tropical Cyclone Warning Centres (TCWCs) in Darwin, Brisbane and Perth. This has made it difficult to use the data temporally in any homogeneous way (see Harper (2002) for examples and discussion of WPR variations across and within TCWCs). Courtney and Knaff (2009) described the current Australian standard WPR, hereafter referred to as the Courtney-Knaff-Zehr relationship (CKZ). Table 1 summarises the changes and advances made over the satellite era.

Table 1.  Timeline of events relevant to TC analysis in the Australian region

Reanalysis efforts can be described in the following projects that were conducted over the years indicated below:

1. 2005–07: corrections for internal inconsistencies and comparisons with external datasets for significant differences, gross error checks and duplicate tracks (Trewin 2008).

2. 2006–12: database repair project to investigate TC intensity (1973–2002) using historical records: Foley (2012), van Burgel (2012), Callaghan (2013) and unpublished work by the lead author.

3. 2015–19: reanalyses using the Hurricane Satellite (HURSAT) archive, especially microwave imagery, with emphasis on initial TC intensity and peak intensity for previously identified cases of interest: unpublished work by the lead author.

4. 2016–19: review of structure parameters and application of objective techniques (Advanced Dvorak Technique (ADT) and various structure datasets) to analyse database completeness and homogeneity: Courtney and Burton (2018) and unpublished work by the lead author.

These projects are expanded on in the sections that follow.

The reanalysis work done in 2005–07 (Trewin 2008) focussed on internal inconsistencies and comparisons with external datasets and included:

1. Checks for excessively rapid or irregular movement

2. Observations at irregular times outside standard synoptic times

3. Consistency between CP and wind speed where both were available

4. A test was also carried out for potential duplicate systems.

The BT was compared to the following datasets and information sources:

1. Joint Typhoon Warning Center (JTWC) dataset (covering 1945–46 to 1999–2000)

2. BoM regional datasets held at Queensland (Qld) and Northern Territory (NT) State offices

3. A reanalysis of TCs in northwest Australia carried out on behalf of Woodside Energy (Harper et al. 2008)

4. The published and working notes of Lourensz (1981)

5. Seasonal summaries, either published as standalone documents or in the Australian Meteorological Magazine (covering 1957–58 to 2003–04, except for 1972–78 and 1992–93)

6. Operational charts and archived TC documents.

Some amendments could be attributed to data entry errors in position or intensity. For example, a longitude of 125.8°E amended from 128.5°E (Gertie 1985). More significantly, a number of new systems were included that appeared to be omitted from the BT record by mistake, such as Doris (1980) and Harold (1997). Tina (1990) was named operationally but not included originally in the BT because it was regarded as a monsoonal low but was reinstated as a TC with comments on its monsoon-like characteristics. An unnamed system in 1980 (197 980_15U) was added, as it was included in Lourensz (1981) and also on the JTWC archive, suggesting it likely reached TC intensity. This was confirmed by a subsequent review of HURSAT data. Another addition was the unnamed TC that devastated the Indonesian island of Flores in April 1973 (197 273_15U); data was retrieved from NT archives and reanalysed during the ‘Woodside’ reanalysis.

Particular attention was given to a number of systems in the JTWC dataset, mostly between 1945 and 1960. The majority of these systems were near the boundaries of Australia’s current area of responsibility, especially in the west. At the time, Australia did not systematically report on or issue warnings for cyclones west of 105°E. For example, Gina (1968) combines BoM data east of 105°E with JTWC data west of 105°E. The JTWC dataset at that time had no intensity indicators, and assumptions therefore had to be made about whether a system was likely to have reached cyclone intensity within the Australian region based on its lifetime and the behaviour of more recent systems in the JTWC database. JTWC systems without other supporting information were included if they met the following criteria:

1. A lifetime of at least 36 hours

2. Did not leave the Australian region within 24 hours of formation

3. Not in close proximity to the coast (it is assumed that near-coastal systems would already have been considered).

For most of these systems it will be impossible to determine definitively whether or not they reached TC intensity, but it is qualitatively estimated that the systems included have at least a 50% probability of having reached TC intensity. This will partially address the negative bias in pre-1970 cyclone numbers over the far western parts of the Australian region, although it is likely that many systems were still missed.

Some systems were also deleted, including some that were not over the ocean at any time during their lifetime, some where the working note of Lourensz indicated that he did not consider these systems to have reached cyclone intensity and some that were outside the Australian region. Others were merged if they were deemed to be duplicates or if they could be analysed as a continuous system. In some cases, this required amending the names (either by merging names or deleting a name), such as Beverley-Eva, Fiona-Gwenda and Doris-Gloria. Some names were added for TCs that entered the region from the Pacific in the late 1950s and early 1960s (Agnes, 195 556_13U, Bertha, 195 859_04U, Beatrice 195 859_05U, Connie 195 859_08U, and Annie 196 263_04U) before routine naming of systems in the Australian region began in the 1963-64 season.

