By now I’m sure that you have seen the headlines claiming that the U.S. just had its “hottest summer on record.” It is front-page news this week from CBS News and CNN to USA Today. Andrew Freedman and CNN meteorologist Brandon Miller report that we have “beat the Dust Bowl era” (Fig. 1).

There is no shortage of news coverage about this. After all, it’s a pretty big story, as such a topic should be. But for simplicity and to not bore you to death (and also, all of these articles read the same), I’m just going to focus on CNN’s reporting. Freedman writes,

“The Dust Bowl summer of 1936 has been long been regarded as the benchmark of hottest summers in the U.S., but has now been equaled or eclipsed twice in the past five years thanks to the influence of human-caused climate change.”

That first part is definitely true. Not only was June–August 1936 incredibly hot, but 15 U.S. states set or tied their “all-time” high temperature records in 1936 alone (Fig. 2), the most in any calendar year. Manitoba, Canada also set theirs that summer. The second part about the 1936 record being equaled in 2021 and broken this year is also true at face value, but there is one reason I’ll detail later on as to why that statistic might not actually be correct.

Freedman continues, stating that the average temperature—which, for the laymen reading, is just the sum of a location’s daily high temperature (Tmax) and low temperature (Tmin) divided by two—was the highest on record since 1895, and that June–August 2026 beat 1936’s record by 0.4°F (~0.2°C). One could even ask whether that difference is even measurable, but I digress. Freedman states,

“Summer featured widespread above-average temperatures across the Lower 48 states, with well-above-normal nighttime low temperatures throughout the country — largely driven by a series of powerful heat domes that affected nearly every state besides Alaska and Hawaii.

The average [temperature] for meteorological summer — from June through August — was 74.4 degrees, which was 3 degrees above average.

Record-breaking heat was especially present across the Southwest, with Arizona, New Mexico, Utah and Colorado each setting records for their warmest summer.”

This is also true (although as a scientist, I absolutely despise the use of the term “heat dome,” but that’s a different can of worms to open and I don’t feel like upsetting the terminally online weather community hall monitors). And therein lies the devil in the details; this “record” was driven primarily by record high overnight Tmin values and repeated transient bursts of record-breaking heat in the Desert Southwest / Four Corners region. NOAA NCEI’s mean temperature percentiles map shown in Fig. 3 below showcases this very well.

However, just because summer 2026 may have had the highest average temperature on record nationwide does not mean that it was the “hottest summer on record,” as widely reported by legacy media. It also does not tell the full story in terms of the conditions that the vast majority of people actually felt.

Reporting that it was the “hottest summer on record” implies that average Tmax was highest on record, not the average Tmin or overall Tavg. By Tmax, 2026 ranks fourth behind 1936, 1934, and 2012, respectively. Some of my readers may argue “tomayto, tomahto,” and say that terminology is trivial. But it absolutely matters. Case in point, NOAA refrains from describing this past summer as having been the “hottest on record” in the CONUS. They explicitly state in their press release that meteorological summer 2026 was “warmest for the CONUS in the 132-year record” (Fig. 4). The adjective “hottest” appears ZERO times in the press release.

Above I stated that this statistic “does not tell the full story.” And that is abundantly clear from a spatial standpoint.

I used the University of Iowa’s Iowa Environmental Mesonet (IEM) page to generate the maps below showing temperature rankings for all 344 districts used in NOAA’s U.S. Climate Divisions database. Notice that there were very few districts in the U.S. that observed their warmest summer on record (marked by No. 1 labels) this year (Fig. 5a). In fact, only SEVEN of the 344 were record-breaking. That’s a little over 2%. And all seven No. 1s (and most top 5s) were confined to the southwest. Now compare that to the summer of 1936 (Fig. 5b) where a whopping 51 (~15% of) districts recorded their warmest summer on record. Make of that what you will.

