Consolidating historical instrumental observations in southern Australia for assessing pre-industrial weather and climate variability

Abstract

Human-induced climate change has resulted in long-term drying trends across southern Australia, particularly during the cool season, with the most pronounced impacts observed in the southwest since the 1970s. Although these trends have been linked to changes in large-scale atmospheric circulation features, the limited number of daily weather datasets that extend into the pre-industrial period have so far prevented an assessment of the long-term context of synoptic-level changes associated with global warming. To address this need, we present the development of the longest sub-daily atmospheric pressure, temperature and rainfall records for Australia beginning in 1830. We first consolidate a range of historical observations from the two southern Australian cities of Perth and Adelaide. After assessing the quality and homogeneity of these records, we verify their ability to capture the weather and climate features produced by the Southern Hemisphere’s key climate modes of variability. Our analysis shows the historical observations are sensitive to the influence of large-scale dynamical drivers of Australian climate, as well as the relationship between southwestern and southeastern Australia. Finally, we demonstrate the ability of the dataset to resolve daily weather extremes by examining three severe storms that occurred in the nineteenth century associated with westerly storm tracks that influence southern Australia. The historical dataset introduced here provides a foundation for investigating pre-industrial weather and climate variability in southern Australia, extending the potential for attribution studies of anthropogenically-influenced weather and climate extremes.

1 Introduction

Historical meteorological observations from long climate monitoring stations around the world are critical for developing a baseline of natural, ‘pre-industrial’ climate needed to better understand the impact of human-induced climate change (Hawkins et al. 2017; Brönnimann et al. 2019; IPCC 2021a). The Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report uses 1850–1900 as an approximation for pre-industrial climate conditions, as this is when many surface observing networks begin, providing sufficiently precise and continuous measurements on a near-global scale (IPCC 2021a). Climate model simulations of the historical period also begin in 1850, making it a convenient and practical start date for examining climate variations in a period with minimal anthropogenic influence (IPCC 2021a).

Recovering daily weather observations from the nineteenth century provides an opportunity to examine how changes in mean-state climate conditions have influenced recent weather and climate variability and extremes (e.g. Alexander and Power 2009; Wang et al. 2013; Ashcroft et al. 2019; Gergis et al. 2020; Ritman and Ashcroft 2020; Slivinski et al. 2021). The international Atmospheric Circulation Reconstructions over the Earth (ACRE) initiative is a global effort to coordinate data rescue activities that contribute to climate applications needed for practical management associated with areas like weather and climate risk assessment (Allan et al. 2011).

To date, the majority of pre-1900 instrumental records identified by ACRE and other initiatives have been located in continental Europe, with substantial data gaps remaining in the tropics and the Southern Hemisphere extending well into the twentieth century (Brönnimann et al. 2019). The lack of Southern Hemisphere records limits scientific understanding of long-term climate variability in the region, making it a high priority for data recovery efforts (Nash and Adamson 2014; Ashcroft et al. 2016a; Lorrey and Chappell 2016; Picas et al. 2019; Brönnimann et al. 2020).

To address this priority, the Australian chapter of the international ACRE data rescue initiative (Allan et al. 2011), ACRE Australia, coordinates efforts to recover and digitise historical meteorological observations from Australian land areas and surrounding oceans (Ashcroft et al. 2016a). Previous efforts have recovered historical weather records from south-eastern Australia (SEA) (Alexander and Power 2009; Alexander et al. 2010; Gergis and Ashcroft, 2013; Ashcroft et al. 2014a, b, 2019, 2021; Gergis et al. 2018, 2020; Bridgman et al. 2019), and most recently, south-western Australia (SWA) (Gergis et al. 2021).

In this paper, we contribute to the ACRE Australia initiative by consolidating southern Australian meteorological data rescue efforts by digitising and statistically evaluating the earliest atmospheric pressure, temperature and rainfall observations for Perth and Adelaide in southern Australia. These data provide a valuable new opportunity to assess the long-term context of recently observed changes in Southern Hemisphere atmospheric circulation and their influence on timescales encompassing multi-decadal climate variability to daily weather extremes (Hope et al. 2010; Cai et al. 2012; Frederiksen et al. 2017; Burls et al. 2019; Garreaud et al. 2020; IPCC 2021b). These new historical weather observations are also useful for verifying and/or calibrating other multi-century datasets including palaeoclimate reconstructions of regional climate variability (e.g. Cullen and Grierson 2009; O’Donnell et al. 2018, 2021) and indices of large-scale climate drivers in the Pacific, Indian, and the Southern oceans that influence the region (Abram et al. 2008, 2020; Fogt et al. 2009; Jones et al. 2009; Dätwyler et al. 2018; Freund et al. 2019). Together, these records provide a basis for assessing changes in multi-decadal variations in the pre- and post-industrial period and their influence on daily weather extremes for the first time in southern Australia.

Accordingly, the objectives of this paper are to:

• 1. Introduce a new, long-term dataset of daily rainfall, pressure and temperature for southern Australia using newly recovered observations from Perth and Adelaide back to 1830 (Sect. 3);

• 2. Outline the data quality control and homogenisation methods applied to the historical observations for Perth and Adelaide (Sect. 4);

• 3. Demonstrate the ability of historical observations to capture contemporary monthly, seasonal and inter-annual climate variability observed in southern Australia (Sect. 5); and

• 4. Evaluate the covariations between SWA and SEA climate, including the influence of large-scale climate drivers of the El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) and Southern Annular Mode (SAM) on regional climate variability (Sect. 6);

• 5. Provide case studies of newly identified severe storms in the nineteenth century observations and the twentieth century reanalysis to demonstrate the ability of the historical dataset to resolve daily weather extremes in southern Australia (Sect. 7).

Before exploring the historical data in detail, we also provide a brief overview of the instrumental, model and palaeoclimate studies of rainfall variability currently available in both regions (Sect. 2). The varying results of these studies highlight the knowledge gaps that may be filled by longer instrumental records for southern Australia. In this study we focus on rainfall, as high interannual variability makes the long-term analysis of human-induced drying particularly difficult for the region (Cai et al. 2012; Timbal and Drosdowsky 2013; Delworth and Zeng 2014; IPCC 2021b).

2 Human-induced climate change and southern Australian rainfall variability

Southern Australia is one of several sub-tropical and semi-arid regions in the Southern Hemisphere that have experienced long-term drying trends since the late twentieth century (Hope et al. 2010; Cai et al. 2012; Burls et al. 2019; Garreaud et al. 2020; IPCC 2021b). In SWA, there has been a 20% decrease in winter (May–July) rainfall since 1970, relative to the 1900–1969 average. This rainfall decline has increased to around 28% since 2000 (Delworth and Zeng 2014; Bureau of Meteorology and CSIRO 2020). There has also been a significant increase in the average intensity of seasonal droughts in SWA since 1911 due to lower rainfall and increased atmospheric evaporative demand driven by an increase in regional temperatures (Nicholls 2004; Gallant et al. 2013; Bureau of Meteorology and CSIRO 2020).

Several studies attribute this drying in SWA to a poleward shift of baroclinic instability and the mid-latitude storm track, which has in turn been associated with a positive trend in the SAM driven by anthropogenic stratospheric ozone depletion and increased greenhouse gas concentrations (Arblaster and Meehl 2006; Timbal and Drosdowsky 2013; Nguyen et al. 2015; Fogt and Marshall 2020; Frederiksen et al. 2017). The intensification and position of the sub-tropical ridge (STR) – the belt of surface high pressure associated with the descending branch of the Hadley Cell that dominates the southern mid-latitudes at mean sea level – has also been implicated in recently observed rainfall declines in southern Australia since the 1970s (Drosdowsky 2005; Hope et al. 2006; Pook et al. 2006; Pepler et al. 2019, 2021). Lastly, modelling studies have shown that local land clearing has exacerbated the rainfall decline and increased temperatures in SWA (Pitman et al. 2004; Timbal and Arblaster 2006; Nair et al. 2011; Andrich and Imberger 2013).

In SEA, cool season (April–October) rainfall has also declined by around 12% since the late 1990s (Rauniyar and Power 2020). This decline has been linked to decreases in the frequency of mid-latitude cyclones and cold fronts, along with an increase in the proportion of those cyclones and fronts that do not deliver rainfall in parts of the region (Pepler et al. 2021). Instrumental rainfall records show that the average length of droughts in SEA has also increased significantly since 1960, lasting between 10 and 69% longer than droughts during the first half of the twentieth century (Gallant et al. 2013).

While observed reductions in SEA rainfall are influenced by significant multi-decadal variability (e.g. Cai et al. 2014), nearly two-thirds of the rainfall decline during the 1997–2009 Millennium Drought was associated with an intensification of the STR (Timbal and Drosdowsky 2013; CSIRO and Bureau of Meteorology 2015). Although there is currently low confidence that recent droughts in eastern Australia can be clearly attributed to human influences (Cai et al. 2014; Delworth and Zeng 2014; Rauniyar and Power 2020), there is emerging evidence that the recent rainfall decline in parts of SEA would not have been as large without the influence of increasing levels of atmospheric greenhouse gases (Rauniyar and Power 2020).