In general, this process did not reassess intensity, except for a few cases where internal consistency checks revealed clear errors or where an individual cyclone had been reassessed by the relevant region (e.g. Pam 1974 peak CP changed from 970 to 925 hPa). Appendix 1 lists some of the more significant changes during this process.

While the database mostly contains fixes at the standard 00, 06, 12 and 18 UTC times, from the 1995-96 season in the western region, fixes were at 01, 07, 13 and 19 UTC, and then, from 1997, this changed to 04, 10, 16 and 22 UTC until being standardised from the 2001–02 season. The addition of the standard fix times was flagged for future consideration.

From approximately 2006 to 2012, a ‘database repair’ project investigated TCs between 1973 and 2002 to provide a complete dataset of V_m and Dvorak CI estimates: Foley (2012), van Burgel (2012), Callaghan (2013) and unpublished work by the lead author. Previously the only comprehensive gauge of TC intensity over the Australian region had been the CP value, but these estimates are inhomogeneous owing to the application of different WPRs. The maximum 10-min mean wind had been included in the database progressively since around 1984, but for a period in the 1980s, some estimates had been incorrectly derived from CI estimates. Comparisons with overseas datasets from JTWC and neighbouring TCWCs were conducted.

The Dvorak method is used globally for estimation of TC intensity, including the Atlantic basin where aircraft reconnaissance is conducted and the technique is used and validated. Briefly determining the intensity involves:

1. Analysing the cloud patterns in infra-red (IR) and visible imagery via a flowchart of decision points to determine the Dvorak term the Final T-number (FT) and then apply rules to determine the term CI number, as described in Dvorak (1984).

2. The CI number is then used to obtain a value of V_m via a look-up table relating the two quantities.

3. The final V_m can be adjusted from the Dvorak according to other inputs, such as surface observations, scatterometry, microwave patterns and objective estimates, such as ADT (Olander and Velden 2019) and SATCON (Velden and Herndon 2020), as summarised in appendix 1 of Courtney and Burton (2020). However, in the Australian region, it is rare that surface observations influence the intensity estimate, and scatterometry, microwave and objective techniques have only been available since the early 2000s, so for TCs from 1973 to the early 2000s, the Dvorak technique was the only method to estimate the intensity in the absence of observations. Deviations from Dvorak are more likely near landfall when there is a greater chance of observations and impact assessments.

4. To obtain the CP, a WPR is applied.

Key issues are the quality of Dvorak estimates and the consistency in its application through time and between analysts. It is truly remarkable and valuable that the Dvorak technique has worked so well for so long. For the Australian region, because of the very few conventional observations to validate the technique, application of the method has been based on validation from the Atlantic, where the deployment of superior observational platforms provides some ground truth.

Reviews of the technique and its strengths and weaknesses are found in Burton (2005), Velden et al. (2006) and Knaff et al. (2010). The latter study validated Dvorak intensity estimates in the North Atlantic against BT data where aircraft reconnaissance data were available, confirming the accuracy of the technique and noting ‘the Dvorak technique has a tendency to slightly underestimate intensities of storms receiving CI estimates ranging from 2.5 to 3.5 by an average of 1.5–2.5 kn,¹ overestimate intensities of storms receiving CI estimate ranging from 4.5 to 5.5 by 1.5–2.5 kn, and underestimate intensity estimates of storms receiving CIs ranging from 6.5 to 7.0 by 4–9 kn.’

While it is not possible to be definitive about biases in the technique in the Australian region, a comparison of BT and the ADT during the objective reanalysis study discussed in section 5 (Courtney et al. 2020), remarkably showed there was just a 0.2 kn bias for the 2003–16 era.

Over the years, the process to determine the Dvorak CI (step 1) has remained generally constant, although differences in applying the Dvorak technique were later addressed during the objective reanalysis project discussed further in section 5.

Being an inherently subjective process, there are many reasons for variations in CI estimation throughout the historical record. Despite this the robustness of the technique has afforded it widespread acceptability in lieu of direct measurement.

Intensities are resolved at 5 kn intervals so that a single CI estimate can match multiple V_m alternatives, allowing the analyst to apply discretion. For example, a ‘weak’ CI of 4.5 could be matched to 65 kn while a ‘strong’ 4.5 could be allocated to 75 kn, and generally, the intensity trend is smoothed between fixes.