There’s one more important thing to consider here and that is that the data NOAA NCEI use above are “homogenized.” That is, the aggregate of station data used for these temperature analyses has been adjusted from the raw data that is handwritten on the original paper observation forms submitted to the federal government by volunteer station observers. These adjustments are not a government conspiracy; they are part of a process known as “quality control.” The problem, however, is that the magnitude of these adjustments may be questionable.

A long-running weather station is rarely, if ever, perfectly consistent for 100+ years. During its operation history, it may be relocated, observe a change in observation time, and instrumentation most certainly changed from mercury- / liquid-in-glass thermometers to electronic Maximum Minimum Temperature Systems (MMTSs). Also, the landscape around a rural station may become developed over time, leading to an artificial warming trend. There is also the issue of missing daily observations; these values are marked with an “M” or “-9999” in digitized data files.

NOAA (and other government agencies) are well aware of these potential issues (e.g., Menne et al., 2009), and so they apply correction factors to the monthly (and annual) data. Daily values are never adjusted unless there are transcription errors brought to NOAA NCEI’s attention (and as I have found in my own research, there are a lot of them that have gone undetected). The largest of these correction factors in terms of its overall magnitude is the time of observation bias (TOB) adjustment.

For some background, prior to the mid-to-late 19th century, weather station observers generally took temperature measurements three times a day: first at 7:00 a.m., 2:00 p.m., and finally at 9:00 p.m. This, of course, was a very tedious task. And while this technique provided a fairly decent estimate of the temperature variation throughout the day, more often than not, we know that Tmax occurs around or a little after 3:00 p.m. locally while Tmin occurs shortly after sunrise (the time of which varies throughout the course of the year). Therefore, a location’s true Tmax and Tmin on any given day were likely under and overestimated, respectively, for much of the 19th century.

By the 1870s, the U.S. Weather Bureau (now National Weather Service) began to use min-max thermometers at stations in major cities (e.g., Washington, D.C., New York City, Richmond), which log both the true Tmin and Tmax over a 24-hour observation period. Traditional versions used a liquid column that pushes small markers as the temperature rises or falls: one marker remains at the warmest point and the other at the coldest point until the thermometer is manually reset, even after liquid contracts or expands upon reaching their daily extrema. An observer then read those values and reset the markers, allowing the thermometer to begin recording the next day’s Tmin and Tmax, respectively. By the 1880s, rural observation sites began employing the min-max thermometers, and by October 1890 when the Cooperative Observer Network (COOP) was launched, essentially all observing stations were using them. The COOP data are integrated into NOAA’s Global and U.S. Historical Climatology Networks (GHCN / USHCN).

However, different COOP stations had different observation times. That is, some station observers reset their min-max thermometers in the morning while others reset theirs in the afternoon or evening. Afternoon schedules can produce warm bias in the monthly means, particularly for Tmax, while morning schedules can produce the opposite effect, particularly for Tmin. Starting in the early 1980s, there was a shift from afternoon to morning stations, which some scientists have argued induced an artificial cooling trend into the unadjusted (raw) CONUS Tavg time series.

It has been argued by the likes of Drs. Zeke Hausfather and Texas A&M’s Andrew Dessler that afternoon resets resulted in a “double counting” of hot days in the past, which means that the raw Tmax observations cannot be trusted at face value and thus that charts showing robust declines in hot summer afternoons are misleading. The justification for this rests on the idea that afternoon resets carried over the high temperature from “day one” into the following day.

To illustrate this better, suppose that a COOP station observer resets his max thermometer once per day at 3:00 p.m. local time. Let’s also suppose that on the afternoon of July 1st, it reaches 104°F (40°C) at 3:00 p.m. The observer goes out at 3:00 p.m., reads the marker, jots the value down on a log sheet, and finally resets the thermometer for the next 24-hour observation period. Now, let’s say a cold front plowed through that evening and the next day, the high temperature was only 86°F (30°C). Well, when the observer goes out to read the max thermometer, it will show the high as being 104°F (40°C) again. However, this does not mean every day of a heat wave is counted twice like Hausfather and Dessler claim. It only potentially adds ONE artificial hot day at the trailing edge of a heatwave event when sharply falling temperatures follow a hot afternoon observation, which was discussed in Christy (2026). According to Dr. John Christy’s paper, hot afternoons in the raw temperature data are only overestimated by ~4% in the “early part of the record compared with the present.”