Given both regions have experienced significant declines in rainfall over the past fifty years, considerable research effort has been focused on identifying the similarities and differences between SWA and SEA climate variability and change (e.g. Bates et al. 2008; Hope et al. 2010; Timbal et al. 2010). The primary influence on southern Australian climate is its location in the prevailing Southern Hemisphere westerly wind belt (Fig. 1), with many systems that pass over SWA moving on to SEA, generally within 2–4 days (Murphy and Timbal 2008; Hope et al. 2010). This results in similarities in the two regions’ seasonal rainfall signature, albeit with a lag reflecting the physical distance of approximately 2,300 kms that separates the western edges of each region (Hope et al. 2010).

Fig. 1

1950–2000 average mean sea level pressure for Australia during the April–October cool season (black solid line) and November–March warm season (blue hatched line). The isobars (in hectopascals) connect regions of equal pressure. The location of Perth and Adelaide are shown as red circles

While the relationship between interannual rainfall in SWA and SEA is largely driven by the link between local rainfall and mean sea level pressure (MSLP), there are years when the influence of large-scale climate drivers result in different climate conditions between the two regions (Hope et al. 2010). For example, Hope et al. (2010) suggest ENSO may strengthen the seasonal link between the two regions, while positive IOD events weaken it (noting, however, that ENSO and the IOD are correlated (Allan et al. 2001; Stuecker et al. 2017)). Hope et al. (2010) conclude that further research on the role of the SAM and the intensity of the STR and southern Australian rainfall is needed before definitive statements can be made about the drivers that strengthen and weaken the relationship between these two regions, especially in the pre-1900 period, which is covered by few studies (e.g. Timbal and Fawcett 2013; Ashcroft et al. 2016b; Ritman and Ashcroft 2020).

These studies illustrate that longer observational records from southern Australia are needed to distinguish any anthropogenic ‘signal’ from the background ‘noise’ of natural variability that is projected to dominate over trends in greenhouse gas emissions in the near future (CSIRO and Bureau of Meteorology 2015). One way to quantify pre-industrial climate variability is to use palaeoclimate reconstructions based on local and remote (but correlated) tree rings, coral and ice core records that can provide extended estimates of multi-decadal climate variability. In both SWA and SEA, palaeoclimate reconstructions have indicated that recent drying trends may not be unusual in the context of the last 500–1000 years (IPCC 2021b). A recent reconstruction of March–July rainfall using tree-ring data suggests that the observed rainfall decline in SWA since 2000 is not unprecedented in terms of its magnitude or duration when compared to the past 668 years (O’Donnell et al. 2021). The authors cite two ‘megadroughts’ of more than 30-year duration in 1755–1785 and 1828–1859 which are greater than the recent rainfall decline in SWA. While no comparison with climate model simulations was conducted to attribute the causes of these pre-industrial droughts, the study concludes that instrumental rainfall records do not capture the full range of natural hydroclimate variability in the region (O’Donnell et al. 2021).

Similarly, palaeoclimate studies suggest that the Millennium Drought in eastern Australia was not unusual in the context of natural variability reconstructed over the past millennium (Palmer et al. 2015; Cook et al. 2016; Kiem et al. 2020). However, shorter palaeoclimate reconstructions indicate a 97% probability that the decadal May–April rainfall anomaly recorded during the 1997–2009 Millennium Drought in southeastern Australia was the worst experienced since 1783 (Gergis et al. 2012). Furthermore, the spatial extent and duration of cool season (April–September) rainfall anomalies in SEA were either very much below average or unprecedented over at least the last 400 years (Freund et al. 2017). Clearly, extended estimates of past climate variability, which cover the pre- and post-industrial periods in as much temporal detail as possible, are needed to reduce uncertainties associated with current limitations in the limited length of instrumental records and the high natural climate variability that characterises the region.

This brief review highlights the need for higher resolution instrumental data to help untangle multi-century climate variations and anthropogenic trends, particularly for rainfall. Sub-daily weather observations are especially important in places like southern Australia, where the long-term interaction of tropical and extra-tropical influences is still poorly understood due to fewer sub-daily weather observations being assimilated into global reanalysis products compared with the continents of the Northern Hemisphere (e.g. Poli et al. 2016; Slivinski et al. 2019).

3 Data

3.1 Historical instrumental data

To extend the availability of daily weather observations for southern Australia, we analyse historical (pre-1900) instrumental data from two locations across the region: Perth in Western Australia and Adelaide in South Australia (Fig. 1). For Perth, daily meteorological observations begin in 1830, and for Adelaide they are available from 1838 (Table 1; see also Supplementary S1 for detail on all historical data sources). While we evaluate all newly recovered variables in Sect. 4 and Sect. 5, this study primarily focuses on the variables of atmospheric pressure and rain days. Given our previous analysis of historical temperature extremes (Gergis et al. 2020; Ashcroft et al. 2021), in this study we target lesser-studied variables needed to assess the pronounced drying trend in the region. Nonetheless, for completeness, we provide an assessment of historical temperature observations to provide a foundation for future analysis.

Table 1 Details of data sources used to compile the datasets for Perth and Adelaide

3.1.1 Perth

We use recently recovered daily pressure, temperature, wind direction, and rain day observations for Perth spanning 1830–1875 (Gergis et al. 2021; hereafter referred to as the Swan River dataset). The Swan River dataset was shown to reliably resolve modern characteristics of Perth’s climate (Gergis et al. 2021). We also use newly transcribed daily pressure, temperature, wind direction, and rain day observations from Perth (hereafter referred to as the Perth Gardens dataset) for the period 1880–1900, and sub-daily pressure observations from the Perth Observatory for the 1897–1908 period. The quality of these new observations is assessed in Sect. 5.1.

3.1.2 Adelaide

We use the homogenised daily temperature dataset for Adelaide developed by Ashcroft et al. (2021) that spans the 1856–2019 period. While inhomogeneities may remain in the early part of this record, Ashcroft et al. (2021) make careful use of 40 years of parallel observations between two thermometer shelters, and data from the Bureau of Meteorology’s homogenised dataset from 1910, to build a continuous dataset for Adelaide. We extend this dataset further into the nineteenth century by using temperature and pressure observations taken by William Wyatt (1838–1847; Gergis et al. 2020) and a newly transcribed set of sub-daily temperature and pressure observations that cover the 1843–1856 period (hereafter referred to as the Survey Office dataset).

In addition to these datasets, we also acquired sub-daily pressure observations that were taken at the West Terrace Observatory in central Adelaide. These observations cover two windows in the mid-19th to early twentieth century (1858–1867; 1876–1907). Unfortunately, the observations that would bridge the gap between 1868 and 1875 have not yet been located (but may be located in historical newspapers). It is believed that the West Terrace Observatory and the Survey Office were located close to each other (see Fig. S3.7).

3.2 Modern instrumental data

In preparation for investigating pre- and post-industrial climate variability in southern Australia, first we assess the quality of the newly digitised historical observations for Perth and Adelaide by comparing them with modern station data from the Australian Bureau of Meteorology (Table 1; Tables S1.1 and S1.2). We then investigate the regional expression of large-scale drivers of Australian climate on Perth and Adelaide climate using climate mode indices for the Pacific, Indian, and Southern oceans.

3.2.1 Perth

We used data from numerous Bureau of Meteorology weather stations in Perth to assess our newly recovered nineteenth century observations. The closest Bureau of Meteorology weather stations to where the historical observations were taken were the Perth Gardens (station number: 009097, 1876–1930), Perth Regional Office (station number 009034; 1880–1992), Perth Airport (station number: 009021; 1944–2020) and Perth Metro sites (station number 009225; 1994–present). The Perth Gardens site is the same station for which we have recovered daily observations (mentioned in Sect. 3.1); however, only monthly averages of daily maximum and minimum temperature are available for this station from the Australian Bureau of Meteorology. The Perth Regional Office site is a composite of three different locations; the Perth Observatory (1897–1963), a nearby location 300 m to the northeast opposite Parliament House (1963–1967) and East Perth, 2 km east of the Observatory (1967–1992). The Perth Regional Office station was discontinued in 1992, which is why we also use the observations made at the Perth Metro station as they provide the most recent observations. Where available, we use temperature and pressure observations available for all of these sites (Table 1).

To produce a continuous rainfall record from 1880 to 2020, we combined the rainfall observations from the Perth Regional Office and Perth Airport. A comparison of overlapping observations of both sites revealed the two to be very similar (not shown, r = 0.98), so this was deemed a suitable approach.

To directly compare the modern rainfall record with historical rain day observations, we convert the modern daily rainfall totals into a simple categorical rain days index. For consistency between the two records, we followed the approach of Ashcroft et al. (2019) and ran a series of comparisons between the seasonality of the historical rain day record and the modern rain day record using different thresholds to define a rain day (e.g. > 0 mm, >  = 0.1 mm etc.). We found a threshold of 0.4 mm or more yielded a seasonality that most closely matched the historical rain day record. Accordingly, we define a rain day as one when 0.4 mm or more rainfall was observed.