However, there are some examples that seem to indicate some discrepancies at the intense end of the scale. Comments in the TC Aivu (1989) report (BoM 1990) indicated that a CI of 6.5 corresponded to a pressure of 935 hPa, but the BT winds at that time were just 95 kn. This happened again during TC Rewa (1993–94) when a pressure of 920 hPa matched 90–95 kn on 2–3 January, even though the report indicated 110 kn; although, for the second intensity peak on 16 January, 920 hPa was matched with 110 kn in the BT. Such deviations may have been the result of interpretations of observations but there is insufficient detail even in the reports. For both Aivu and Rewa, the intensity was adjusted to 110 kn following reanalysis.

Occasionally, the availability of observations appears to have influenced the Dvorak estimate, at odds with the ‘pure’ satellite nature of Dvorak. For example, for TC Justin (1997), a large system, the initial operational Dvorak estimates were as high as 3.5–4.0 with a remark ‘primarily from observations’. A reanalysis of imagery suggested that Dvorak values during this period were 2.5–3.0.

There are situations where Dvorak is not valid or unreliable, and this reinforces the fact that even re-engineered CI numbers could not be applied blindly to the Australian TC database simply based on CP or V_m values. This applies to the following situations:

1. Over land, either during the development stages before moving over water or after making landfall and weakening. Dvorak is not generally applied during overland trajectories, although the Dvorak estimate is often still considered in the absence of other information in the hours following landfall or when the track remains close to the coast.

2. Moving outside the tropics and losing their tropical cloud characteristics. Dvorak will underestimate the intensity and is not applicable for most sub-tropical systems or systems undergoing extra-tropical transition (ETT), such as TC Alby, 1978 (Fig. 2), documented in Foley and Hanstrum (1994).

3. Monsoonal systems exhibit fundamentally different characteristics to TCs in their dynamics, structure and cloud signatures, yet they can generate gale (34 kn) to storm-force (48 kn) winds. These can be included in the Australian record even though Dvorak does not strictly accommodate such a genre.

4. Strong synoptic forcing can result in higher winds than Dvorak suggests, particularly during formation and when moving south towards the sub-tropics, associated either with a strong monsoon to the north or a strong ridge to the south. Both of these features occurred during Kelvin (1991) in the Coral Sea, and while the maximum CI was 3.0, Willis Island recorded V_m to 65 kn. Estimated winds were at least 55 kn for an extended period from 25 to 27 February and again on 3–4 March. Observations, particularly scatterometry since 2001, can support BT V_m being higher than the associated Dvorak CI value, but the situation is more ambiguous in the absence of such observations. Historically, Dvorak remained the primary metric, but on some occasions some allowance is made so the V_m is higher than the CI. For example, Betsy and Daman (1992) were both large systems that were estimated to remain at 60 kn intensity even though the CI fell to 3.5 as they moved south and eventually transitioned into the mid-latitudes.

5. Eyewall replacement cycles (ERC). Dvorak estimates in general do not resolve such changes in the eyewall that are best viewed through microwave imagery, as discussed in Sitkowski et al. (2011).

Fig. 2.  Satellite imagery of TC Alby (1978) illustrating the very different cloud characteristics of a TC undergoing ETT – cloud patterns not accounted for in the Dvorak technique (BoM).

The database fields cyc_type and max_wind_spd_method (BoM 2021) allows for these cases to be identified, and values have been added to the record since 1973 during the reanalysis process.

The relationship between CI and V_m (step 2) has not changed since Dvorak (1975), other than a transition from 1 to 10-min wind conversion factors and some rounding issues. However, the practice during the early years was to arrive at the CP, and the CI and V_m values were not always saved. Furthermore, the WPR applied varied and was revised for each region to reflect experiences. Table 2 shows the range of WPRs used in the Australian region since the introduction of the Dvorak technique, as discussed in detail in Harper (2002).

Table 2.  Conversions from Dvorak CI to V_m and CP from different WPR. Winds are in kn and CP is in hPa

While all regions initially applied Dvorak 1973 and then 1975 WPRs, the western region adopted the Atkinson and Holliday method (A&H) from Emma (1984) onwards; the northern region applied a local version from 1985, L&M, after Love and Murphy (1985), on the basis that the region has cyclones on the smaller end of the size distribution and have relatively high CPs; while the eastern region adopted their local WPR in 1984. The latter had two scales for ‘normal’ and ‘small’ TCs, the normal being somewhat similar to A&H whereas the small are more like Love and Murphy (1985). The different methods are a significant factor in the inconsistencies in the TC record.

As a result, although the base Dvorak technique was providing consistent estimates, the diagnosed values of V_m and CP were often lost or inconsistent. It is this inconsistency in the TC record that has been reviewed and repaired, as summarised by Foley (2012) and van Burgel (2012). The WPR for each TC was scrutinised, which uncovered many variants from that described above. For a period from 1976–77 to 1977–78 in the western region, the mid-point of the northwest Pacific and Atlantic was used, except for Trudy, Vern and Alby.

