On Sunday afternoon the station up here read 36.9 degrees. That is the highest September number in the whole record. My first instinct was not to celebrate it, it was to distrust it, because a single station producing a single spectacular number is the oldest trap in amateur weather. So I went to control it properly. Two days later I have a reading that stood up, and a different problem that I did not expect to find.

What actually happened on Sunday

The day started soft. 18.5 at four in the morning, a light southerly, nothing unusual for the first week of spring. By five the wind had swung around to the west and from that moment the whole day was decided.

Westerly air in September is inland air. It comes across the Downs and down over the range, it has already dumped whatever moisture it had, and it arrives dry and hot. You can see it in the dewpoint on the chart below more clearly than in the temperature. The dewpoint sat around 16 or 17 all morning, then through the afternoon it collapsed to 8.7. Relative humidity at the peak was 28 per cent. For a valley that usually sits in the fifties and sixties at that hour, that is properly dry air.

The peak came at 13:45. Not late afternoon, which is when our hot days normally top out. Just after lunch, because the change was already on the way. By 14:30 the wind had backed to the south and the temperature dropped three degrees in half an hour. Then a second, cleaner change came through around half past eight in the evening, wind to the east southeast, dewpoint back up from 8.7 to 16.5 in about twenty minutes. That is the sea air coming back in.

And then this morning, 18.8. An eighteen degree swing in two days. In the weekend the fans were on all afternoon, now I am looking at a jumper.

Why it happened

No mystery here, and the Bureau said it plainly at the time. A high sitting out in the Tasman extended a ridge back over the mainland. Settled, clear, no cloud to knock the sun down, and the flow around the northern side of that high comes at us from the west. The sea breeze that normally saves us in the afternoon never got in, because the westerly was stronger than it was.

The Bureau was forecasting 4 to 6 degrees above the September average across the eastern districts, and there were local fire bans for Ipswich and the Lockyer Valley. So this was a well understood, well forecast synoptic setup. Nothing about the cause is strange, which is exactly why the size of the number bothered me.

Trying to break the reading

There is a second automatic weather station in Dayboro that publishes monthly summaries. Different owner, different hardware, different spot in the valley, no connection to mine at all. If my station had gone silly on Sunday, that station would not have followed it.

It read 36.9.

The same number, to the decimal, on the same day. That is about as clean a confirmation as an amateur network can give you. Two instruments that have never spoken to each other, agreeing on the hottest afternoon of the spring so far. Sunday's reading is real, and I am not going to pretend otherwise just because the number is dramatic.

Their record starts in January 2020, so it is shorter than mine. It also has a bad patch of its own, sixteen days in March 2021 where it reports 48.9 degrees and above, which is not weather, that is a fault. I threw those days out. What is left is 1,700 days where both stations were running at the same time, and that is enough to answer a much more interesting question than "was Sunday real".

The awkward part, and it is not what I expected

Line the two stations up day by day and my station is not the one that has been reading hot. It has been reading cool.

Across 1,004 days from 2022 to January 2025, my daytime tops averaged 0.98 degrees below the other station, and my overnight lows averaged 0.61 degrees above it. Both ends squeezed inward. A compressed day. That is the signature of a sensor that is not getting enough air past it, so it lags the real swing in both directions.

Then in February 2025 it changes, and it changes fast. Across the 548 days since March 2025 my daytime tops average 0.12 degrees off the other station and my overnight lows 0.43 degrees off. That is two stations agreeing. Whatever happened in February 2025 did not break my station, it fixed it.

My own logs agree that something physical happened in that month. Average wind speed at the station jumped from 0.75 metres per second in January 2025 to 1.45 in February, and has stayed in that band since. The solar radiation channel logged a maximum of 63,730 watts per square metre in February 2025, which is not a reading, the real ceiling is around 1,400. Something was moved, replaced or rewired. I do not have a maintenance note for it, which is my own fault and is now on the list.

Actually the direction of this matters more than the fact of it. If my station had been reading hot, Sunday's 36.9 would be suspect. Because it was reading cool, and only stopped in early 2025, the numbers that are suspect are the old ones. Every hot afternoon in my archive before February 2025 is probably understated by around a degree. Which means I cannot honestly say "hottest September day since 2008" off my own record, because the record has a seam in it and the seam runs the wrong way for that claim.

So is it a record or not

Not on the evidence I can defend.

To get a yardstick that does not care what happened to anybody's hardware, I used reanalysis data. That is a model that takes every observation available across the region and reconstructs the atmosphere on the same grid, by the same method, every day back to 1995. For the grid point over Dayboro it puts Sunday at 32.0 degrees.

Against that, out of 316 days falling in the first ten days of September between 1995 and 2026, Sunday ranks second. Only 1 September 2024 was hotter. The average for that window is 23.3, so we were more than eight degrees above where early September normally sits.

Across all 936 September days in that record, Sunday sits at the 97th percentile. Uncommon, indeed. But the September ceiling in the same series is 36.9 on 28 September 2017, then 36.0 on 30 September 2000. Late September regularly does worse than this.

One more thing that came out of the two station comparison, and it corrects something I would have got wrong on my own. Both stations sit about 3.7 degrees above the reanalysis grid cell in 2026. Mine is plus 3.75, theirs is plus 3.74. When two independent instruments in the same valley are both that far above a model grid square, that is not two faults, that is the valley. A reanalysis cell averages a large area of mixed terrain and it clips local extremes. So the gap between 36.9 and 32.0 is mostly geography, not error, and I would have blamed my thermometer for it if the second station had not been there.