3.2.2 Adelaide

For Adelaide, we use the homogenised daily temperature record developed by Ashcroft et al. (2021) which extends from 1856 to present (pre-1910 observations are merged with a homogenous series covering 1910–2019). We also analyse all available daily pressure and rainfall observations for Adelaide sourced from the Australian Bureau of Meteorology (Table 1). Pressure observations are sourced from two stations: West Terrace (station number 23000; 1955–1977) and Kent Town (station number 023090; 1977–2020). Daily rainfall totals are also used from both West Terrace (1839–1977) and Kent Town (1978–2019). No adjustments were made to produce this time series as the difference in rain days between the two sites was found to be minimal (not shown). For consistency with the approach used for Perth, we define rain days as days where 0.4 mm or more of rain was observed. This provides a long-term rain days time series for Adelaide that can be compared with the consolidated rain day record for Perth.

3.2.3 Climate driver indices

To evaluate the relationship between our historical meteorological observations for Perth and Adelaide and large scale drivers of Australian inter-annual climate variability associated with Pacific, Indian and Southern Ocean influences (Risbey et al. 2009), we use monthly climate indices for ENSO and the IOD sourced from the Australian Bureau of Meteorology (http://www.bom.gov.au/climate/influences/graphs/) to examine tropical Indo–Pacific influences on southern Australian climate. While we note that simultaneous and lagged correlations exist between ENSO and Australian climate variability (e.g. McBride and Nicholls 1983; Allan 1988), lagged correlations are strongest for eastern and northern Australia (McBride and Nicholls 1983). Accordingly, we focus this assessment on zero-lag correlations which are highest for our study region of southern Australia. For the SAM, we use the Marshall (2003) index (1957–2020), as a seasonally resolved reconstruction that covers the entire twentieth century is not available (Fogt et al. 2009; Jones et al. 2009). For ENSO we use the Southern Oscillation Index (SOI; 1876–2020), and for the IOD we use the Dipole Mode Index (DMI; 1900–2020).

3.3 Verification with other historical observations, reanalysis, and palaeoclimate data

For further evaluation of the daily historical observations analysed in this study, we also acquired monthly MSLP observations for Perth and Adelaide (Table 1) from the United States of America’s University Corporation for Atmospheric Research (UCAR) research data archive dataset ds570.0 (Quayle 1989; Fogt et al. 2016a). Monthly pressure observations for Perth Airport cover 1876–2018, while data for Adelaide spans the 1857–2018 period. According to Quayle (1989), monthly weather observations were digitised as part of an American philanthropic effort by John Wolbach of the Harvard College Observatory, eventually contributing to the development of the World Monthly Surface Station Climatology dataset (https://rda.ucar.edu/datasets/ds570.0/#!description). It is unclear exactly which sites or observation times were used in this dataset but it is assumed that the monthly pressure values are a mean of all available observations.

To identify the atmospheric conditions associated with historical severe storms, we use the ensemble mean from the latest version of the 20th Century Reanalysis (20CRv3; Slivinski et al. 2019). This release contains recently recovered surface pressure observations from the 19th and early twentieth century from Australia compared with previous versions (Ashcroft et al. 2014a), potentially improving its regional performance over this period (Slivinski et al. 2019, 2021).

We also use recently developed palaeoclimate and instrumentally-based reconstructions to assess the pre-1900 variations in our historical dataset and large-scale climate drivers that influence the region. For ENSO, we use a seasonally resolved coral-based reconstruction of the Niño Warm Pool Index (NWP) indicative of sea surface temperature conditions in the central–western equatorial Pacific for the 1830–2008 period (Freund et al. 2019). For the IOD, we use the July–December coral-based reconstruction available from 1846 to 2019 (Abram et al. 2008, 2020). While SAM exerts a considerable influence on cool season rainfall (Hendon et al. 2007), unfortunately a long-term reconstruction for this season is currently unavailable (Dätwyler et al. 2018). This data limitation highlights the critical need for a long-term cool season SAM reconstruction to better understand long-term changes in southern Australian weather and climate variability.

4 Data quality assessment and homogenisation

Instrumental weather observations must be closely examined to ensure that all fluctuations are due to weather and climate variations, rather than non-climatic influences (Peterson et al. 1998). The process of assessing this, and making adjustments where required to remove the effect of non-climatic influences, is known as homogenisation. This analysis is particularly important for historical records which were often taken with non-standard instruments using non-standard techniques (Brázdil et al. 2010). Additionally, historical observations were hand-written in journals which then need to be transcribed to be digitally analysed. The transcription process can introduce erroneous values where the person transcribing the observations mistakenly transcribes the wrong value. This can occur for many reasons; for example, the handwriting can be difficult to read, or the journals can be in poor physical condition due to their age (Ashcroft et al. 2018). Consequently, a thorough data quality and homogeneity assessment must account for both potential sources of error.

To identify any data quality and homogeneity issues in the newly transcribed historical observations, we use a range of widely used statistical techniques summarised in Table 2. The range of methods account for the different observing frequency (e.g. sub-daily, daily) and the different variables of the newly transcribed datasets. We only assess the quality of the unpublished datasets that are not part of the Australian Bureau of Meteorology’s official climate record, as the Bureau of Meteorology records have already undergone internal quality assessment.

Table 2 Methods used to assess the statistical quality and homogeneity of recovered sub-daily and daily historical observations

While some of the methods we use to assess the quality of the historical observations will detect transcription errors, we also attempted to minimise potential transcription errors through the use of the citizen science platform Zooniverse (https://www.zooniverse.org/) for the data we recovered as part of the ACRE-Australia initiative. Zooniverse has been used to recover historical weather observations from the United Kingdom (Craig and Hawkins, 2020) and New Zealand (Andrew Lorrey, pers comm) but had not been used to recover Australian historical weather observations until now. We used Zooniverse to recover historical daily observations from Perth Gardens (https://climatehistory.com.au/2021/04/20/australias-longest-weather-record-from-climate-change-hot-spot/) and the Adelaide Survey Office (https://climatehistory.com.au/2020/09/30/help-australias-longest-weather-record/). William Wyatt’s pressure observations, the West Terrace Observatory pressure observations and the Perth Observatory observations had already been digitised by experienced data entry volunteers.

From a data quality perspective, the advantage of Zooniverse is that it allows projects to access a global volunteer base, substantially increasing the number of skilled volunteers potentially available to help digitise weather journals. Based on the advice from the researchers involved in managing the Southern Weather Discovery project run in New Zealand (https://www.zooniverse.org/projects/drewdeepsouth/southern-weather-discovery), we opted to have each journal page digitised eight times. They found eight to be the optimal number that minimised transcription errors (Andrew Lorrey, pers comm). We then compared the transcription made by each volunteer per journal page to extract the ‘consensus value’ (e.g. the value most commonly transcribed, for a given day). If five or more volunteers agreed on a value, it was accepted as being a correct transcription. If less than five entries agreed, ambiguous values were verified manually before further analysis.

The historical pressure observations we recovered were taken by mercury barometers, which require further correction to be comparable with modern-day pressure observations (World Meteorological Organization 1971). We implement the corrections recommended by the World Meteorological Organization (1971) so that the newly recovered historical observations can be compared with modern observations (for a detailed description of the corrections applied to the pressure observations, see Section S2). Daily and monthly pressure means were computed by taking the average of all available observations.

Finally, some of the nineteenth century pressure observations we recovered had been reduced to MSLP, while some had not. For consistency, we reduce all historical station pressure observations to MSLP for comparison with modern observations. To do this we followed the method implemented by the Australian Bureau of Meteorology (Treloar, 2009).

5 Perth

5.1 Perth Gardens temperature and pressure observations (1880–1900)

The journals containing the historical Perth Gardens observations (National Archives of Australia series number: PP430/2) are detailed, and provide some insight into the location of where the observations were taken. Based on local expert knowledge (Craig Bowers, pers comm) and the altitude noted in the journals, it is highly likely that these observations were taken from approximately the same location as the Swan River record (Gergis et al. 2021; Glenn Cook, pers comm). This site was historically referred to as the Botanical Gardens (Cooke 1901), now known as the Stirling Gardens in modern Perth (Bowers, pers comm.; Glenn Cook, pers comm). This site is very close to the Swan River and has an elevation of 14 m above mean sea level. As reported on the page containing the January 1885 observations, and elsewhere (Cooke 1901), the site was moved at the beginning of 1885 to a location 17 m above mean sea level. However, historical research has found (Bowers, pers comm; Glenn Cook, pers comm) that this site move was to a location less than 150 m from where the 1880–1884 observations were made.

A compilation of Western Australian meteorological observations from 1876 to 1899 (Cooke 1901) reported that this site move led to slightly cooler temperatures. While we do find that the maximum temperature (tmax) observations are slightly cooler (− 0.6 ℃) after the site move, we also note that the minimum temperature (tmin) observations are slightly warmer (+ 0.3 ℃). It is difficult to attribute these changes solely to the site move. Differences may also be caused by the years sampled in the calculation of these means or by changes to the conditions in which the instruments were kept. We find no significant changes to the climate record before and after the site move. As seen in Section S3, both the temperature and pressure observations from the Perth Gardens were found to be homogenous, supporting the conclusion that the location of the new site was very close to the old site.

