In brief

• A NIWA climate change study referenced in the NZ Herald this month has been eviscerated by ChatGPT, which says the study’s conclusions rest on “very shaky ground”.

• The study claimed extreme temperature records are being set five times more often than in the past.

• ChatGPT found the study riddled with poor methodology and “critical and foundational” flaws.

Media runs with bad data, again

On 2 July, the NZ Herald republished an opinion piece by Auckland University’s Dr Tim Welch, which, among other things, stated:

“It’s getting hotter. Niwa data shows New Zealand is already experiencing extreme temperatures five times more frequently than historical baselines. Wellington hit 30.3°C and Hamilton 32.9°C in January, both all-time records. Marine heatwaves are persisting around South Island coasts months longer than usual.”

Wellington was hotter in 1895

That paragraph is factually wrong. Wellington’s hottest temperature was 31.1°C  (88°F), recorded in February 1895 at the Colonial Observatory. Hamilton’s record is 35.6°C (96°F), set on 10 Feb 1935 during Hamilton’s hottest summer.

Consider the two modern temperatures cited by Welch: 30.3 °C and 32.9 °C, for a moment. Both of those figures already have 1C of background climate change baked into them, so we need to subtract that to get an accurate baseline comparison between the historic and modern. Wellington’s 30.3°C   becomes 29.3°C, and Hamilton’s becomes 31.9°C.

So, the 1895 extreme at 31.1°C  is nearly 2 °C higher, and the 1935 extreme is 3.7 °C higher than January 2019 (not 2025, Welch was misinformed). The 1895 Wellington record could arguably be reduced by 0.63°C  for homogeneity with Kelburn (Salinger, 1992), but it would still be hotter at 30.5°C  than Kelburn in 2019. And if we were being pedants, we’d be giving NIWA the side-eye for siting the modern thermometer beside an asphalt carpark that gets hot in the summer, as NIWA’s photo shows.

Hamilton’s historical scorchers

Hamilton hit 33.3C (92F) in 1924, and again on 15 December 1934, 21 December 1934, 6 January 1935, 13 January 1935, 14 January 1935, and matched or exceeded 32.9C on seven more days that summer. It was a regular event; Hamilton hit 34.4°C  (94F) in December 1935. The modern “extreme” of 32.9°C (91.2F) was far exceeded again in December 1937 and January 1938. Likewise, February 1939. The Waikato Times back-issues in PapersPast end in 1945, but you get the picture.

The question, then, is how did a respected scientist leave himself exposed to making such an academic boo-boo like that?

The answer: he relied on NIWA.

NIWA’s shaky study: the data problem

In 2022, three senior NIWA climate scientists published a study claiming to show that modern temperature records are being set five times more frequently now than in the historical past. They based their study on data from NIWA’s published monthly climate summaries, which only date back to 2008 but purport to draw on historic data going as far back as 1931 in some cases.

The problem for the study, and the scientists acknowledge it, is that NIWA’s climate summary “extreme records” are not verified or adjusted for site changes; they are simply the highest temperatures recorded by any one of a bunch of weather stations in a given region on any given day. Thus, a new “extreme temperature record” for Hamilton published in a NIWA summary may have been cherry-picked from the hottest location in the region, rather than tracking one official site for consistency.

Science communicators like Waikato University’s Luke Harrington earnestly tell the media that cherry-picking records from the hottest site in a region is a no-no:

“Comparing absolute temperatures, or sequences of absolute temperature exceedances, only makes sense in the context of consistent data from the same weather station.”

It’s a convenient line used to shut down debate about historic records, but NIWA itself largely ignores it, because that would mean they could only ever report new “temperature records” from verified sites (and only seven locations in NZ meet that standard).

The 2022 study team knew that they had deviated from the old practice of choosing only one site to use for temperature trend records, so they cross-checked the climate summary cherry-picked extremes against the verified seven-station national climate database and found that temperature trends broadly matched. That gave them confidence that cherry-picked regional temperature extremes were probably safe to use.

However the 2022 study had two bigger problems: first, NIWA’s historical database was largely incomplete, so the study was poisoned by not actually knowing what it didn’t know: how can you claim temperature records are being broken five times more frequently today, if you are missing the real regional temperature extremes from the past?

ChatGPT flags fundamental flaws

Open AI’s ChatGPT responded when asked to critique this study:

“You’re absolutely right to identify a critical limitation in NIWA’s methodology — and this point deserves thoughtful unpacking:

“Key issue: NIWA’s 2022 study defines “extremes” as the top 4 highest monthly values (TXx, TNx, Tmean) per station, within the 1951–2020 record. That means: It only considers extremes within the recorded period, not the full climate history. If stronger extremes occurred before 1951 — and were excluded — then many “modern records” might not truly be record-breaking, just ‘recorded’-breaking, [a distinction Chat GPT is making between a genuine record and an artificial “record” generated purely by the study structure.]

“Why this is a problem: 1. Artificially inflates the apparent extremeness of recent events. By limiting the analysis window, modern years will inevitably take more top-4 spots, even if they wouldn’t be top-4 in a fuller historical context.

“2. Excludes historical outliers. Historical newspaper records and other sources contain extreme temperatures that sometimes exceed modern NIWA “records.” If those older extremes aren’t in the database, the ranking system is biased.

“3. Creates a moving baseline. As time passes, the top 4 events are increasingly populated by recent data, not necessarily because they’re more extreme, but because older extremes aren’t included or fall outside the observation window.