The original Dvorak CI to wind relationship used a 1-min mean wind. As noted by Harper (2002), a conversion factor of 0.88 was applied to correct to the Australian and World Meteorological Organization standards 10-min V_m. When winds began to be reported in the TC database from around 1984, it appears that the western region used 1-min winds instead of 10-min winds. The most likely cause would be from translating Dvorak CI numbers using a 1-min wind table from Dvorak’s guidance material. This gave around a 10% positive bias in wind values from 1984 to nearly 1992, with some exceptions. All these V_m estimates were corrected with appropriate checks for extra information considerations.

The task was to provide a complete dataset of V_m and Dvorak CI estimates from 1973 to 2002. Rather than perform a complete reanalysis of all TCs from satellite imagery, the approach was to build on the skill found in the original estimates and then consider the exceptions and inconsistencies detected. This is based on the knowledge that the Dvorak technique has been applied assiduously and skilfully during the post-event BT analysis by all three TCWCs and upon the premise that the application of the technique has been consistent throughout the record, as discussed in the previous section.

However, CI data were not incorporated into the BT until 2003 nor were they routinely stored in a digital archive. The retrieval of these CI data involved a search through archives in the different states in order to glean information, looking not only for the operational and BT Dvorak ‘worksheets’ that were a part of the determination process but also any evidence of what WPR and Dvorak tables were being used for in a given season in each TCWC.

A total of 352 TC over the entire Australian region contained in the database between 1973 and 2002 were reviewed. This period encompasses the most volatility in respect of WPR for all TCWCs (see Table 1).

Of the 212 TCs identified in the western region record between 1973 and 2002, the search revealed 102 cyclones where BT CI or FT data were retained in the archive. There are subtle differences imposed by Dvorak between CI and FT values in respect of the intensity of a TC, particularly during weakening, which required consideration.

Another 20 cases were found where the Working Track (WT) worksheets only were available. While these Dvorak operational worksheets are useful as a guide to the intensity of the TC, they do not constitute the final BT version, as some significant adjustments can be and were made in the post-event analysis. Therefore, WT-CI numbers cannot be related directly to the TC CP found in the database. The remaining 90 TC archived cases in the western region dataset that were examined revealed no readily locatable record of Dvorak data.

Similar searches of the BOM archives in Darwin and Brisbane yielded reasonable results. For the 38 northern region cyclones, 10 BT intensity analyses were located, eight cases had no Dvorak data and the remainder offered operational Dvorak material of variable quality. As a lot of northern region cyclones form very close to land and have short life spans, a number of operational sheets only contained sparse data entries. The Brisbane archive had data for all but 19 of the 110 cyclones. The majority were operational intensity estimates, but nine were BT estimates, mostly in the early days of Dvorak when CI or FT were tabulated in official summary reports. In lieu of no local Dvorak data being identified, Satellite Bulletins from the US JTWC or NOAA Satellite Services Division were noted as an additional guide if they were available.

While it has been fruitful to collect these Dvorak data, it became apparent that merely finding the CI or FT values in the archive and simply applying them directly to the database would not be the most appropriate or encompassing approach because of the different WPRs being used. A ‘reverse-engineering’ approach was proposed where the CIs are regenerated by applying the known Dvorak relationship used in the TCWC to the CPs given in the archive. Examination of CPs in the database (and after about 1984 supplemented with V_m values) was used to identify the key pressure values that corresponded the post-analysis BT CI number (Table 3). This gave confidence that TCs with no locatable archived Dvorak data can also be reverse-engineered to derive a relevant set of CI numbers given that the relationships used to derive those values are known.

Table 3.  Excerpt from the report of TC Jean (1973) showing the Dvorak FT and CP estimates for each of the Environmental Science and Services Administration-8 (ESSA 8) satellite images received
GMT is the same as UTC and mb is the same as hPa (www.bom.gov.au/cyclone/history/pdf/jean.pdf)

All TCs in the Australian region dataset were processed in this manner. The key pressure values – points of correspondence between CP and/or V_m and CI number according to the Dvorak table being used – were identified for each TC and the dataset was then interpolated to obtain the CI numbers at each time step in the cyclone’s history (Table 4). Once this was in place, the prescribed V_m values could be documented.

Table 4.  Excerpt from the BT record of TC Jean (1973) showing the interpolation of Dvorak CI numbers (last column) across the temporal record using the key threshold numbers determined from the Dvorak CI/P relationship
Time is in UTC; RE, reverse-engineered

Although observational data is traditionally sparse over the area where TCs occur, occasions do arise when the intensity estimate is influenced by a surface observation. The database field max_wnd_spd_method (BoM 2021) allows for these cases to be identified. For the majority of cases, this field will be ‘Dvorak estimate’ but could also be land estimate or derived from pressure measurement, for example.