A comparison between monthly averages of tmax and tmin computed from the recovered daily observations with those extracted from the Australian Bureau of Meteorology Perth Gardens station (station number: 009097) revealed the two sets of observations are virtually identical (see Section S3.1 and Fig. S3.2). Bureau of Meteorology metadata indicate that the elevation of this station is unknown and the coordinates are known only to 0.01 degree precision, suggesting that there is still potentially some uncertainty about the precise location where these observations were taken.

Annual volumes collating weather observations from across Western Australia prepared by Government Astronomer W. Ernest Cooke from 1894 contain valuable information about the meteorological instruments used at the time (Cooke 1901). From the available metadata we infer that the thermometer used to observe maximum and minimum daily temperature was housed in a Glaisher stand which was enclosed in a thermometer house (e.g. Fig. 1 of Ashcroft et al. 2021). Thermometers were typically housed in Glaisher stands in Australia at this time (Nicholls et al. 1996) to minimise the impact of solar radiation while still allowing air to circulate around the thermometer. At the top of the page containing the observations for August 1885 it was noted that the ‘instruments [were] shifted to new house’. This was also noted by Cooke (1901), suggesting that the Glaisher stand was moved to a new house (although there was no significant change in the location of where the observations were made). Atmospheric pressure was observed using a barometer by E. Sawtell as noted by Cooke (1901). Pressure observations were made twice daily, once in the morning and afternoon at 0900 and 1500, respectively, though observations were occasionally made at 0700 and 1200.

To evaluate the quality of the Perth Gardens observations, we compared them with the parallel observations that were made at the Perth Observatory for the 1897–1900 period. An assessment of the quality of the temperature observations from the Perth Gardens for 1880–1900 revealed very few errors, and the data were found to be homogenous (see Section S3.1). Given these results, we are confident that the recovered daily temperature observations for the Perth Gardens contain few transcription errors. Figure 2 shows the good agreement of the seasonal cycle of tmax and tmin computed from Perth Gardens and modern observations from the Perth Regional Office. The historical tmax observations are likely slightly too warm during summer, perhaps due to the exposure of the thermometer used to take the historical temperature observations. The historical tmin observations are slightly cooler than the modern observations. This offset may be partially explained by the increase in Australian minimum temperatures since the late nineteenth century.

Analysis of the parallel observations from the Perth Regional Office station and the Perth Gardens site was conducted to assess the potential to homogenise the temperature observations. The results show that while the Perth Gardens temperature observations are slightly warmer than those of the Perth Regional Office (Fig. S3.9), they contain similar seasonal variations seen in Fig. 2. We therefore opt to homogenise the Perth Gardens observations using the Perth Regional Office to create a continuous daily tmax/tmin record from 1880 to 1992.

Fig. 2

Daily maximum temperature (tmax; top panel) and daily minimum temperature (tmin; bottom panel) grouped by month for the newly digitised Perth Gardens observations (see Sect. 3 and Table 1) and the nearest Australian Bureau of Meteorology station, Perth Regional Office (station number: 009034). Outliers are defined as 1.5* interquartile range

The newly digitised pressure data from the Perth Gardens were found to be homogenous and contained very few transcription errors (Section S3.1). We applied the gravity and temperature corrections recommended by the World Meteorological Organization (1971) for mercury barometers described in Section S2, and reduced the observations to MSLP. Figure 3 shows that the seasonal cycle of MSLP computed from historical and modern observations for the Perth Gardens closely agree. The wide distribution of observations during the cool season shows that the Perth Gardens record captures the cool season storminess that characterises the region.

Fig. 3

Daily mean sea level pressure (MSLP) grouped by month for the Swan River dataset (Gergis et al. 2021), Perth Gardens dataset, Perth Observatory dataset and Australian Bureau of Meteorology stations Perth Regional Office (station number: 009034) and Perth Airport (station number: 009021). Values are given in hectopascals (hPa). Outliers are defined as 1.5* interquartile range

In summary, we find that the newly digitised historical temperature and pressure observations for 1880–1900 from the Perth Gardens reliably reproduce modern features of Perth’s climate. Combined with the recently recovered Swan River dataset (Gergis et al. 2021), this dataset now provides a nearly continuous daily weather record for Perth from 1830 to present (see Sect. 6).

5.1.1 Perth Observatory pressure observations (1897–1908)

The journals that contain the Perth Observatory observations contain extensive detail about the site and the meteorological instruments used. The meteorological instruments were placed on 1 January 1897, though the Observatory building was not completed until 1898. The Perth Observatory was located on top of Mount Eliza, which is located ~ 2 km west of the city centre of Perth in a field elevated at 60 m above mean sea level. The journals note that no trees or buildings were permitted to be located near the field so that the climate record would remain homogenous. This observing standard, along with the other details captured within these volumes, indicates the well-considered weather monitoring practices of the Government Astronomer and their staff at the time.

A Fortin barometer produced by Negretti and Zambra was used to observe atmospheric pressure, while a thermometer located in a Stevenson screen was used to observe daily air temperature at 09:00 and 15:00. In 1908, these observations were made three times each day with an additional observing time of 21:00. Along with temperature and pressure, humidity, evaporation, rainfall and wind direction and speed were also observed, though we have not digitised all variables. These journals contained observations that had already been reduced to MSLP. The pressure observations were found to be homogenous (Fig. S3.3), with 41 (< 0.01% of the total number of observations) found to have been incorrectly transcribed (Table S3.2).

The pressure observations taken at this site were incorporated into the record of the Perth Regional Office station (station number: 009034) and the similarity in the seasonal variations between the Perth Observatory and the Perth Regional Office shows this (see Fig. 3). As observed in other pressure data sources, the 1897–1908 observations are more variable during the cool season, indicative of the storminess associated with cool season rainfall that characterises the region. In summary, we find that the 1897–1908 historical pressure observations for the Perth Observatory site capture the expected seasonal pattern. The close agreement between the historical and modern observations suggests it would be possible to develop a long-term homogenised pressure record for Perth.

5.2 Adelaide

5.2.1 William Wyatt pressure observations (1841–1847)

The daily weather observations made by William Wyatt at Adelaide are the earliest available for the colony of South Australia, which was officially proclaimed on 28 December 1836 (Gergis et al. 2020). While the temperature observations made by Wyatt have been examined previously (Ashcroft et al. 2014a; Gergis et al. 2020), this is the first time the daily pressure observations have been analysed in detail. Atmospheric pressure was observed three times per day at 09:00, 15:00 and 21:00.¹ Unfortunately, there is no metadata information on the instrument used or exactly where the observations were taken, however Fig. S3.7 shows an approximate location (Mac Benoy, pers comm).

The pressure observations contain one changepoint, also identified in Ashcroft et al. (2014a). Both the magnitude of the step change and the significant increase in standard deviation after the step change suggest the influence of non-climate factors such as a move in the observing site or a change in instrumentation. Consequently, we opt to homogenise this record to minimise non-climatic influences. As there are no metadata available for this record, we were unable to determine the specific reason for this changepoint. As a result, the absolute uncorrected pressure observations between 1841 and 1847 should be interpreted cautiously.

A comparison of these pressure observations with others from Adelaide shows a significant high pressure bias in the Wyatt observations (Fig. S3.4). Wyatt’s observations were only made to the precision of a single decimal point (approximately 3.4 hPa). This precision is less than that specified in the World Meteorological Organization (1971) recommendations that mercury barometer observations be taken to the nearest 0.002 inch (0.07 hPa), and may explain the high pressure bias of the Wyatt pressure observations.

As there is a 3 year and 5 months overlap between the Wyatt record and the Adelaide Survey Office (presented below), we attempted to minimise these biases by correcting the Wyatt record to the Survey Office. To do this we use the method implemented in Gergis et al. (2021) which is a quantile matching correction method originally introduced by Trewin (2018). This method computes the value that corresponds to each fifth percentile (5th, 10th, …, 95th) for both datasets during their period of overlap. These values are used to construct transfer functions which are applied to correct each value of the Wyatt dataset (see Trewin (2018) for further detail on this method). In line with the treatment of other historical pressure observations for Adelaide, we applied the gravity and temperature corrections recommended for mercury barometers (see Section S2) and reduced them to MSLP. Following homogenisation and correction to the Survey Office, the seasonality of the Wyatt record is similar to the other historical sources of pressure presented in Fig. 5 (see Fig. S3.4 of the Supplementary Material to view the Wyatt observations before these corrections were applied).

5.2.2 Adelaide Survey Office temperature and pressure observations (1843–1856)

The observations that constitute the dataset we refer to as the Adelaide Survey Office were made in central Adelaide at the Adelaide Surveyor General’s Office during the period 1843–1856. We were unable to locate comprehensive metadata for the Adelaide Survey Office record. However, we do know that the observations were made at the Adelaide Surveyor General’s Office (Ashcroft 2013) which was located in central Adelaide (45 m above sea level; see Fig. S3.7). No information about the instruments used to take the historical pressure and temperature observations was available. The temperature observations begin in 1843, while the pressure observations do not start until 1844.