When the TOB adjustments are applied (each station data file contains metadata with their observation times), they are applied to the monthly means, ultimately inducing a significant cooling tendency that is arguably not entirely real. As a result, NOAA’s TOB adjustment adds ~0.2–0.4°C (0.4–0.7°F) of warming to the unadjusted U.S. temperature record. The upper bound of the TOB adjustment is larger than the margin by which the U.S. supposedly broke the previous 1936 and 2021 June–August Tavg record. TOB adjustments, while arguably necessary, should only be applied to the handful (at most) of specific calendar days on which an erroneous reading was logged and only to COOP stations that have switched from afternoon to morning observation times. Instead, the adjustment is applied to monthly averages, which effectively erases both legitimately hot afternoons as well as erroneous observations. It may be an overcorrection.

There is another significant adjustment made to the raw station data that attempts to correct for spurious trends. These are false statistical patterns where data over time appear(s) to show a statistically significant upward or downward trend, but the movement is actually caused by some third external factor as opposed to being an indication of change in the physical quantity that is being measured. In the case of temperature data, this is typically caused by station movement over time; instrumentation changes; and perhaps most importantly, long-term urbanization around the observation site. NOAA (and other groups like Berkeley Earth) use what’s called the “pairwise homogenization algorithm” (PHA) to achieve this.

The PHA compares a station’s monthly series with numerous neighboring stations. Weather-related fluctuations such as a regional heat wave should appear at many neighboring stations. Simply put, a sudden change found primarily at one station is more likely non-climatic in origin and thus must be corrected for. PHA is designed to detect breakpoints (discontinuities) associated with station siting and instrumentation changes. The previous urban correction factor used in USHCN Version 1 (which contains the COOP data) was eliminated, and NOAA argues that PHA indirectly corrects for urban sprawl in Version 2.5. But PHA, being a breakpoint detector, is poorly suited to that gradual urban heat island (UHI) effect signal. In fact, PHA may actually blend urban warming into rural data, something that Dr. Roy Spencer has written about and documented extensively on his blog.

Urban blending happens with PHA because if a rural station sits next to growing towns or airports that have weather stations to “pair” it to, those neighbors already carry a slowly positive UHI slope. Because that slope is shared across the neighboring stations, it does not manifest as a sharp breakpoint in the time series. So, the algorithm does not remove it. Instead, PHA adjustments can pull the rural time series toward the urban ones. Fig. 6 below from Dr. Roy Spencer shows this well. Stations that had no urban-related warming trend in the raw data now have a warming trend after PHA adjustments are applied.

Another problem with the PHA is that there are breakpoints the algorithm inserts into station time series that don’t have a documented cause (e.g., location or instrumentation change) in their metadata. The algorithm does not require a station move, instrument change, or time of observation note. It only needs a statistically unusual jump in a difference series. O’Neill et al. (2022) found that only 19% of GHCNm Version 4 breakpoints (18% in Version 3) landed within a year of a documented event such as a move or instrument change. While that does not prove that the adjustments were unnecessary, it most certainly calls them into question.

While summer 2026 was indeed the “warmest on record” in the CONUS since 1895 using NOAA’s homogenized data product, it was not the “hottest” (i.e., by the average Tmax). While the goal of homogenized data NOAA NCEI publishes is to ensure quality control, there are legitimate concerns that the adjustments applied to raw data may overcorrect for certain biases (e.g., time of observation) and/or not completely remove others (e.g., urban warming), which in aggregate act to artificially cool the early and mid-20th century (potentially meaning that 2026’s reported “record” isn’t actually a record) while artificially warming more recent years with urban blending.

You can discuss amongst yourselves in the comments.

Categories: Climate

Tags: Climate