The honest summary: Sunday was a genuine 36.9 in Dayboro, it was unusual for the date, it was not unusual for September, and the phrase "all time September record" is not one I can put my name to while my own archive has a seam in it.

Did we forecast it

Three forecasts had a go at Sunday. The scorecard is not flattering, but it is not uniform either.

All three verified against the same station reading. The first two are day ahead forecasts, the Inigo Jones number was generated on 1 August, five weeks out.

The short range forecast was excellent. Six tenths of a degree on a day that ran nine degrees above normal is about as good as it gets, and the Bureau was not far behind. Neither of those is a surprise, a day ahead forecast has the actual synoptic setup to work with.

The long range forecast was a disaster, adn I want to be specific about how. The Inigo Jones monthly product for September, generated on the first of August, called for near average temperatures and stated that no significant severe weather was anticipated for the month. It gave Sunday a top of 25.0 with a 50 per cent chance of showers. It gave every single day of September a 50 per cent chance of showers, which is the tell. When I opened the file, the planetary events list was empty, the lunar phases list was empty, and the analog years list was empty. With no astronomical inputs it had fallen back to a plain seasonal curve. Its own confidence score was 0.35.

It was not a one day miss either. Verified against the station, the same forecast was 10.7 degrees low on Friday the 4th and 7.7 degrees low on Saturday the 5th. If you want the longer version of why I keep testing this method anyway, I wrote it up in what I found when I tested Inigo Jones, and the method itself is laid out in the long range weather guide.

The 12 month agricultural product is worse and I am going to say so here rather than quietly fix it. It currently reports zero heat wave events for the whole period, and an overall risk score of zero, described as "excellent conditions for agriculture and outdoor work". That is not a forecast, that is a broken counter. It is on the bench this week.

One product did get it, sort of. The 5 year seasonal outlook has Spring 2026 at plus 0.5 degrees, above average, with bushfire risk high and heatwave risk moderate. Directionally right. In fairness to itself, that file also carries a note saying its measured skill against plain climatology is currently negative, so I am not going to stand here and claim a win on the strength of one season.

So what does it mean

Less than the headline number suggests, and more than nothing.

It means the first week of September can deliver a day that behaves like late October, and that the dry westerly setup which produces it can arrive before anyone has thought about fire season. Ipswich and the Lockyer had bans on Sunday. In the first week of spring. That is the practical version of what I laid out in the bushfire risk piece last month, arriving earlier than I would have liked.

For anyone with stock or a garden, the thing to take from Sunday is not the temperature, it is the 28 per cent humidity combined with the west wind. That combination pulls moisture out of soil and leaf far faster than a still 37 degree day would. One afternoon like that will undo a week of decent soil moisture, and if you did not water on Sunday evening you will be seeing it in the garden by now. The soil moisture and evaporation numbers are tracked daily in the Farmers Almanac.

• Sunday's peak was 36.9 at 13:45, with 28 per cent humidity and a west wind. A second Dayboro station read the same 36.9 independently, so the number is sound.
• On homogeneous data it was the second hottest early September day since 1995, behind 1 September 2024. September as a whole has done considerably worse, in 2017 and in 2000.
• My station has a seam in it at February 2025. Before that it read about a degree cool on hot afternoons. Older station maxima are understated, so I am not making an all time record claim off my own archive.
• The Dayboro Model was 0.6 degrees out. The long range Inigo Jones forecast was 11.9 degrees out, from an input file with no astronomical data in it.
• Fire weather arrived in the first week of spring. Whatever you were planning to do about fuel around the house in October, bring it forward.

What I am doing about it

Quantify the pre 2025 offset properly and publish a corrected series, so the archive is comparable across the seam instead of quietly stepping in the middle. Reconstruct what was changed in February 2025 from whatever notes and receipts exist. And keep the second station in the loop permanently, because two days of cross checking told me more about my own data than three years of looking at it alone did.

Anything that leans on my station's daytime temperatures is affected, including the forecast accuracy scoring. The direction of the error is at least reassuring, the corrections will make older hot days hotter, not recent ones cooler. But it is still a break in the record and I would rather say that out loud in a post than have somebody find it in the data later. Last month's August review sits on the newer, better behaved side of the seam, for what it is worth.

The daily readings, the full archive and the forecast comparisons all live on dayboro.au, and the detailed weather reports are part of the membership at $3.95 a month or $39 a year. If you garden here, Garden Buddy uses this same weather data to work out what to plant and when. And for everything else happening around town, LocalBuddy.

Method. Station data is five minute records from the Dayboro weather station, queried 7 September 2026. The second station is a separate automatic weather station in Dayboro publishing monthly NOAA style summaries, 1,700 overlapping days after removing 17 days where it reported above 42 degrees. The homogeneous comparison series is ERA5 reanalysis daily maximum at 27.19 south, 152.82 east, 1 January 1995 to 6 September 2026, obtained through the Open Meteo archive. Forecast verification uses the recorded variation tables, which store actual minus forecast for each day. The February 2025 change is inferred from the data and from the station's own wind and radiation channels, it has not been confirmed against a maintenance record.