1. Why test this again

We already put Inigo Jones’s sunspot and planet cycles through the wringer in our first test of his method. None of 90 cycle tests survived. A fair reply to that is: “you tested your version of his method, not him.” He worked from notebooks we do not have, and he adjusted by judgement. So the honest test is the one that does not need his method at all. Take the forecasts he published, with the date they were published, and see whether the weather followed.

That is now possible because the National Library of Australia has digitised millions of old newspaper pages in Trove, and his forecasts were syndicated widely. If you want the background on how he worked, our page on the basics of the Inigo Jones method covers it.

2. Data

• Forecasts. Every Trove article with “Inigo Jones” in it, 1890s to 1960s, harvested through the Trove API: headline, newspaper, date and the short search snippet. 23,896 of them carry a headline between 1920 and 1954, the period we scored.
• Measured rain. SILO daily rainfall for the grid cells over Dayboro and Brisbane, from 1889 (Jeffrey et al., 2001), plus the Bureau of Meteorology’s own Crohamhurst gauge, station 040062, which is the gauge at his observatory.
• What the papers reported. A second harvest of Queensland newspaper articles whose headline names a flood, cyclone, storm, drought, heat wave, heavy rain or bushfire and whose text names a south east Queensland place: 60,480 distinct articles across 678 searches.
• Official flood list. The Bureau’s chronicle of known floods in the Brisbane and Bremer River basin (Bureau of Meteorology, n.d.), used to check the newspaper flood record.

3. Method

3.1 Pulling the calls out of the headlines

Our API key gives headlines and snippets, not the full article text, so a call is only used when the headline or snippet itself carries it. Headlines like “RAIN IN NOVEMBER. MONSOONAL PROSPECTS.” (The Daily Mail, Brisbane, 20 May 1925, Trove) give a target month and a direction. The same forecast often ran in a dozen papers, so syndicated copies were merged into one call. That left 346 distinct monthly calls: 299 wet and 47 dry. We hand checked 30 of them against scanned page and the extraction was right in 26, about 85%.

Event calls were pulled the same way. A headline like “MORE STORMS DISTURBANCE DUE. FLOODING PROBABLE. February Rain Groups” (Glen Innes Examiner, NSW, 4 February 1928, Trove) is a dated call: it names a month. One like “FLOODS LIKELY. HEAVY RAINS SOON.” (Bundaberg Daily News and Mail, 18 February 1937, Trove) is undated, and we gave it the 30 days after publication. That gave 194 storm or cyclone calls, 58 flood calls and 81 heavy rain calls.

3.2 Scoring the monthly calls

A wet call is right when that month’s rain came in above the long run median for that calendar month, a dry call when it came in below. A forecaster with no skill scores 50%. As a yardstick we also scored “last month repeated”: if last month was wet, call this one wet. It’s the dumbest forecast that still uses the weather, and anyone wiht a rain gauge could have issued it.

3.3 Building the event record, and checking it

Newspapers are not a measurement, so before we used them as “what happened” we tested them. Articles were grouped into dated events by type: 2,161 storm, rain, cyclone, heat, fire and drought events, and separately 660 flood events in the Brisbane and Bremer catchments. The flood list found all 9 Brisbane City floods from 1873 to 1955 in the Bureau chronicle. For the rain types we compared measured SILO rain inside each event window with the same window one year before and one year after.

The rain types pass well. 35% of heavy rain events had a 3 day total at or above the 95th percentile, against 3% of the control windows. Going the other way, 37 of the 40 wettest 3 day spells between 1889 and 1969 fall inside a storm, cyclone, heavy rain or flood event in the papers. The biggest, 513 mm in the 3 days ending 6 February 1931, shows up under all four. Drought did not pass, so we didn’t use it.

3.4 Scoring the event calls

A call is a hit when an event of its type starts inside its window. The trap here is that storms in summer are common, so a forecaster who says “storms in February” will nearly always be right. To deal with that every call is compared with the same calendar window in each other year from 1920 to 1954. The average hit rate in those other years is the “expected with no skill” figure. The p value is one sided: how likely a no skill forecaster is to score this many hits or more.