The time of observation also varies across the record. For temperature, initially observations were made at 10:00, 12:00, 14:00 and 16:00 on every day except for Sundays and public holidays. From 1852 they were then only taken at 10:30 and 15:30. For pressure, the initial observations were largely taken at 10:00 and 16:00, and then from 1852 were also taken at 10:30 and 15:30. These observations were accessed from the National Archives of Australia, series number: AP810/58.

The homogeneity of temperature for each observation time for the Adelaide Survey Office was examined (Section S3.2). This revealed the occurrence of a changepoint in 1848 which coincides with anomalously cool conditions (see Section S3.2.2 and Figure S3.5). Another identified changepoint in 1850 appears to coincide with a very dry period. Unfortunately, no metadata for the record exists for us to understand which non-climate factors, if any, may have caused these changes.

To facilitate a detailed comparison between the Adelaide Survey Office temperature observations and the record of Ashcroft et al. (2021) and others, we estimated tmax and tmin from the sub-daily Survey Office temperature observations. For this we implement a slightly modified version of the method that is described by Ashcroft et al. (2014a). This method calculates a seasonal linear regression between tmax and tmin and the times at which temperature was observed. The goal is to identify the relationship between the sub-daily temperature observations and tmax and tmin. We calculate this regression using modern observations (we use data from the Kent Town Automatic Weather Station, station number 023090) and apply this relationship to the historical observations to estimate tmax and tmin.

This method of estimating tmax and tmin is most effective when the observing time is close to when tmax and tmin might occur. This means estimates of tmax from historical observations are considerably more accurate than estimates of tmin. For the Survey Office, the afternoon temperature reading was taken at 15:30 or 16:00 (close to when tmax might occur), while the earliest observing time was 10:00 (hours after tmin might occur).

The seasonal cycle of tmax and tmin computed from the Survey Office estimates and the Ashcroft et al. (2021) record presented in Fig. 4 illustrate the suitability of this method. The Survey Office tmax seasonal cycle is remarkably similar to that of Ashcroft et al. (2021). It does not exhibit as much month-to-month variability but the interquartile range is typically within those of Ashcroft et al. (2021). The December Survey Office temperatures are slightly too warm and this may be related to the exposure of the Survey Office thermometer or a reflection of natural variability. For most months the Survey Office tmax estimates are cooler relative to Ashcroft et al. (2021) and this is expected noting that the latter includes the post-1960 period which was much warmer than the nineteenth century. Despite the good agreement between the monthly average tmin values, the Survey Office tmin is unable to capture the variability of the Ashcroft et al. (2021) data (Fig. 4). It is also too warm in December and January. Consequently, the annual average tmin values can be treated with confidence, but the extreme values should be treated with caution. As there is only a single month of overlap between the Adelaide Survey Office record and the homogenised record of Ashcroft et al. (2021), we are unable to combine the records. However, the similarity of the tmax observations of both records suggests that the two could be combined if sufficient overlap became available in the future.

Fig. 4

Daily estimates of maximum temperature (tmax; top panel) and minimum temperature (tmin; bottom panel) grouped by month from observations of Adelaide’s Survey Office alongside tmax and tmin observations from the consolidated daily temperature dataset for Adelaide (Ashcroft et al. 2021). Outliers are defined as 1.5* interquartile range

The Survey Office pressure observations are homogenous (Fig. S3.6), and contained no transcription errors (see Sect. S3.2.2). We applied the gravity and temperature corrections recommended for mercury barometers (see Sect. S2) and also reduced them to MSLP. As seen in Fig. 5, Adelaide’s seasonal cycle of barometric pressure calculated from historical and modern observations shows that the historical record is quite similar to modern observations. There is, however, a consistent low pressure bias observed in all months, suggesting that the bias is likely to be due to instrumentation or siting issues. Seasonally, the Survey Office observations vary similarly to the other pressure observations, showing greater variability during the cool season (MJJASO) and reduced variability in the warm season (NDJFM).

Fig. 5

Daily mean sea level pressure (MSLP) grouped by month from the Wyatt, Survey Office and West Terrace Observatory datasets for Adelaide. Combined monthly means of MSLP from Bureau of Meteorology stations West Terrace (station number: 023000) and Kent Town (station number: 023090). Values are given in hectopascals (hPa). Outliers are defined as 1.5* interquartile range

5.2.3 West Terrace Observatory pressure observations (1858–1867; 1876–1907)

The West Terrace Observatory (now the site of the Bureau of Meteorology’s West Terrace station, station number: 023000) was established in 1856 by South Australian Government Astronomer, and meteorologist, Charles Todd (Gergis et al. 2020; Ashcroft et al. 2021), but only began taking atmospheric pressure observations in June 1858. The pressure and parallel temperature observations for the 1858–1867 period were digitised from the original handwritten meteorological records that were kept by Charles Todd (http://www.met-acre.net/MERIT/). The observations from 1876–1907 were recovered from annual volumes that collated the observations made at the West Terrace Observatory. The majority of these volumes were located in the Hathi Trust Digital Library (https://catalog.hathitrust.org/Record/012261912), with the remaining volumes accessed from the National Library of Australia (https://catalogue.nla.gov.au/Record/2778034). The two records provide varying amounts of information about the instruments used to observe pressure, and the observing conditions. The handwritten meteorological records (1858–1867) contain no information about the instrument used, while the annual volumes (1876–1907) comprehensively detail the instruments used to measure each variable of interest and their specifications.

As these records contain observations made at the same site, which were managed under the direction of the same Government Astronomer, Charles Todd, we assume the conditions and the instruments used to make their observations are similar. It is likely a Glaisher stand was used, as was common practice at the time (Ashcroft et al. 2021). The 1858–1867 observations were found to be homogenous (see Sect. S3) and we corrected 38 transcription errors (Table S3.2). The gravity and temperature corrections had already been applied to these observations, though we reduced them to MSLP. During the 1858–1865 period, pressure was observed twice daily at 09:00 and 18:00. From 1865 to 1867 the observing frequency increased to four times a day: 09:00, 12:00, 15:00 and 18:00.

The 1876–1907 observations were also found to be homogenous, with 51 observations (< 0.01% of the total number of observations) having been transcribed incorrectly (Table S3.2). For most of the record (1876–1901) the journals contain pressure observations taken three times a day at 09:00, 15:00 and 21:00. For 1902–1907, only the 09:00 pressure observation was recorded. It is likely that pressure was observed more frequently over the 1902–1907 period; however, we could not locate the journals which contain the original observations. As with the Perth Observatory pressure observations (1897–1908), the annual volumes from which we recovered the 1876–1907 data for Adelaide only recorded pressure observations that had been reduced to MSLP.

As shown in Fig. 5, the seasonality of the 1876–1907 West Terrace pressure observations and those from the 1858–1867 period closely mirror the seasonality computed from modern observations for Adelaide. As was observed for Perth, the historical observations capture higher variability in the cool season associated with storminess. Resolving this characteristic of the region’s climatology is critical as it allows for the extension of pre-industrial estimate of southern Australia’s weather and climate variability.

6 Assessing southern Australian climate variability

6.1 Identifying optimal seasonal window

Before examining the long-term variability of our newly consolidated historical observations for Perth and Adelaide, we first identify the optimal seasonal window to focus our analysis on. As outlined in Sect. 1, the clearest surface weather and climate expressions of changes to atmospheric circulation in southern Australia have been observed in cool season rainfall. To further investigate the long-term context of this pronounced rainfall decline, this section focuses on the newly consolidated historical pressure and rainfall observations developed in Sect. 5. To identify the optimal seasonal window needed to examine the influence of atmospheric circulation changes on southern Australia’s weather and climate variability we evaluate the relationship between 1) local rainfall and MSLP in Perth and Adelaide and 2) the influence of the large-scale drivers of variability (e.g. ENSO, IOD and SAM) on MSLP.

As seen in Fig. 6 and Figure S3.8, the monthly correlation coefficients between MSLP and rainfall in Perth and Adelaide show distinct seasonal variation. Figure 6 demonstrates the strong negative correlation between MSLP and rainfall (low pressure corresponding with high rainfall) in both Perth and Adelaide particularly over the April–October cool season (Adelaide r = − 0.52, Perth r =  − 0.63, significant at the 99% level over the 1880–2018 period). This confirms the winter-dominated seasonal cycle reported elsewhere in the literature (e.g. Timbal et al. 2006; Bates et al. 2008; Murphy and Timbal 2008; Hope et al. 2010).