While the initial review done in 2005–07 used different datasets to make additions and corrections, the database repair process made further alterations. About 15 TCs had major portions of their track added in sections of at least 6-hourly frequency. Examples were the sub-tropical phase of TC Herbie (1988), which impacted the Shark Bay area of Western Australia, and TC Hector (1986) and TC Jason (1987) where the missing northern region sections were located and inserted. Some TCs abruptly commenced when they were near TC intensity already, so fixes were added to better describe the origins.

Further typographic and systemic issues were addressed. For example, some winds erroneously mixed ms^-1 with kn, such as Steve (2000) in the northern region.

Comparisons with other datasets identified cases warranting particular attention using the methodology described in the following sections. As mentioned earlier, the revised BT was compared with theJTWC archive and the 2002 reanalysis study for western region TCs performed on behalf of Woodside by ex-BoM meteorologists (Harper et al. 2008). A consistency check against these two datasets highlighted exceptions that were investigated. Table 3 of the Woodside report documented 35 cases when their CI numbers differed from the derived CI in the BT (as of 2000) by more than 0.5. Following the database repair process, this number reduced to 13 and then further to just three by the end of the reanalysis, as discussed in section 4.

For some, such as Alex (1990), the reanalysed BT agreed with Woodside, and for many others the final CI was somewhere between the earlier BT and Woodside, whereas for others the earlier BT was supported.

There were only three events (Maud 1973, Linda 1976 and Carol 1980) where the difference was 1.0 or more. BT winds for both Maud and Linda were estimated higher than the original CI of 2.5 would suggest, but the imagery was not available to review. For Carol, reanalysis of the imagery supports the BT CI estimate of 6.5 over the Woodside value of 5.5. TC Tina (1990) was categorised as a monsoonal system without any BT CI estimates even though the Woodside CI was estimated at 4.0.

The JTWC dataset (in conjunction with the Neumann reanalysis set) as well as those areas of intersection with the La Reunion Regional Specialised Meteorological Centre’s (RSMC) database in the western region and RSMC Wellington in the eastern region can be compared using the US National Climate Data Center International Best Track Archive for Climate Stewardship (IBTrACS) website (Knapp et al. 2010), although it must be remembered that JTWC use a standard of 1-min mean wind compared with the BoM 10-min mean wind and does not provide CP values. CP is used by RSMC’s Wellington and La Reunion and in the Neumann reanalysis set as the principal intensity metric.

The IBTrACS site lists the 25 TCs in each basin with the largest deviations or ‘disagreements’ between contributing agencies for the same TC. For the South Indian Ocean, there were eight TCs that occurred after 1973. JTWC winds were less than the BT for Hazel (1979), Amy (1980), Fred (1980) and Jacob (1985), and reanalysis mostly confirm BT, especially as Hazel and Amy both had documented land impacts. Jacob was updated upon reanalysis and is now consistent with JTWC. The upgraded BT winds are also within tolerance with JTWC for Leon (1989) and Jane (1992), whereas for Idylle (1973) and Annette (1984), the differences are with La Reunion and well west of 90°E outside the Australian region and were not investigated.

Similarly, in the South Pacific Basin, only five TCs were listed in the top 25 differences after 1973 and within the Australian region. Two of these (Kerry and Rosa, 1979) were well scrutinised at the time, having National Oceanic and Atmospheric Administration (NOAA) aircraft reconnaissance data, suggesting that the BT should have significant credibility. Elinor (1983) was reanalysed to be similar to RSMC Wellington at peak intensity.

The other two TCs, Simon (1980) and Harry (1989), continue to have significant points of difference between agencies. TC Simon (1980) was analysed as a more intense cyclone on 25 February by RSMC Wellington (930 hPa) than either BoM (reanalysed at 85 kn) or JTWC (75 kn). While the imagery would support the BT, there is sufficient ambiguity in the observations not to discount a higher intensity. Observations included Heron Island (70 kn on 25 and 26 February), Sandy Cape Lighthouse: (75 kn at 00 UTC 27 February) and ship ‘Poolta’ off Fraser Island (100 kn at 00 UTC 27 February) but of unknown quality. TC Harry (1989) was held at its most intense by RSMC Wellington (lowest CP 925 hPa), which is longer than BoM or JTWC, but reanalysis confirms weakening. The reanalysis also increases peak intensity from 100 to 110 kn, closer to JTWC’s peak of 117 kn (when converted to 10-min wind).

