There are papers that confirm expectations, and then there are papers that quietly force you to re-check the assumptions everyone has been repeating for the past decade. This new study by John Christy falls squarely into the latter category.
The paper, “Declines in hot and cold daily temperature extremes in the conterminous US, 1899–2025”, takes on a topic that has become almost ritualized in public discourse: temperature extremes. The narrative most people encounter is straightforward and rarely questioned, that heat extremes are increasing rapidly, cold extremes are fading away, and both trends are tightly coupled to greenhouse gas emissions.
What Christy has done here is step back and ask a more basic question: what do the actual station observations, extended as far back as possible and examined without heavy-handed adjustments, say about extremes over the full historical period?
The answer is not quite what the headlines would lead you to expect.
First, a word about the dataset, because this is where much of the value of the paper lies. Christy revisits the U.S. Historical Climate Network (USHCN), originally assembled in the 1980s as a high-quality subset of stations with relatively stable observing conditions. As my two surfacestations studies have shown, that network has not been maintained in a particularly rigorous way in recent decades, with nearly half the stations closed since 2000. Rather than accept that degradation, the study reconstructs and extends the records, “threading” nearby stations with high correlations and minimal bias to fill gaps.
The result is a dataset of 1,211 stations with at least 92% completeness, spanning from 1899 through 2025, and consisting of more than 40 million daily observations. Importantly, these are actual observations, not homogenized monthly products. As the paper states, “all values utilized are actual, observed temperatures… without any spatial or temporal interpolation or other types of homogenization methods applied to the station observations.”
That alone should get the attention of anyone who has followed the long-running debates over adjustments and data handling. Now to the central findings.
The headline result is almost disarmingly simple: both hot and cold extremes have declined over the full period of record.
The abstract puts it plainly: “metrics for extreme summer heat… show modest negative trends since 1899… Extreme cold temperature metrics also indicate a decline… In sum, instances of both hot and cold extreme metrics have declined since 1899.”
That is not a typo, and it is not a selective statistic. The study looks at multiple definitions of extremes, including single-record events, daily record frequencies, anomaly magnitudes, and multi-day heat and cold waves. Across these metrics, a consistent pattern emerges.
Start with the simplest question: when did the most extreme events occur?
According to the results, the standout years for heat are not recent. The year 1936 dominates the record, accounting for about 22% of stations experiencing their all-time hottest day. The next most prominent years are clustered in the early to mid 20th century, with 1934, 1930, and 1954 all ahead of most modern entries.
Cold extremes tell a similar story in reverse, with 1899 standing out due to the well-known Arctic outbreak that still holds many records.
Already, this should give pause to the idea that current extremes are unprecedented. The observational record suggests that the most severe events, at least in the U.S., are not a modern phenomenon.
Moving to a more robust metric, the study examines the frequency of daily record highs and lows. Here again, the 1930s dominate for heat, with 1936 producing an average of 6.7 daily high records per station, far above the expected random rate. What happens after that peak is more interesting than the peak itself. There is a sharp decline in record highs from the 1950s through the 1970s, followed by a partial recovery in recent decades. But even that recovery does not reach the levels seen in the earlier period.
The paper notes:
“the most recent 15-year period… was slightly above the expected value… yet well below the highest of 35.1 in 1925–1939.”
In other words, recent increases exist, but they are modest relative to the historical range.
Cold records show a different pattern, with a pronounced decline since the 1990s. That part aligns more closely with expectations of a warming climate, though the paper is careful to note that non-climatic influences such as urbanization may play a role.
One of the more telling results comes from examining the magnitude of extremes. When comparing the hottest and coldest days each year, the difference between them has decreased by about 6°F over the past century.
That is a reduction in variability, not an amplification. The system, at least in terms of these metrics, appears to be moderating rather than becoming more extreme.
Heat waves and cold waves provide another lens. The study defines these as runs of at least six consecutive days above the 90th percentile or below the 10th percentile. Over time, the total number of such extreme days has declined by roughly 30% since the early 20th century.
The peak period for heat waves was 1930 to 1944. The minimum occurred in the 1960s and 1970s. Recent decades show some increase, particularly in the western U.S., but again not to historical maxima.
The cold wave story is simpler: a steady decline across most regions. Taken together, these results point to a long-term decrease in extreme temperature events in the U.S., with regional variations and some recent upticks in certain areas.
Now, before anyone jumps to conclusions, the paper spends considerable time discussing caveats, and this is where things get more nuanced.
Non-climatic influences are a recurring theme. Station moves, instrumentation changes, and especially urbanization/siting issues can introduce biases. The Fresno case study is particularly illustrative. There, minimum temperatures increased by over 5°F relative to nearby rural stations, largely due to local development.
That matters because many extreme metrics rely on minimum temperatures, especially for cold events. If nighttime temperatures are artificially elevated by urbanization, cold extremes will appear to decline even if broader atmospheric conditions have not changed as much.
Christy acknowledges this explicitly, noting that urbanization effects are “well-documented” and disproportionately affect minimum temperatures.
At the same time, maximum temperatures, which drive heat extremes, are less sensitive to these local effects. That makes the lack of a strong upward trend in heat extremes more difficult to attribute to measurement artifacts.
Another point worth noting is the role of natural variability. The paper repeatedly emphasizes the magnitude of early 20th century extremes, particularly the 1930s heatwaves. These events set a high bar for comparison and complicate attempts to attribute recent changes to specific forcings.
As the author puts it, “the magnitude of local and regional short-term natural variability… is greater than the magnitude of warming due to GHGs” in these metrics.
That is a statement about signal-to-noise. Even if greenhouse gases are contributing to warming, their influence on extremes in the U.S. may be small relative to the inherent variability of the system.
The paper also takes the unusual step of directly comparing its findings to claims in the Fifth National Climate Assessment. This is where things get a bit uncomfortable for the standard narrative. The NCA5 asserts that climate change is increasing the frequency and severity of heatwaves. Christy tracks this claim through the report and finds that, when specified more precisely, it applies to certain regions and time periods, particularly since 1960.
When evaluated against the dataset, the CONUS-wide trend in heatwave days since 1960 is positive but small, on the order of 3%, and not statistically robust. Regionally, increases are concentrated in the Southwest, while other areas show little change or even declines.
This does not mean the NCA5 is entirely wrong, but it does suggest that broad, generalized statements can obscure a more complex reality.
One particularly interesting example involves the use of threshold metrics, such as days above 95°F. These tend to emphasize regions where such temperatures are common, skewing the overall picture. When percentile-based metrics are used instead, the spatial patterns become more coherent and less dominated by a few hot regions.
It is a reminder that how you define an “extreme” can significantly influence the conclusions you draw.
So where does this leave us? The key takeaway from this paper is not that nothing is changing, but that the story of temperature extremes in the U.S. is more complicated than often presented.
There are declines in cold extremes, some regional increases in heat extremes, and an overall reduction in the frequency and magnitude of the most severe events when viewed over the full 127-year record.
There are also substantial uncertainties related to measurement practices, station environments, and data completeness, all of which can influence the results. And perhaps most importantly, there is the persistent influence of natural variability, which can produce swings in extremes that rival or exceed those associated with long-term trends.
For those interested in the intersection of climate science and policy, this matters. Decisions are often justified by claims about increasing extremes, yet those claims are rarely examined in the context of the full observational record.
Christy’s work does not settle the debate, but it does provide a detailed, data-driven perspective that is difficult to dismiss. It raises questions about attribution, about the role of local versus global influences, and about the reliability of commonly cited metrics.
In short, it is the kind of paper that invites closer inspection rather than easy conclusions.