4. Results

4.1 Monthly rain calls

Right = rain on the called side of the median, chance 50%. Strict test (upper or lower third, chance 33.3%): 32.8% to 35.0% across the three sites.

One subset looks good on its own. Calls made more than 60 days ahead came true 26 of 37 times (70%, p 0.01). We don’t believe it, and here is why. We ran 11 subsets, so one of them reaching p 0.01 by luck is not surprising. After a Bonferroni correction for the number of subsets it fails. And the same 37 calls score 22 at Crohamhurst and 21 at Brisbane, so it does not repeat even at his own gauge. A real skill would show up at all three sites.

4.2 Storm, flood and heavy rain calls

Hit = an event of that type in the SEQ newspaper record starts inside the call’s window. Expected = the same window in each other year, 1920 to 1954.

The named month calls, the ones a farmer could actually plan by, carry no skill for any type. The undated “soon” calls do a little better than chance. We reckon the likely reason is timing: a “rain soon” headline is easiest to write when it is already raining somewhere, and wet spells last. That is our reading, we have not tested it directly. Under stricter controls the storm result moves between p 0.02 and 0.10, and as you will see below the heavy rain edge does not survive measured rain.

4.3 The big floods

The flood calls look best on paper: all 58 together hit 44 against 35.9 expected (p 0.01). But the most of the flood “events” in the papers are a single article about a local creek in fresh. The floods that matter are the ones the whole region wrote about. Restricted to the 33 floods with 10 or more articles, the calls naming a month hit 2 of 16 against 2.3 expected (p 0.70).

On the picture the grey spikes tell it. January 1927 (130 articles), January 1931 (139), April 1928 (66) and January 1951 (49) had no call at all. One of the two dated hits, February 1928, came from a forecast in the Glen Innes Examiner that was very likely written for northern New South Wales. The other, December 1931, was a three month summer window (“WET SUMMER? … FLOODS LATER?”, The Maitland Daily Mercury, 3 November 1931, Trove). Generous windows like that are easy to hit in a Queensland summer.

4.4 Heavy rain against measured rain

The newspaper record is good but it has a weakness: papers write more about weather when there is more to write about, and coverage of Trove changes from year to year. So for heavy rain we dropped the papers and used SILO measured rain. A call hits when the wettest 3 days in its window reach the 95th percentile (61 mm) or the 99th percentile (143 mm) of all 3 day totals.

SILO daily rain, wetter of the Brisbane and Dayboro cells. Expected = same window in the other years 1920 to 1954.

None of these reach significance. We also tried a tighter control, using only years close to the call. With years within 3 of the call the undated calls look strong (31 against 22.9, p under 0.01), with years within 5 it weakens (25.1, p 0.04). We looked into why. The windows in those nearby years happen to be drier than the 1920 to 1954 average (42% against 50% at p95), so the strong result comes from the control, not from his call years. We report it but do not count it.

4.5 Do his wet months bring the events?

If his monthly wet calls meant anything, the windows he called wet should hold more storms, floods and heavy rain than usual. They don’t. Wet call windows held storm events 235 times against 235.3 expected, floods 182 against 184.1, and heavy rain 150 against 166.5.

5. Limits of this study

• We have headlines and snippets only. Many calls are inside the article text, which our key cannot read, so we test a subset of what he wrote. A request to the National Library for full text access is drafted but not yet sent.
• Headlines were written by sub editors, not by him. A headline can sharpen or soften the forecast underneath.
• For many calls the region is unknown. A forecast printed in a Queensland paper could be for the Darling Downs or for New South Wales. Restricting to Queensland papers did not change the monthly result (50.6%).
• Trove coverage varies by year and by paper, which is why we checked heavy rain against SILO as well.
• His calls are heavily wet biased (299 wet, 47 dry). In a subtropical summer that alone scores a fair number of hits, which is exactly why every result here is against the same window in other years, not against zero.
• Heat calls were not tested. There is no daily temperature series for the area before 1991 that we trust for this.
• The flood matching threshold was set by hand. We report results at 10 and 25 articles; at 25 or more the calls hit 3 of 58 against 2.0 expected (p 0.32).
• Official reviews of his forecasts were made in 1939 and 1953. We have not compared our results with theirs yet.