In reviewing the outcomes of the previous projects, it was clear that some TCs warranted extra attention. The HURSAT satellite archive (Knapp and Kossin 2007) provided Advanced Very High Resolution Radiometer (AVHRR) and Geostationary Meteorological Satellite (GMS) imagery from 1978 as well as microwave imagery (Special Sensor Microwave Imager (SSMI) from 1988 and Tropical Rainfall Measuring Mission (TRMM) from 1998) that was not considered in the BT process until about 1999 when NRL provided operational access via their website. Fig. 3 is an example of the 37 GHz and 85 GHz SSMI microwave imagery alongside a visible image for Barisaona (1988) that was used to increase the intensity. Callaghan (2013) selectively reanalysed notable parts of 31 TCs in the eastern and northern regions. Results from a more comprehensive reanalysis of TC Tracy by Courtney and Shepherd (2010) were also incorporated. Adjustments to the BT were made on the basis of these investigations.

Fig. 3.  Satellite imagery for TC Barisaona from HURSAT on 11 November 1988. (a) SSMI 37 GHz at 0018 UTC. (b) SSMI 85 GHz at 0018 UTC. (c) AVHRR visible image at 0808 UTC.

The findings from previous reviews and reanalyses meant that all TCs in the HURSAT era could be reviewed by the lead author. This process resulted in substantial changes to the intensity and extra fixes being added. Appendix 2 is a summary of selected reanalysed V_m estimates, and Appendix 3 has extended descriptions for a smaller selection: Pam (1974), Tracy (1974), Trixie (1975), Dominic (1982), Elinor (1983), Kathy (1984), Jacob (1985), Barisaona (1988), Orson (1989), Felicity (1989), Kelvin (1991), Neville (1992), Rewa (1993–94) and Rosita (2000).

Overall, the changes mostly resulted in an increase to the intensity. Those cases having a notable increase in peak intensity include Ivor (1990; 55–90 kn at landfall); Barisaona (1988; 30–60 kn); Neville (1992; 80–110kn); Rewa (1993–94; 80–110 kn on first peak); Nina (1992; 60–80 kn at Cape York landfall); Aivu (1989; 95–110 kn); Rosita (2000; 100–120 kn at landfall), Dominic (1982; 90–115 kn at landfall) and Felicity (1989; 50–75 kn). The cases of Winifred (1986; 70–85kn; BoM 1986), Aivu (BoM 1990) and Rosita (2000) underwent particular scrutiny given they were heavily studied at the time with official reports issued. Several events were revised to commence TC intensity at an earlier stage, notably Barisaona (1988), Kelvin (1991), Neville (1992) and Theodore (1994), and all three required earlier fixes to be added.

Microwave imagery permitted the detection of ERC, which typically resulted in an initial increase in winds but then a reduction of winds when the secondary eyewall dominated. Notable cases included Esau (1992), Drena (1997), Thelma (1998) and Vance (1999). Microwave imagery was instrumental in discerning intensity characteristics during Dvorak covered centre patterns, which often resulted in many previous analyses of CI of 5.0 being reduced. For example, Jacob (1985) was held too high at 5.0 for too long as was a tendency during the 1980s after the adoption of the

EIR technique. During this process winds were often reduced during weakening phases owing to the change in the Dvorak weakening rule to use 6 h instead of 12 h for holding the intensity higher.

Courtney et al. (2020) details the intensity decreases from the application of the Dvorak covered centre patterns, especially during the 1980s, and the change in the weakening rule.

As noted earlier, cyclones for much of the 1970s were named when the CI reached 2.5 based on the application of the 1-min wind Dvorak relationship. Each of the seven cases (Paula (1973), Natalie (1973), Maud (1973), Wanda (1974), Linda (1976), Lily (1977) and Nancy (1977)) were investigated in detail. Aside from the reasons of infrequent satellite passes, for all of these events, there are valid reasons to indicate gales were present either because of observations and/or synoptic forcing. In some cases, it is possible that gales may not have extended more than halfway around the centre, such as for Lily and Nancy in 1977.

In addition, Ernie (1989) in the Coral Sea was reanalysed to have failed to reach a CI of 3.0, and there were no observations available to support TC intensity. However, the reanalysis concluded that it was possible that gales occurred around the system for a 24-h period to justify TC status.

TC Pam (1974) was downgraded from 110 kn (CI = 6.5) to 105 kn (CI = 6.0) upon reanalysis as described in Appendix 3.