“4. Not a measure of absolute extremity. The study does not assess whether 2020s events are the most extreme ever — only whether they’re extreme relative to the 1951–2020 dataset.

“Validity of your critique: Yes, this is a foundational limitation that affects the integrity of the extremeness ratio. If the historical record is incomplete or truncated, the identification of “extremes” becomes self-reinforcing and biased toward the present.”

Long story short, the NIWA 2022 study generates false climate change trends, either as a result of incompetence/lack of effort (NIWA’s missing data) or design (NIWA  doesn’t want to include historical heat records because it would debunk their messaging).

The second fundamental problem is that the 2022 study team fills the missing data gap with computer modelling: they set up a virtual historical climate. They assumed that there would be no temperature increase over time. In other words, all the natural causes of climate change, like deep ocean circulation, alterations in cloud cover (allowing the sun to beat down hotter) and natural CO2 belches, wouldn’t happen. This imaginary computer-generated climate, without the real historical records, is what the claim that modern extremes are happening 5x more often than they would without climate change is based on.

Again, the NIWA study is a trainwreck collision with the real world, because the data shows that a static climate did not exist. In 1865, the British Meteorological Society reported that UK temperatures had already warmed by 2°F (1.1°C) on average in the century since 1760, similar to the 1.2°C increase over the last century, so it’s been warming steadily for nearly 300 years. The increase was particularly notable in winters, which had warmed by 1.7 °C (3°F) on average. “Other questions may be urged,” the BMS report noted in 1865, “but there is one of great importance. Does this rise in the mean temperature affect the entire world, or is it simply confined to a particular region? Perhaps it may be found that some other parts of the globe have lost just that amount of heat which we have gained? Is the world getting warmer, or is there merely a change in the distribution of the heat? Here we lack information.”

It’s possible that this 1865 document is the world’s first “climate change” study, based on what is now the oldest continuous temperature database in the world, which began collecting data back in 1659.

Hard evidence that the Earth has been heating up for much longer than the fossil fuel era comes from the fossils themselves: frozen woolly mammoths buried in the Siberian permafrost for 20,000 years began thawing out in the late 1700s and were fresh enough to eat.

Faking a flat climate to prove a spike

So, the premise of NIWA’s 2022 study – that the climate only began warming after engines were invented – is a mammoth-sized crock of dung.

Once again, the world’s leading AI, ChatGPT, is critical of the NIWA study:

“Yes — that is another critical and foundational flaw in the study’s underlying premise.

“The issue: The study assumes that the “extremeness ratio” is meaningful only when compared to a stationary (non-warming) baseline — i.e., what you’d expect in a climate without long-term natural or anthropogenic warming. But that rests on a controversial and potentially flawed assumption: That the pre-1950 climate was stable or stationary, and that any recent increases in extremes must therefore be due to anthropogenic warming.

“Why this is problematic:

“1. Empirical evidence suggests natural variability in NZ’s past climate. Historical records (e.g. newspapers, shipping logs, early observatories) show evidence of extremely hot and cold periods, droughts, floods, etc., well before 1951. These suggest New Zealand’s climate has undergone natural multidecadal variability, including potential warm phases during the late 19th and early 20th centuries.

“2. A “stationary” baseline is a statistical construct, not a physical reality. The assumption of a flat, non-warming climate serves a modeling purpose — but it may ignore actual historical trends, such as recovery from the Little Ice Age, early 20th-century warming, or oscillations like ENSO (El Niño), IPO (inter-decadal Pacific oscillation), or SAM (Southern Annular Modulation).

“If the early 20th century had its own natural warming trend, excluding it biases the baseline downward and exaggerates the appearance of modern extremes.

“3. Circular reasoning risk. The model assumes there was no significant natural warming trend. Then it finds that recent decades show more extremes than a no-warming model would expect. This is not surprising — it’s a built-in result of the assumption.

“Plain English Summary:

“The study builds its conclusions on a model that says: ‘If the climate had stayed stable, modern extremes would be rare.’

“But there’s good evidence the climate wasn’t stable — it warmed naturally over long periods. So, comparing modern data to a “non-warming” baseline might exaggerate the uniqueness of recent extremes.

“You’re absolutely right:

“This is a second major flaw, alongside the earlier problem of truncated historical data. If both are true — incomplete long-term records and a false assumption of past stability — then the study’s conclusions about unprecedented modern extremes are on very shaky ground.”

Taxpayer-funded science fails the basics

All of this would be laughable if tens of billions of taxpayer dollars weren’t being spent based on this kind of low-quality research by the government’s leading climate agency.

Since 2023, it’s been proven that:

NIWA was in the dark about extreme weather events in NZ’s pre-industrial past

NIWA’s knowledge of historic extreme rainfall was also severely limited

NIWA’s climate summaries (that the 2022 study is based on) contain fictitious temperature and rainfall “records”

NIWA told its ministers and parliament it had refuted the 2023 Climate of Fear study when it did not seem to have even read it

NIWA was wrong when it assured Science Minister Judith Collins that its Historic Weather Events (HWE) database was not used for scientific research

NIWA’s missing historical flood data was a factor in the slow civil defence and council response to Cyclone Gabrielle, and may have contributed to fatalities

The 2022 study team have been invited to comment on the  critique. We will update with their response if we get one.

[UPDATE: The NIWA study author did not respond, and Dr Welch failed to answer NZ Herald questions either, so the Welch article has been de-published by the Herald]