Fig. 6

Monthly correlation between MSLP–rainfall (top left), MSLP–Southern Oscillation Index (SOI; top right), MSLP–Dipole Mode Index (DMI; bottom left), and MSLP–Southern Annular Mode (SAM; bottom right), for Adelaide (blue) and Perth (orange), Australia. Correlation with rainfall is calculated over 1880–2018. Correlations with the climate modes of variability are calculated over a common reference period (1957–2018). Asterisks denote the months where the correlation is significant at the 95% level. ^ denotes a statistically significant relationship at the 99% level

Correlations are strongest for Perth, particularly during the winter months. This reproduces the results of Hope et al. (2010) who report the strongest correlation between MSLP and rainfall in May–July (MJJ) from 1890 to 2007 in SWA and southwest eastern Australia (encompassing western Victoria and southeastern South Australia which are more influenced by storms embedded in the westerlies compared to the rest of southeastern Australia). However, in contrast to Hope et al. (2010), we find a stronger relationship between MSLP–rainfall in Perth than in Adelaide. Note that Hope et al. (2010) used gridded monthly rainfall and pressure datasets to calculate these correlations over broad regions that encompass the cities of Adelaide and Perth, which may account for this difference and prevent a like for like comparison between our results based on station data. The study attributes the strongly correlated rainfall relationship between the two sites with the long-recognised relationship with MSLP variations in the mid-latitudes (Allan and Haylock 1993; Hope et al. 2010).

The results shown in Fig. 6 suggest that an analysis of the MSLP relationship with key climate drivers is a logical way to examine the link between atmospheric circulation changes and local weather and climate in SWA and SEA. While we do not have long term numerical rainfall data, we do have extended pressure datasets, and these are highly correlated with regional rainfall. Hope et al. (2010) explore the variability in the rainfall relationship between SWA and SEA, but they did not extensively explore the specific drivers of this variability. Building off their effort, we use the extended instrumental datasets described in Sect. 3 and 5 to assess how the dominant modes of large-scale climate variability, namely the ENSO, IOD and SAM, influence the local climates of SWA and SEA and the relationship between the two regions.

As seen in Fig. 6, there is a strong relationship between MSLP, averaged over winter (July–August) and an extended cool season (April–October) in Perth and Adelaide, and the SOI across all months. This confirms the strong influence of ENSO over the region noted elsewhere in the literature (e.g. Risbey et al. 2009), with marginally stronger results noted for Perth. Statistically significant negative relationships between the SOI and MSLP at Perth and Adelaide are noted for both seasonal windows, peaking at r = − 0.41 during April–October (significant at the 99% level) in Adelaide.

There is a significant relationship (p < 0.05) between cool season MSLP and the DMI, which is strongest from May–October, with the highest correlations observed in early spring consistent with the winter–spring influence of the IOD on Australian climate (Risbey et al. 2009). The correlations are higher in Adelaide in the months of September and October, with the correlation peaking at r = 0.21 during April–October (p < 0.01). The influence of SAM on MSLP is strongest from April–September over the 1957–2018 period at both locations, with slightly higher correlations noted again for Adelaide. While winter correlations are highest for Adelaide, correlations are significant (p < 0.01) for the protracted cool season at both locations. Importantly, these results demonstrate that our historical pressure observations from southern Australia have the potential to be used to make inferences about the likely state of cool-season SAM during the nineteenth century.

Based on these results, we use the April–October seasonal window for all subsequent analysis. This window accounts for the peak relationship between rainfall and MSLP (Fig. 6), while also encompassing a period where all three modes of climate variability significantly influence the atmospheric dynamics of southern Australia.

6.2 Associations between southwest and southeast Australian climate variability

To use the newly consolidated historical observations to infer long-term changes in the weather and climate of southern Australia, a relationship between the climates of Perth and Adelaide must first be established. A close relationship between the two locations has been previously demonstrated by Hope et al. (2010) over the 1890–2007 period. Using the seasonal window defined in Sect.  6.1, we extend this analysis to cover the 1841–2018 period using the newly consolidated historical observations to examine the long-term relationship between the weather and climate of Perth and Adelaide.

Our assessment of the covariations in SWA and SEA pressure presented in Fig. 7 shows that the two regions are highly correlated. A comparison of the normalised annual April–October anomalies between Adelaide and Perth displays a statistically significant relationship at the 99% level. The results are very similar to those reported by Hope et al. (2010), who also found a significant correlation between May–July rainfall variability in the two locations between 1890 and 2007, noting periods of weakened correlation in the 1930s, 1940s and 1970s. Importantly, Fig. 7 also shows that the relationship observed in the post-1900 period is very similar to the relationship observed in the pre-1901 period, providing another indication of the quality of the 19th century pressure observations. These results clearly demonstrate that the pre-1901 pressure observations can be used to infer changes in atmospheric circulation in the pre-industrial period across southern Australia.

Fig. 7

Scatter plot of April–October annual normalised anomalies of MSLP for both Adelaide and Perth for 1841–1900 (orange) and 1901–2018 (blue) periods. The correlation coefficients for both periods are statistically significant at the 99% level (denoted by ^)

6.3 Long term variability of mean sea level pressure and rainfall in southern Australia

Following the extensive data quality checks and homogenisation procedures outlined in Sect. 4, we consolidated rain days and MSLP observations from all available sources for Perth (Fig. 8) and Adelaide (Fig. 9) from the early nineteenth century to present. This provides a basis for inferring long-term changes in southern Australian climate. When daily observations were not available, monthly values were used where available to provide as continuous a record as possible for each site. This was especially the case for pressure observations in the first half of the twentieth century, as discussed in Sect. 3. Although every effort was made to homogenise variables where sufficient overlaps exist (see Figs. 8 and 9), the lack of overlap between constituent datasets prevented homogenisation across the entirety of the records for all variables. As such, where homogenisation was not possible, variables should be interpreted as being useful for inferring relative rather than absolute values when considering historical climate variability and extremes. As justified in Sect. 6.1 , we focus our subsequent analysis on the April–October cool season.

Fig. 8

Consolidated April–October pressure (top) and rain day (bottom) observations for Perth, Australia, over the 1830–2020 period. Solid lines denote daily data, dotted lines denote monthly data as per the sources given in the legend. The normalised MSLP anomalies are calculated by subtracting the mean from each dataset, respectively. The rain day anomalies are calculated using a 1850–1900 base period

Fig. 9

Consolidated April–October pressure (top) and rain day (bottom) observations for Adelaide, Australia, over the 1839–2019 period. Solid lines denote daily data, dotted lines denote monthly data as per the sources given in the legend. The normalised MSLP anomalies are calculated by subtracting the mean from each dataset, respectively. The rain day anomalies are calculated using a 1850–1900 base period

Figure 8 reveals good temporal coverage for Perth for MSLP and rainfall back to 1830, albeit with some small gaps. As identified in Table 1, there are missing rainfall and pressure observations from 1836 to 1837 and 1851–1853 in the Swan River dataset. These gaps are unlikely to be filled as we currently have no leads on possible data sources (Gergis et al. 2021). Daily rainfall observations are missing for the years 1876–1879, preventing a link between the Swan River and Perth Regional Office/Perth Airport rain day record. We did, however, calculate rain days for the Perth Gardens site for these years from Cooke (1901) to provide a continuous monthly rain days record from 1854 to present. Daily pressure observations are most incomplete, with data missing from 1909 to 1941 (Table S1.1). However, monthly observations are available during this period (Fogt et al. 2016a, 2016b), providing continuous data coverage to present. Note that sub-daily pressure observations for the Perth Regional Office (BoM station number 009034) are available from the National Archives of Australia (series K994), but will require an adequately funded digitisation effort to recover the missing 1909–1941 period.

As seen in Fig. 8, there are considerable decadal variations in the April–October anomalies of Perth’s rainfall and pressure observations. This is consistent with independent rainfall reconstructions (O’Donnell et al. 2021; Zheng et al. 2021) for SWA, suggesting that our dataset is useful for inferring long-term climate variability in the region. By using a 1850–1900 reference period to calculate rain day anomalies, we are able to examine the 20th and 21st variability relative to the nineteenth century pre-industrial period.

Figure 8 shows that aside from some wet years during the 1840s and 1870s, the nineteenth century was relatively dry compared with the twentieth century. Our results align with those of O’Donnell et al. (2021) who find that the periods 1828–1859 and 1876–1888 were the driest since 1350 in their tree-ring chronology. Interestingly, Fig. 8 shows that the 1830s were distinctly dry, appearing to be even drier than the well-documented rainfall decline that has been observed in the Perth region since the 1970s. Zheng et al. (2021) also report the 1830s as being very dry in their reconstruction of SWA rainfall derived from an Antarctic snowfall reconstruction. The corresponding pressure observations for this period provide an insight into the atmospheric conditions which produced these significant rainfall deficits. Figure 8 clearly shows that persistent high-pressure conditions were observed throughout the 1830s, matching the independent results of the palaeoclimate records.

The other notable covariation in Perth’s rainfall and pressure observations throughout the nineteenth century is the marked reduction in decadal variability throughout the century. In the earliest part of this record there is a sharp swing from dry conditions in the 1830s to wet conditions in the 1840s, which also coincides with marked variability in the pressure observations, oscillating from high pressure anomalies in the 1830s to low pressure anomalies at the beginning of the 1840s. From the 1850s until the 1870s there is much less interannual variability observed in both the pressure and rainfall observations. Interestingly, O’Donnell et al. (2021) find 1876–1888 to be the driest period since 1350. We do not find this, with Fig. 8 suggesting that this was a relatively wet period in the instrumental observations presented here. A possible reason for this disagreement is discussed below.