The Objective Tropical Cyclone Reanalysis (OTCR) project (Courtney and Burton 2018; Courtney et al. 2020) aimed to produce a homogeneous TC dataset by applying the ADT to IR imagery for all Australian TCs between 1981 and 2016. The ADT was run in two modes: (1) incorporating passive microwave imagery (PMW), which is the current operational mode and (2) without PMW given that it only became available in the 1990s. Comparisons of the ADT with PMW and BT for the most accurate BT era between 2003 and 2016 showed that the bias was just 0.2 kn, thereby providing confidence in the ADT dataset. Differences of at least 20 kn identified in 12 events during this period and were investigated further. In all but two of these cases, observational and microwave imagery supported the BT. For the other two cases, Fay and Wati, the BT winds were shown to be too high during weakening phases based upon a re-examination of satellite imagery and applying the Dvorak 6-h weakening rule.

Upon further analysis of the ADT without PMW, a series of adjustment rules were incorporated to address shortcomings of the dataset and to cater for handling missing images in the 1980s, as described in Courtney and Burton (2018). Following these adjustments, there remained 30 cases (1981–2016) having differences of at least 30 kn. These cases have since been investigated in response to the project’s recommendations. For 11 of these (Fran (1992), Nina (1993), Katrina (1998), Chris (2002), Larry (2006), Floyd (2006), Glenda (2006), Hamish (2009), Ului (2010), Kate (2014) and Quang (2015)), a reanalysis of HURSAT imagery supported the higher BT winds. For many of these cases, ADT did not sufficiently resolve the eye pattern, sometimes during rapid intensification.

For seven cases (Elinor (1983), Lance (1984), Felicity (1989), Kelvin (1991), Betsy (1992), Justin (1997) and Dylan (2014)), the BT V_m was higher than satellite imagery suggested because of synoptic enhancement, which was supported by observational evidence, particularly in the Coral Sea.

For seven cases (Neil (1981), Paddy (1981), Naomi (1983), Isobel (1985), Victor (1986), Kay (1987) and Cathy (1998)), the BT V_m was amended to be lower, mostly during weakening. This may be partly attributed to a conservative weakening approach, which was most evident in the 1980s.

For five cases where the BT V_m was lower (Bernie (1982), Jane (1983), Polly (1993), Sharon (1994) and Norman (2000)), a reanalysis using HURSAT indicated that the BT V_m required amending to be higher based on the combination of microwave patterns and revised Dvorak classification. The revised Dvorak CI was determined to be higher in the intensifying period for Bernie and Jane, whereas for Polly, microwave imagery resolved the contraction of an eye that was not identifiable on IR or visible (Vis) imagery.

The OTCR project also recommended the systematic application of the early microwave data to the BT V_m, which has mostly been achieved using the HURSAT dataset. This study also raised the possibility that the BT may have a high bias for the 1981–2003 era because of the application of the 12-h weakening rule that was adjusted to just 6 h from about 2003 and recommended. This has since been addressed for the above cases having major differences, but it is noted that most of those changes were quite minor on the order of 5–10 kn for short periods.

The OTCR project concluded that the BT intensity from 2003 onwards is the highest quality dataset, while users requiring a longer homogeneous record could use the adjusted ADT dataset from 1989 to 2016.

The major focus of reanalysis efforts has been on intensity (V_m and CI), however other parameters were also considered to different extents. The following database fields were completed during the final reanalysis process: surface_code, cyc_type, max_wind_spd_method and the comments field when appropriate. Other fields, such as wind gust, were partially inserted (BoM, 2021). The OTCR project examined wind radii (radius of gales (R34), storm-force (R48) and hurricane-force (R64)), CP, radius of maximum winds (RMW), environmental pressure (P_e) and radius of outermost closed isobar (ROCI).

The BT includes R34, R48 and R64 winds by quadrants (NE, SE, SW and NW). Values have been routinely included since 2003, with some early exceptions. Aside from infrequently available surface observations, gale radii were determined by scatterometry supported by microwave, Vis and IR imagery. QuickSCAT (QSCAT) imagery was only incorporated into the BT process in 2001 and ASCAT commenced in November 2007.

Earlier wind radii estimates were heavily biased towards available surface observations, including ship observations and otherwise inferred from cloud patterns and extent in IR imagery. The first values of R34 in the BT occurred in the 1984–85 season in the western region and initially on an opportunity basis when a surface observation was available, but then it was used more regularly from about 1988.

The early data has some very large values that included synoptically enhanced gales not directly associated with the TC circulation and were biased to use with general values, such as 60, 90, 120 and 150 nm. A comparison of the two eras shows a reduction in R34 estimates from 173 km (pre-2001) to ~140 km when scatterometry and microwave was available. The decrease agrees with anecdotal evidence from analysts who conducted BT in both eras. The recent average is consistent with the study using QSCAT imagery from 1999 to 2009 (J. C. L. Chan pers. comm. 2018) that found the average gale radius in the Australian region was 145 km.

While some historical data has been updated on an ad hoc basis during the reanalysis period, there has been no systematic process to reassess wind radii prior to 2003.

