Victor Conrad’s missing data

Ever since reading Robert Dewar and James Wallis’ ‘Geographical Patterning of Interannual Rainfall Variability in the Tropics and Near Tropics’ about two decades ago, it has occured to me that when the authors say:

“Conrad mapped variability in precipitation for the entire world, using records from 384 stations. For the area between 30˚N and 30˚S, he had 149 stations.”

Yet, Dewar and Wallis fail to note that for many of the high-variability regions they discuss, Conrad had no stations whatsoever. This is true particularly for:

1. Queensland
2. eastern Melanesia (the “Fiji–New Caledonia” region of Ropelewski and Halpert)
• of the regions discussed by Ropelewski and Halpert as having coherent El Niño Southern Oscillation precipitation responses, “Fiji–New Caledonia” is the only one where all stations in Dewar and Wallis’ database are “highly variable”
3. northwestern coastal Australia
4. coastal Angola
5. eastern Kenya and Somalia (“Greater Somalia”)
• if you read ‘Geographical Patterning of Interannual Rainfall Variability in the Tropics and Near Tropics’ you will see that Dewar and Wallis do discuss the Horn of Africa as noted by Conrad
• in reality, Conrad had no stations between Zanzibar and Aden, nor in present-day Ethiopia or Eritrea
6. Baja California
• of all the high-variability regions in the 1999 dataset, Baja is undoubtedly the most “excusable”, at least in the sense that no data for the region existed as of 1928 when Conrad’s data was collected

Asking the question on Google, I have discovered that the reason most of Australia and eastern Melanesia were unsampled in Victor Conrad‘s 1941 study. Apart from a few stations published in Quarterly Journal of the Royal Meteorological Society — which were those listed by Conrad — Australia’s large assembly of meteorological records was some of the world’s most extensive but also most fragmented:

1. before the formation of the national Bureau of Meteorology (BoM) in 1908, rainfall data was collected independently by individual Australian colonies (states like New South Wales, Queensland, etc.)
2. each colony or state published its own monthly meteorological summaries
3. these were mostly distributed locally or held in national and state archives and libraries rather than in global meteorological circulars
4. records were held locally or in state archives like the Australian Bureau of Meteorology’s state volumes
1. the great majority of stations in Australia (and Melanesia) thus had their records published only in state-specific histories and never in global summaries
2. the vast majority of high-quality long-term records remained “buried” in colonial-era journals, ship logs, or government documents
3. global sources like Smithsonian’s World Weather Records rarely carried data from these regions

Most Australian and Melanesian meteorological records when Conrad was writing were thus localised and inaccessible to international researchers, being not yet integrated into widely-circulated international compendiums. The most important of these were the UK’s Meteorological Office and the Smithsonian Institution’s World Weather Records.

Even so, one might argue that articles like Steven Sargent Visher’s ‘Variability Versus Uniformity in the Tropics’ published nineteen full years before Conrad’s work should have provided enough data for Onslow and perhaps other parts of tropical Australia. It is possible that the raw data used by Visher were not kept in his archives, but the basic information should have been available to Conrad yet he clearly did not use it.

What I will do below is:

1. select a representative list of stations in areas of Australia not covered by Conrad
1. only data from 1928 and before — when his first data were compiled — are included
2. data will be done in a calendar year format as they were done by Conrad
3. data will be tabulated as Conrad did for his actual stations, although I have omitted latitude and longitude
4. I have deviated from Conrad in ordering stations by region in a clockwise order, rather than purely by longitude as Conrad did
2. compile:
1. mean annual rainfall by calendar year to 1928
2. average departure from the mean, and then comparing it in two ways with Conrad’s expected value:
1. by simple difference, as Conrad did
2. by ratio to expected average deviation minus 100 percent

Method 2) was added because studying Conrad’s article does suggest that areas of high rainfall but abnormal variability are not easily identified by mere difference from expected value. Of the few high-variability regions identified by Dewar and Wallis for which Conrad did have data, two — the northern South China Sea region and lowland eastern Indonesia — were not identified by Conrad as regions of abnormal variability, not discussed as such. This despite the fact that Nha Trang and Ambon Island were shown as having a variability that was clearly exceptional, whilst nearby stations had sufficiently high variability that a pattern could quite likely have been recognised. Conrad’s failure to recognise the northern South China Sea and lowland eastern Indonesia as areas of unusual rainfall variability related to his use of arithmetic rather than geometric (as used by Dewar and Wallis and by Pierre Camberlin’s 2010 ‘More variable tropical climates have a slower demographic growth’) departure.