The wettest period over the entire Perth record is observed around the 1900s–1920s (Fig. 8). This period also coincides with persistent negative pressure anomalies and dampened interannual pressure variability. A significant La Niña event in 1916–1918 also occurred during this period, influencing much of eastern and southern Australia including the Perth region (Bureau of Meteorology 2021). This agrees with the palaeoclimate findings of O’Donnell et al. (2021) who show 1909–1922 to be the wettest period over the past 700 years. Zheng et al. (2021), however, do not identify the early twentieth century as particularly wet and this may be due to the other influences on Perth’s rainfall (e.g. Pacific Ocean) which may not be resolved by this Antarctic ice core record.

Finally, our results show that while the magnitude of dry anomalies on annual to decadal timescales in the post-1975 period was not the greatest since 1830, the number of below average years from 1975 is unmatched over the 1830-present period. The lack of significantly wet recharge years is what distinguishes the post-1975 cool season rainfall decline from other dry periods. For example, from 1975 there were 27 years of below average rainfall. In contrast during the 1830–1845 period there were 11 years of below average rainfall and 2 years of above average rainfall (Fig. 8). Unfortunately, the journals containing the observations for 1836 and 1837 have not yet been located so are missing from this analysis.

This significant decrease in cool season rainfall since 1975 has been largely attributed to increasing surface pressure (Timbal et al. 2010) which is also clear in our results presented in Fig. 8. Zheng et al. (2021) also find the post-1975 decline to be anomalous in the context of the past 750 years. In contrast, O’Donnell et al. (2021) conclude that the post-2000 rainfall decline is not anomalous relative to variations over the past 700 years. It is worth noting that the O’Donnell et al. (2021) reconstruction is from a region ~ 500 km inland from Perth. The recent decline in rainfall has been most severe along the coast where Perth is located (O’Donnell et al. 2021), and this may explain why the post-1975 rainfall decline is not completely captured by the O’Donnell et al. (2021) tree-ring based rainfall reconstruction derived from inland SWA. This may also explain the other disagreements between the historical dataset and the palaeoclimate record, like the discrepancy in the 1876–1888 period noted above. This also demonstrates the value in having multiple lines of evidence to verify historical climate variability and extremes.

Turning to Adelaide, Fig. 9 displays excellent temporal coverage with rainfall and pressure back to 1839 and 1841, respectively. The rainfall record is particularly noteworthy because it is continuous throughout the entire period at a daily resolution. The pressure observations are continuous at monthly resolution back to 1857; however, some daily readings are missing, most notably between 1901–1954, 1868–1875, and during parts of the 1850s (see Table S1.2).

The consolidated pressure and rainfall record for Adelaide presented in Fig. 9 does not exhibit as much interdecadal variability as Perth (Fig. 8). This may be due to the additional tropical and extratropical influences that impact Perth’s climate. The covariations in pressure and rainfall in Adelaide (r =  − 0.52) are less tightly coupled than in Perth (r = − 0.63). As in Perth, Adelaide's rainfall and pressure remain closely related throughout the record (Fig. 9), reflecting the influence of the large-scale drivers on Adelaide’s climate (Fig. 6). For example, the 1860s were the wettest period during the nineteenth century and are associated with persistent low pressure conditions (Fig. 9). The year 1877 was one of the driest in the entire record, produced by some of the largest positive pressure anomalies observed over the record. The conditions experienced during 1877 occurred during one of the strongest historical El Niño events (Gergis and Fowler 2009; Huang et al. 2020), demonstrating the ability of this record to resolve the surface expression of the large-scale drivers of Australian climate variability.

Further evidence of the strength of the rainfall-MSLP relationship is the clear signature of the 1914–1915 El Niño event that was followed by the 1916–1918 La Niña (Bureau of Meteorology 2021). The 1914–1915 El Niño produced significant rainfall deficits across eastern Australia, with particularly strong impacts in SEA including the Adelaide region (Bureau of Meteorology 2021). In Adelaide, one of the largest positive pressure anomalies observed occurred during the 1914–1915 El Niño, along with significant rainfall deficits (Fig. 9). The 1916–1918 La Niña which followed produced record rainfall across a similar area (Bureau of Meteorology 2021). During this wet period, one of the largest negative pressure anomalies in the entire record is observed for Adelaide, along with significant positive rainfall anomalies (Fig. 9). Independent studies of instrumental and palaeoclimate records confirm that the early twentieth century was characterised by several protracted ENSO episodes (Allan and D'Arrigo 1999; Gergis and Fowler 2009); with prolonged La Niña conditions recorded from 1907–1910 and 1916–1918, interrupted by an extended El Niño episode that spanned 1911–1915 (Gergis and Fowler 2009).

The Federation Drought (1895–1902), one of the worst in Australian European history (Verdon-Kidd and Kiem 2009), is also well-captured by the Adelaide record (Fig. 9). This drought was characterised by continuing El Niño activity (Verdon-Kidd and Kiem 2009) which produced the sustained positive pressure anomalies seen over this period (Fig. 9) and the resulting prolonged rainfall deficits. However, the late twentieth century rainfall decline is clearly much more significant in terms of both magnitude and duration. Consistent with the results noted for Perth in Fig. 8, a substantial post-1980 rainfall decline is also clearly observed in Adelaide (Fig. 9), corresponding with a marked increase in surface pressure over this period. This sustained period of below average rainfall post-1980 and the increasing surface pressure trend are anomalous over the 1839–2020 period.

7 Analysis of southern Australia daily weather extremes—severe storms

Having demonstrated that the newly consolidated historical dataset can resolve large-scale inter-annual and decadal climate variations in southern Australia, next we assess its ability to capture weather variability on daily timescales. Given the dominance of the westerly storm track in this region, we focus our attention on severe storms. We also concentrate our analysis on the periods when there are daily pressure observations available for both Perth and Adelaide during the nineteenth century. This identified the following four windows for analysis: January 1841–February 1851; March 1854–December 1856; July 1858–December 1867; and January 1880–December 1900. To identify severe storms that impacted both Perth and Adelaide we compared monthly sums of the daily rain days records from both locations throughout these four windows. This analysis revealed that particularly significant rainfall events occurred in both Adelaide and Perth during the very wet months of July 1847, July 1862 and August 1900.

To understand the dynamical characteristics of these events, we also compute the MSLP anomalies for each month from the 20CRv3 (Slivinski et al. 2019). We also use these MSLP anomalies to evaluate the ability of the 20CRv3 to resolve the synoptic conditions associated with historical extreme weather events in southern Australia during an especially data-poor time period. As noted previously, high rainfall across southern Australia is influenced by variability in the large-scale circulation features of the ENSO, IOD and the SAM. To understand the role these drivers may have played in producing these wet months, we also examine reconstructions of the ENSO (Freund et al. 2019), IOD (Abram et al. 2008, 2020) and SAM indices (Dätwyler et al. 2018) which represent the modes of climate variability in the Pacific, Indian and Southern oceans.

While the reconstructions of the ENSO and IOD indices have sufficiently high resolution to examine the conditions of the Pacific and Indian Ocean during the identified stormy months, there are no cool season reconstructions of SAM, constraining our ability to determine the role of Southern Ocean variability in influencing weather and climate conditions before 1905. Unfortunately SAM reconstructions between 1830 and 1904 are only available for summer and autumn. As variations in southern Australian rainfall are most strongly influenced by SAM during winter (Hendon et al. 2007; Fig. 6), these warm season reconstructions were unable to be used for this analysis. Additionally, the southern Australia cool season rainfall decline since the 1970s (seen in Figs. 8 and 9) has been linked to the increasing trends in SAM over this period (Cai and Cowan 2006). Therefore, to more completely understand the weather and climate variability of the nineteenth century, a cool season reconstruction of SAM is required, and should be considered a high priority for Southern Hemisphere climate variability and change research.

Following the approach of Gergis et al. (2021), we make extensive use of documentary records to independently verify the occurrence of the three severe storms and their societal impacts. We use historical compilations of severe storms for Western Australia and South Australia (Hunt 1918, 1929), supplementing these records with a detailed historical newspaper analysis using the National Library of Australia’s TROVE database (National Library of Australia 2021) provided in Section S4. We also use the remarks, wind direction, rain days and temperature observations available from the historical weather journals to further examine the local weather conditions that characterised these very wet months.

7.1 July 1847

During the month of July 1847, multiple significant rainfall events resulted in flooding and considerable damage to infrastructure and agricultural crops. In fact, June–July 1847 is the wettest on record in Adelaide’s West Terrace record beginning in 1839, with Table 3 also highlighting extremely anomalous low pressure conditions. The impact of the wet weather was most clearly observed in the second half of the month with widespread reports from both Perth and Adelaide of flooding and major impacts. For example, in Perth on 21 July 1847, The Inquirer noted that ‘On the York road the bridges were washed away…It is apprehended that the crops of the low flat lands will suffer much, many acres of growing wheat being already submerged’. While two days later in Adelaide on 23 July 1847, The South Australian reported ‘The late heavy rains and hail storms have had the effect of raising the river higher than it has been for eight years. At six o’clock last evening, the fine new stone bridge, recently built over the Torrens, gave way’.

