Are Australia’s automatic weather stations any good? Part 2: Issues affecting performance 1 Dr. Bill Johnston Summary Said to be one of the most important weather stations in the Bureau of Meteorology’s automatic weather station (AWS) network, data for Cape Leeuwin was used to demonstrate a physically based rapid screening technique for evaluating the fitness of maximum temperature (Tmax) data for use in climate studies. Surrounded by ocean, the site is very exposed. Data were affected by ‘wind shaking’, which could reset instruments, cause breaks in thermometer mercury/alcohol columns and damage Stevenson screens. Mist, sleet, fog, rain and sea-spray blasted into screens affects measurements, which are assumed to be dry-bulb. Wind-driven rain is not measured accurately and moving the AWS to an up-draft zone close to the drop-off above the ocean in April 1999 reduced rainfall-catch by 26% and caused Tmax to be biased high relative to pre-1994 manual data. Tmax appeared to increase with rainfall and seven of 27 datapoints were identified as outliers. Non-parametric LOWESS regression visualised the shape of Tmax-rainfall relationships, which are expected to be significant, linear and negative in sign (Pearson’s linear correlation coefficient, Punc <0.05) and aided identification of out-of-range values. Tmax data for Cape Leeuwin don’t reflect the weather and are unlikely to be useful for detecting trends and changes in the climate. 1. Introduction Part 1 of this series outlined the utility of a physically based empirical reference frame for evaluating the fitness of maximum temperature data (Tmax) for detecting trends and changes in the climate. Amberley RAAF (Bureau ID 40004) was the case study and analysis of six additional widely dispersed sites showed the methodology was rigorous, statistically robust and widely applicable. Using metadata to identify non-climate changes is subjective and can be influenced a priori; for instance by making arbitrary adjustments for changepoints that make no difference or by selectively ignoring (or not documenting) those that do. In contrast, the statistical approach is straightforward, replicable and objective. Summarised from daily data, annual datasets used in the study are listed in Appendix 2 (pp. 11-16). In this Part 2 of the series the same thermodynamic principles are used for exploratory data analysis – screening individual datasets for biases and outliers, with the main focus being automatic weather stations (AWS) located in southern and eastern Australia. The overarching question is: are Australia’s automatic weather stations any good? 2. Background While some stations were automated earlier, since AWS were designated primary instruments on 1st November 1996, the Bureau has converted most of their network to automatic operation. However, like thermometers, platinum resistance probes used to monitor temperature (T-probes) may deteriorate in service. According to site summary metadata the original T-probe at Cape Leeuwin (Figure 1) operated for 11 years; Cape Otway, 17 years; Cobar AP and Double Island Point 1 Former NSW Department of Natural Resources research scientist and weather observer. 2 (Qld), 18 years; HMAS Cerberus and Deniliquin AP, 26 years; Armidale AP, 9 years; Flinders Island, 14 years and Cunderdin since June 1995 (25 years). It is concerning also that ACORN-SAT weather stations used to calculate Australia’s warming1 receive only two site visits per year for inspection and maintenance2, which is probably insufficient. Stevenson screens operating unsupervised under adverse conditions accumulate dust, grime, wind-borne salt-spray etc; paint peels off and they become damaged which affects their performance often resulting in upper-range bias on warm days. A salty sheen detected on the 60- litre screen at Cape Leeuwin in September 2015 by the Author for instance, may bias observations. Highlighted in Part 1 is that weather station documentation is mostly inadequate, out-of-date and often faulty. Site-summary metadata does not routinely show when 60-litre Stevenson screens replaced 230-litre ones for instance; for most stations site photographs are not available; positioning of T-probes in front of or behind the supporting frame within screens is important but not documented and maintenance issues such as the use of herbicide to destroy natural ground- cover in lieu of regular mowing (which is the standard) are rarely noted. Figure 1. Looking southwest, the Almos automatic weather station at Cape Leeuwin (Western Australia) is close to the 13-metre drop-off to the ocean in an up-draft zone (left). Horizontal rain can’t be measured accurately by tipping-bucket raingauges buffeted by wind and after the site moved 30 m from the protected position between the sheds (circled right) on 14th April 1999, average rainfall declined 26%. Around 50% of days experience mist, fog, sleet, rain