The OTCR project attempted to derive estimates of gale radius but concluded that none of the objective methods reviewed had sufficient skill to extend the reliable record of R34. Hence data from late 2003 onwards should be considered as a reliable dataset of sufficient accuracy and completeness.

As previously discussed, estimates of CP are derived from V_m via a WPR. The current CKZ method is based upon V_m, latitude, P_e, translation speed and gale radius (size parameter). The unavailability of reliable gale radii estimates prior to 2004 prevent applying this method to the earlier era, and hence, CP estimates have not been updated.

Values for RMW have only routinely been included in the BT since 2004 and are estimated by a combination of available observations, radar, scatterometry, microwave supported by IR and Vis imagery. Although the HURSAT microwave dataset, in conjunction with more infrequent radar and observations, can be used to estimate RMW, this would only provide sporadic estimates and was out of scope of the reanalysis. Potentially, missing values could be derived by applying a climatological relationship of V_m and latitude. Further details on RMW determination and BT values are discussed in Courtney and Burton (2018).

Values for P_e and ROCI are complete from 2004 but only sporadically prior to then. The OTCR project identified a dataset from 1979 of POCI derived from the European Centre for Medium-Range Weather Forecasts reanalysis dataset that could be used to insert P_e values, but this has not been implemented. The same study noted that the accompanying ROCI dataset underestimated, when compared to BT, and were not suitable.

As discussed in the OTCR project (Courtney and Burton 2018; Courtney et al. 2020), the BT dataset from 2003 is the period of highest quality, aside from CP when 2007 onwards is the best quality. Undoubtedly the efforts outlined in this report have made the intensity dataset far more homogenous, especially since 1981. This year is significant because of the improved satellite resolution, both temporally and spatially, and the consistent application of the EIR Dvorak technique. In particular, the eye pattern for EIR generally results in higher intensity than the visible method, especially for the smaller TCs characteristic of the Australian region. The significant improvements in satellite quality and processing, especially since the 1980s, produced more accurate analyses, but it is difficult to quantify any bias in the earlier era arises given that that there are factors to both increase and decrease intensity estimates.

Improvements to the database permit more accurate climatological studies. For example, using a 2007 dataset before any changes were made (as shown in Fig. 4), the number of severe TCs (hurricane-force winds above 63 kn) has increased by 20 (almost all between 1972–73 and 1984–85) compared to the previous method of using 970 hPa as the threshold (e.g. Dare and Davidson 2004). As in previous studies, the overall number of TCs has decreased in time, with the average over the entire record being 10.7, but decreased from 12.2 in the first half of the record to 9.3 in the second half of the record. The percentage of severe TCs to all TCs is 50 percent, and this slightly decreased across the record from 51.6 to 46.6 from the first half to the second half of the record.

Fig. 4.  Seasonal numbers of TCs and severe TCs compared to the number of severe TCs using CP of 970 hPa or less using a 2007 dataset prior to any database changes being made.

The objective of a series of projects from 2005–19 has been to generate a more homogeneous and consistent TC BT dataset for the Australian region, which could be reliably used for weather and climate research and other TC-related work. Specific efforts have:

1. Removed inconsistencies from the dataset.

2. Determined a complete record of Dvorak CI values since 1973–2002 by recovering archived records and applying a reverse-engineering process using CP and the various WPRs used across the record. Where necessary, V_m values were then updated to arrive at a more homogeneous and consistent set of V_m values. Validation of the new dataset (a) by experienced, skilled analysts, and (b) against other existing BT estimates, indicate that the new values are much more consistent.

3. Reanalysed each TC using the HURSAT archive, especially microwave imagery, with emphasis on initial TC intensity and peak intensity for previously identified cases of interest.

4. Applied the objective techniques (ADT and various structure datasets) and reviewed structure parameters to analyse database completeness and homogeneity.

For many adjustments, future work could include:

1. Re-deriving appropriate CP estimates, following investigations into appropriate size estimates (POCI and R34).

2. Ensuring there are fixes at the standard times: 00, 06, 12 and 18 UTC. From the 1995–96 season in the western region fixes were at 01, 07, 13 and 19 UTC, and then from 1997 this changed to 04, 10, 16 and 22 UTC until being standardised from 2001–02 season.

3. Further reanalysis to add or adjust other parameters, including wind radii and metadata.

Furthermore, the database quality would benefit from closer collaboration between the BT analysts and users of the database through regular (annual) workshops and database updates being documented and made available. The availability of annual database versions would allow the impact of the updates to be assessed.

The authors declare no conflicts of interest.

The support of (a) the Australian Climate Change Science Program and (b) the US Office of Naval Research, under Award No. N000141010139, is very gratefully acknowledged for elements of this study.