6. Conclusion

We tested what Inigo Jones actually published, not our rebuild of it. Over 35 years his monthly rain calls were right half the time at Dayboro, at Brisbane and at his own Crohamhurst gauge, which is what tossing a coin gives. Simply repeating last month did better. His storm, flood and heavy rain calls that named a month landed on the number expected from the season alone, and he did not call any of the four largest floods of the period. The small edge in undated “soon” calls is the easiest thing in weather forecasting: saying more rain is coming while it is already raining. Against measured rain even that edge is gone.

This fits our earlier test, where none of 90 cycle tests survived. Two independent routes, his method and his published record, give the same answer. We have the greatest respect for the man’s energy and for the rainfall records he kept at Crohamhurst, which are still useful today. But on the evidence, his long range forecasts carried no skill.

We hold our own forecasts to the same rule: if the numbers say it doesn’t work, it doesn’t work. For local weather in general, start at the Dayboro weather guide.

7. Update: a modern follower, tested the same way

People still forecast with his method today, so the obvious next question is whether a modern version does better. We took a current Telegram channel that forecasts in the Inigo Jones tradition and has posted dated “rain dates” for every month since October 2022. We are not naming the channel. The point is the method, not the person, and the person behind it has clearly put years of work in.

We pulled out 142 dated rain windows from October 2022 to October 2026 and fixed the rules before looking at a single result: one extra day each side, and a hit when any day in the window reached 10, 25 or 50 mm. Each window was then compared with windows of the same length dropped at random in the same month of the same year, 20,000 times. That keeps the season and how wet that year really was, which matters in four years that were mostly wet. We scored our own Dayboro station, plus SILO daily rain at 22 towns from Cairns to Gippsland and inland to Charleville.

Full table for all 22 towns and the 50 mm results are in our research notes. Around 140 windows were scored per place; a few fell on data gaps and were skipped, never guessed.

Same answer as the old newspapers. The called windows catch rain about as often as random dates do. Across 75 tests two came in “significant”, Sydney at 25 mm and inland at 10 mm, and with that many tests you expect about three by luck alone. Neither held up at the other thresholds. October 2026 shows how that feels on the ground: both dates called heavy were dry at Dayboro, and our 228.6 mm on the 22nd came one day after a date called minor.

It does explain something, though. The windows cover about four days in ten, so somewhere on a coast from Cairns to Gippsland gets rain inside most of them, 86% of the time at 10 mm. That is why “right on my rain date” posts are so common and so convincing. Random dates would collect jsut as many.

Why we still like Inigo Jones

None of this makes us think less of him. He kept careful rain records at Crohamhurst for six decades, his own gauge holds the Australian daily rainfall record, and he asked a question well worth asking: does our weather move in long cycles? It does, through El Niño, La Niña and the Indian Ocean Dipole, only not on the planetary clock he chose. He also wrote his method down clearly enough that a hundred years later we can still test it, which is more than most forecasters ever leave behind. We reckon the respectful thing is to take a man’s ideas seriously enough to check them. We have done that three ways now, and on this moment the numbers say the rain dates carry no skill. If a future test says different, we will publish that too.

References

1. Bureau of Meteorology. (n.d.). Known floods in the Brisbane & Bremer River basin, including the cities of Brisbane and Ipswich. Retrieved from https://www.bom.gov.au/qld/flood/fld_history/brisbane_history.shtml
2. Jeffrey, S. J., Carter, J. O., Moodie, K. B., & Beswick, A. R. (2001). Using spatial interpolation to construct a comprehensive archive of Australian climate data. Environmental Modelling & Software, 16(4), 309–330.
3. National Library of Australia. (n.d.). Trove: Digitised newspapers. https://trove.nla.gov.au/

How we did it: Trove API harvest of headlines and snippets (no full text), serial and inside the published rate limit. SILO daily gridded rainfall for the Dayboro and Brisbane cells; BOM station 040062 Crohamhurst. Monthly calls scored against calendar month medians; event calls scored against the same calendar window in each other year 1920 to 1954, one sided binomial p values. Research by the Dayboro team, November 2026.