and sea-spray and moisture blasted into the screen by strong to gale-force winds directly off the Southern Ocean affects temperature measurements, which are assumed to be dry-bulb. 3. Site changes and metadata Being the most southwesterly point of the continent, Cape Leeuwin is one of Australia’s most iconic lighthouses. Signage proclaims the weather station (Bureau ID 09518) is “one of the most important weather stations in the AWS network” and commencing in 1907 the dataset is the longest same-locality climate record in Western Australia. According to Simon Torok3, in January 1919 the screen was a “small old observatory pattern”; in November 1926, “still old screen; thermometers 2-feet from the ground”; a new screen was installed in June 1936; by July 1964 its door (which faced south into the weather) had become 1 Australian Climate Observations Reference Network – Surface Air Temperature 2 See: Review of the Bureau of Meteorology’s Automatic Weather Stations (2017). Appendix B, Table 1 (p. 54). Bureau of Meteorology, Melbourne, 77 p. 3 Torok SJ (1996). Appendix A1, in: “The development of a high quality historical temperature data base for Australia”. PhD Thesis, School of Earth Sciences, Faculty of Science, The University of Melbourne. 3 damaged and in October 1978, “Large screen replaces small one. 250 m move towards residence” … to a more sheltered position between the sheds, which should reduce the effect of wind-shaking on instruments. As metadata is unavailable between 1926 and 1936 (10 years), 1936 and 1964 (28 years) then until the site moved 14 years later in October 1978, it is impossible to gauge the state of the site, the screen or instruments without additional research and analysis. ACORN-SAT metadata states the small screen in Figure 1 replaced (a previous) large one on 31 October 1978 and that an AWS installed on 3 February 1993 moved a small distance westward on 14 April 1999. Site-summary metadata (26 July 2020) placed the original site northwest of the lighthouse (at Latitude –34.3742, Longitude 115.1344) but failed to mention the moves in 1978 and 1999. Further, as thermometers were removed on 3 February 1993 there was no overlap between manually observed data and data provided by the rapid-sampling AWS. An undated 1950s site plan located in a file at the National Archives of Australia shows the windgauge (anemometer) about 50 m northwest of the lighthouse but places the original meteorological enclosure in a slightly more sheltered position about 40 m northeast. Looking south from the cottages, 1975 photographs (Appendix 1) show a 230-litre Stevenson screen and anemometer pole located there (not the small screen mentioned by Torok (1996)). Further, a high- level aerial photograph dated 11 November 1943 showed a concrete path leading from the lighthouse to the enclosure, a distance estimated to be 30 m (Figure 2). Figure 2. Joined and reoriented 1943 high-altitude aerial photographs of Cape Leeuwin (left) and a same-scaled Google Earth Pro satellite image (17 November 2017) show the lighthouse (L), the position of the windgauge (w), a met-enclosure serviced by a concrete path (1), the subsequent site between the sheds (2) and the current site atop the steep drop-down to the ocean (3). Although site-summary metadata placed the original site at ‘X’ (Latitude -34.3742, Longitude 115.1344) it was most likely to have been adjacent to the windgauge. Photographs were directly overlaid and saved at 100% and zero% opacity. Apparent shoreline differences are due to the angle of exposure and state of the tide at the time images were acquired and close inspection of individual protruding rocks and shoals showed no material change in sea level during the 74-year period between photographs. 4 The original small observatory pattern screen was most likely near the windgauge, where it was overexposed to gale-force southerly and southwesterly winds direct from the Southern Ocean. Wind-shaking noted by Torok (1996) could re-set thermometers or cause breaks in their alcohol and mercury columns. Photographs and the site plan show the site had moved east of the road before 1943 (to Latitude -34.3747, Longitude 115.1365) and that a standard 230-litre screen was used at least before 1975. That site whose coordinates were not noted in metadata was said to have moved 250 m north (more likely 195 m) in October 1978 to a 60-litre screen installed between the two sheds, with the raingauge about 3 metres to the west surrounded by lawn. ACORN-SAT metadata states the screen and raingauge relocated to their current position near the drop-off on 14 April 1999. Corresponding with the move average rainfall abruptly declined 267 mm/yr or 26% (Figure 3). 1919 Sm1926 Sc 2' 1936above new GL Sc 1964 damaged1978 move; Sc 1993 60-L 1999AWS Sc AWS move 1500 1983 1000 2000 - 1009.6 mm -267 mm 500 Rainfall (mm) 1900 1920 1940 1960 1980 2000 2020 Figure 3.