Explanatory Shading:

1. Departures above 20 percent or more than twice Conrad’s expected value have been shaded in dark red
2. Departures above 10 percent but below 20 percent or more than 1.5 times but less than than twice Conrad’s expected value have been shaded in red
3. Departures above 5 percent but below 10 percent or more than 1.25 times but less than than 1.5 Conrad’s expected value have been shaded in pink
4. stations who departure is less than 5 percent and/or less than 1.25 times Conrad’s expected value are unshaded

Representative Stations from Areas of Australia Unrepresented in Conrad’s Study (Courtesy Australian Bureau of Meteorology):

Region	Station	Elevation		Mean		Average deviation		Percent	% deviation from expected	departure as % of expected
		feet	metres	inches	millimetres	inches	millimetres
Tropical Queensland	Boulia	532	162	10.6	269	5.53	141	52%	+28%	+117.43%
	Burketown	20	6	27.9	709	10.04	255	36%	+18%	+99.70%
	Coen	653	199	46.7	1,187	11.47	291	25%	+9%	+58.37%
	Cooktown	20	6	69.9	1,775	17.23	438	25%	+10%	+64.36%
	Innisfail	33	10	142.8	3,627	26.13	664	18%	+4%	+30.71%
	Townsville	13	4	47.3	1,203	13.59	345	29%	+13%	+79.46%
	Rockhampton	36	11	39.2	995	10.70	272	27%	+11%	+70.65%
	Barcaldine	876	267	19.8	502	6.86	174	35%	+16%	+87.67%
Murray–Darling Basin	Roma	981	299	23.6	599	6.87	174	29%	+11%	+61.80%
	Tamworth	1,326	404	26.8	682	5.07	129	19%	+1%	+4.95%
	Dubbo	853	260	22.2	563	5.22	133	24%	+6%	+30.82%
	Albury	515	157	27.9	709	4.72	120	17%	-1%	-6.02%
	Wentworth	121	37	11.8	301	3.13	79	26%	+4%	+19.96%
Western Interior	Daly Waters	696	212	26.4	670	6.71	170	25%	+7%	+41.29%
	Tennant Creek	1,237	377	14.4	366	5.10	129	35%	+15%	+77.03%
	Moonaree	787	240	7.4	187	2.44	62	33%	+6%	+22.76%
	Wiluna	1,709	521	9.6	244	3.18	81	33%	+8%	+32.48%
	Mount Magnet	1,398	426	9.4	239	3.53	90	38%	+13%	+50.14%
	Halls Creek	1,181	360	21.0	532	6.04	154	29%	+11%	+60.25%
West Coastal	Geraldton	10	3	18.5	469	3.87	98	21%	+2%	+13.36%
	Carnarvon	16	5	9.5	241	3.61	92	38%	+13%	+52.57%
	Onslow	13	4	9.0	228	5.46	139	61%	+36%	+143.67%
	Roebourne	39	12	11.7	297	4.78	121	41%	+19%	+86.16%
	Derby	26	8	25.8	655	8.01	204	31%	+13%	+72.53%
Southern Coastal	Bega	164	50	33.5	852	9.25	235	28%	+11%	+62.23%
	Melbourne	102	31	25.5	648	3.78	96	15%	-3%	-17.66%
	Hobart	171	52	23.9	608	4.26	108	18%	0%	-1.18%
	Eucla	305	93	10.0	253	2.13	54	21%	-3%	-10.88%
	Albany	10	3	37.3	947	4.98	127	13%	-3%	-18.98%

Results:

The results clearly show that if Conrad could have obtained existing data for tropical Queensland — extending into the extreme northern Murray–Darling Basin represented by Roma — and the north of Western Australia, he would have recognised them as regions of abnormal rainfall variability analogous to Northeastern Brazil and the northwestern Indian subcontinent. This is true even if we use Conrad’s arithmetic departure method. By geometric departure, as I presumed, tropical Queensland appears somewhat more variable relative to Conrad’s calculated expected value, and the southern arid zone less so. This difference, however, is less than I anticipated.

It is quite possible that had Conrad some of the data tabulated above, he would have seen the two as one region of unusual rainfall variability: the figures for the central Northern Territory [Tennant Creek and Daly Waters] make this quite plausible. However, mechanistically, Queensland is more closely related to lowland eastern Indonesia and eastern Melanesia than to northwestern Australia. The high variability of the latter region is purely due to dependence for rainfall upon random tropical cyclonic disturbances that frequently produce a year’s rain in two or three days. Contrariwise, lowland eastern Indonesia, eastern Melanesia, and almost all of Queensland owe their high variability to lying in the core of the “ENSO horseshoe” where convection is most sensitive to El Niño and La Niña events. The geometric departures in Boulia and Burketown, it might be noted, are only slightly higher than Conrad tabulated for Ambon Island [+87 percent].

The south coast of New South Wales, represented by Bega, is similar to northwest Australia in owing its high variability to distinctly random Tasman Sea cyclones producing exceptionally heavy rainfall.

The remaining unsampled regions of Australia — tabulated here for both fairness and completeness — do not show much surprise in light of later studies like those of Dewar, Eddie van Etten and Camberlin.