Table 3 Palaeoclimate reconstructions and historical observations characterising the three storm case studies presented in Sect. 7

As seen in the upper panel of Fig. 10, the synoptic conditions that helped produce these flood conditions show a persistent low pressure system located over Tasmania. Interestingly, the 20CRv3 results do not show pronounced anomalous atmospheric conditions over Western Australia during this time. Nonetheless, the documentary evidence and independent weather observations made at Perth during this month clearly show the occurrence of many severe storms throughout this month.

Fig. 10

Mean sea level pressure anomalies of selected weather systems that occurred during the stormy months of July 1847 (top), July 1862 (middle) and August 1900 (bottom) from the ensemble mean of 20CRv3. Anomalies are calculated relative to a 1981–2010 reference period to clearly isolate the synoptic signal

Additionally, Table 3 shows that significant negative pressure anomalies were observed in the historical pressure observations taken at Perth during the severe storm observed in Fig. 10. Table 3 also shows temperatures were much cooler than normal during this event, dropping to 4.7℃ below average. 20CRv3’s failure to resolve this event is most likely due to the very limited in-situ pressure observations from the Western Australian region assimilated into 20CRv3 over this time, impacting its ability to resolve this particular event in Western Australia, although the use of the ensemble mean instead of individual model members in extreme event analysis (e.g. Pepler et al. 2017) may also be a factor. This highlights the dynamical importance of recovering historical observations for this region (Gergis et al. 2021) and submitting them to publicly accessible databases such as the International Surface Pressure Databank that the 20CRv3 draws its pressure observations from.

Oceanic conditions around Australia at this time show the Indian Ocean may have provided the moisture for the winter storms which occurred throughout July 1847 (Table 3). The Pacific Ocean during this time exhibited El Niño-like conditions, which for eastern Australia, are typically associated with reduced rainfall. However, significant rainfall can still occur during an El Niño event during negative IOD conditions (Meyers et al. 2007). This highlights the complex mix of competing factors that influence the weather and climate of southern Australia, and the need for a winter SAM reconstruction.

7.2 July 1862

The documentary record for July 1862 reveals significant rainfall at both Perth and Adelaide, with Perth in particular experiencing some of the most severe flooding since the founding of the colony. On 9 July 1862, The Inquirer and Commercial News reported: ‘During the past week we have had a flood higher than has been known since 1830…All the lowlands in the neighbourhood of Perth were, of course, inundated and much destruction of property will follow’. These storms also had widespread impact across the Perth region. Later in the month on 23 July 1862 The Inquirer and Commercial News published a report from a correspondent located north of Perth who writes: ‘The weather has been truly dreadful. I may say we have not had a fine day for the past month, and the rain comes down with greater violence here than at the Swan [Perth]. A similar winter has not been known since the first settlement of this district [1829].’

Days later in Adelaide, on 27 July 1862, the South Australian Register notes: ‘the weather, which had for some days previous been very showery, became exceedingly tempestuous. The rain occasionally fell in torrents, accompanied by high winds, lightning, and thunder… the River Torrens has in consequence risen to a height unknown for several years.’

As seen in the middle panel of Fig. 10, the pressure anomalies for 8–10 July 1862 show the progression of a notable low pressure weather system across southern Australia during this period. This system is likely to have contributed to the stormy conditions observed in Perth, and would have also influenced the antecedent conditions which then influenced the flooding in Adelaide reported later in the month (mentioned above). A pronounced low pressure system can be seen migrating eastward along the southern coast of Australia, drawing cool air from the Southern Ocean, resulting in below average temperatures. The observations made of this at Adelaide and Perth validate these results, indicating temperatures 3–4 °C below average and extremely low atmospheric pressure (Table 3), with Adelaide registering its third-wettest July on record. As seen in Table 3, both IOD and ENSO are in their neutral phase, indicating that another large-scale driver like the SAM may have been responsible for producing these storms. While this storminess is consistent with negative winter SAM conditions, a cool season reconstruction of SAM is required to confirm this.

7.3 August 1900

Several severe storms were observed during August 1900, which brought significant rain that was much needed for the livelihood of those in southern Australia who were experiencing a significant multi-year Federation drought that spanned 1895–1902 (Verdon-Kidd and Kiem 2009). Wet conditions in August followed a severe storm which affected the southwest of Western Australia in late July (https://www.museum.wa.gov.au/maritime-archaeology-db/wrecks/katinka). In The West Australian, comments by the Government Astronomer W.E. Cooke demonstrate how anomalously wet this month was: ‘August was a remarkably wet one in coastal districts…Not only was the amount registered considerably in excess of the average, but the actual duration of rain was much greater than usual…this month there was a succession of “lows” following one upon the heels of another, leaving only very short internals of fine weather in between’.

This series of low pressure systems culminated in one of the wettest Augusts observed in Adelaide. For example, according to the South Australian Register: ‘A year ago the dry districts of the colony were suffering from want of rain, and fervent prayers for a much-needed downpour were offered. How different are the conditions now. This is the wettest August South Australia has known for thirty-six years’. This observation is confirmed using our newly digitised instrumental rainfall observations. August 1900 was more notable for the frequency of rainfall than for the magnitude of individual events. In Perth rain was observed on 27 days, while in Adelaide rain was observed on 23 days. In total Perth Observatory had 202.7 mm for the month of August with a highest daily total of 22.9 mm, while Adelaide recorded 105.2 mm and 18.0 mm respectively.

As seen in the bottom panel of Fig. 10, the synoptic features of one of these storms shows a significant low pressure system that stretched across southern Australia, with a clear west–east progression. This system is typical of the types of rain bearing fronts that are embedded in the westerly storm track, which travel across southern Australia during the cool season. As seen in Table 3, this storm led to significant negative pressure anomalies in both Perth and Adelaide. The historical observations also reveal temperatures that were 5℃ below average. The sea surface temperatures of both the tropical Pacific and Indian oceans were warmer than normal – indicative of combined La Niña and negative IOD conditions – likely supplying the additional moisture that fuelled this particular severe storm.

8 Conclusion

This paper presents a newly consolidated dataset of historical meteorological observations for the two sites of Perth and Adelaide from southern Australia back to 1830. After comprehensively assessing the quality of the historical observations, we find they are able to resolve key features of the contemporary climates of SWA and SEA. This is significant as this new dataset for southern Australia is now able to support estimates of pre-industrial climate and weather variability and extremes over the 1850–1900 period.

Significantly, we provide a ‘proof of concept’ demonstration that this new historical dataset for southern Australia can resolve changes in large-scale circulation that have impacted daily weather extremes during the pre-industrial period. We show that the historical observations are able to resolve transcontinental daily weather extremes using the example of severe storms. We assessed the ability of the 20CRv3 reanalysis to resolve significant synoptic systems in the Southern Hemisphere in the pre-1901 period, while providing an analysis of the Indo–Pacific interactions that may have influenced these conditions. Our study also highlights the benefit of using documentary records to verify the societal impacts of weather extremes like severe storms during the nineteenth century.

Unfortunately, the lack of a cool season SAM reconstruction limited our ability to completely evaluate the complex drivers of these severe storms in the pre-industrial period. While there have been significant advances in understanding the historical behaviour of SAM (Fogt and Marshall 2020), because of its highly seasonal impacts on southern Australian rainfall and its intra-seasonal variations, a higher temporal resolution reconstruction is needed to better understand the long-term influence of SAM on southern Australian climate particularly during the cool season.

Another key constraint is the lack of digitised daily pressure observations for the first half of the twentieth century for virtually all Australian capital cities, including our study sites of Perth and Adelaide. Although previous data rescue efforts by the Bureau of Meteorology were undertaken in the 2000s, pressure observations for capital cities were not digitised. Instrumental pressure observations from ships sailing past Perth and Adelaide are also yet to be assimilated into 20CRv3. Our effort to consolidate this southern Australian dataset is a significant contribution to bridging the gap between nineteenth century daily pressure observations and modern instrumental records. Nonetheless, key gaps in the early–mid twentieth century pressure records still remain and should be prioritised in future data rescue efforts in the region.

The newly consolidated dataset of historical meteorological observations from southern Australia presented here can now be used to infer changes in large-scale circulation and the resulting influence on the westerly storm track, which influences daily climate variability across the region. Further work is needed to assess how this relationship has varied over the pre- and post-industrial periods, with a particular emphasis on noting any non-stationarities between climate variations in the two locations; for example, the weakening of the relationship between rainfall SWA and SWEA during the 1940–1980 period noted by Hope et al. (2010).

Future research will be directed towards a better understanding of the long-term behaviour of the climate modes (inferred from palaeoclimate reconstructions) and an investigation of the variability of the STR calculated using an extended network of historical pressure observations. This effort will help determine the dynamical drivers of pre-industrial weather and climate variability across southern Australia, and the long-term context for anthropogenically-driven climate change now influencing the region.