July 2026 Sets a CONUS Record for the Warmest Month

A lot is being said about the fact that July 2026 eclipsed July 1936 as the warmest month on record for CONUS temperatures. Objections from contrarians are voluminous, but a post on X from Chris Martz seems to capture the gist of the objections I'm hearing. According to him, "That claim is misleading for two reasons:  

1. This is due to minimum temperatures only, which skew the 'average.'  
2. The real, measured thermometer data (area-weighted) don’t support this claim."

Each of these objections are seriously uninformed and misleading, and it won't be hard to demonstrate why. After rebutting Martz's claims, though, I'll point out some ways that this record could be reported in misleading ways, and what I think might do a better job of showing what's happening in the US.

Objection #1: The Record is Due to Minimum Temperatures Only

This is objectively false. The record is for average temperature (Tavg), not minimum temperature (Tmin), and it should be patently obvious that Tavg doesn't skew the average, it is the average. Let's help Martz out and restate his objection to something that is at least mildly intelligent. It's true that since GHG-induced warming prevents thermal energy from escaping to space, we should expect Tmin to increase more rapidly than Tmax, and this is also exactly what we observe. Both Tmax and Tmin are increasing, and Tavg, which is just (Tmax + Tmin)/2, is increasing. This is confirmation that current warming is driven primarily by increases in greenhouse gases like CO2. The Dust Bowl years, by contrast, were driven by a very different mechanism that made daytime highs (Tmax) much warmer while nighttime lows (Tmin) were comparably less affected. It would be fair to say that the 1930s were hot more because of Tmax than Tmin, and the 21st century has been hot more because of Tmin than Tmax. Because of this, it is actually more remarkable that Tavg in July 2026 has eclipsed the previous record from July 1936, since the factors that drove extremely hot summer highs in the 1930s (also made worse by human activity) are no longer at work. For Tmax, only once during the Dust Bowl years did a 12-month running mean of CONUS temperatures eclipse 66°F, but this has become a regular occurrence since the beginning of the 21st century.

The only way to get record highs in the 1930s is to engage in a bit of cherry picking. After limiting ourselves to CONUS, we must then limit ourselves to Tmax (throwing out half of our data), and then we must limit ourselves to the Summer months (throwing out 75% of the remaining data). Here, the Summer (JAS) of 1936 still has the highest Tmax record (though neither August nor September are in the books for 2026 as I'm writing this). Even here, though, of the top 5% JAS Tmax years (n=7), 6 occurred since 1998 and 1 occurred in the 1930s.

Using this cherry pick, we can zoom in on the area most affected by the Dust Bowl (Summer CONUS particularly in the Midwest) and minimize the fact that outside of this season and general area, the 1930s simply were not that hot. Essentially, Chris Martz wants us to zoom in on an outlier while at the same time complaining that scientists are skewing the average. I find that a bit disingenuous.

Tmax was Hot in the Midwest. Not so Much Elsewhere

So yes, if you engage in a sufficient amount of cherry picking, you can see the exceptional heat in Summer Tmax in the midwestern states. But that just makes it even a bit more remarkable that CONUS has so quickly eclipsed that July 1936 record. 

Objection #2: "Measured Thermometer Data" Don't support This Claim

The record is in NOAA's nClimDiv dataset which is "measured thermometer data (area weighted)," so this is again just false. I think we need to help Martz out here again and improve his objection to the data. His real complaint is not that nClimDiv isn't "measured thermometer data." His real complaint is that nClimDiv uses bias correction that removes non-climatic biases that affect CONUS temperature trends.  He would rather us use the raw (biased) data before NOAA uses its homogenization algorithms to remove those biases. Using raw data with known biases simply because these allow you to get your preferred results is never good practice in any scientific discipline, but it's a common objection to sound science among contrarian influencers like Martz. And his statements about homogenization are at best misleading (see links to further evidence and citations to the relevant literature):

1. Homogenization does not technically "cool the past." Homogenization removes biases that artificially warm the past in the US. Globally, bias correction has the overall effect of increasing temperatures prior to 1940 by removing cold biases affecting SSTs. Raw temperatures globally show about 0.12°C more warming than bias corrected datasets.
2. Homogenization does not "inflate recent warming by blending urban biases into rural stations." If anything, the exact opposite is the case. The "raw" USCRN dataset shows marginally more warming than the homogenized nClimDiv dataset for the years they overlap (2005 to present). And multiple lines of research confirm that the most rural stations (both in CONUS and globally) are warming at the same rate as all stations, meaning necessarily that there no remaining bias from urbanizing areas adding spurious warming to homogenized datasets. In fact, even Roy Spencer has shown that to find any bias in CONUS temperatures you have to use the raw station data. This bias is removed in homogenized datasets, and this can be seen in maps of temperature anomalies that show cities warming at the same rate as rural areas.
3. Time of observation (TOBS) bias is very real, and is well-documented in the literature (Zeke Hausfather has an interactive demo here). Changing TOBS does change the frequency of double-counting either hot days or cool nights. This has been quantified and even simulated in USCRN data. Different datasets correct for this by different means and come to the same conclusions. The bias is well-quantified and therefore can be corrected.
4. Martz has no familiarity with the scientific literature on bias correction (including homogenization). His claims come more from the likes of Tony Heller. I've compiled a bibliography of the literature on bias correction and urban heat islands here. Anyone can read this and check up on my claims.

Even if we help Martz out by improving his attempts at objecting to this CONUS record, his attempts fail. But that doesn't mean that all reporting of this record has been stellar. I think we should have some caveats here:

1. This is a record for CONUS (2% of the planet), not the globe. It's significance is not as great as record highs in GMST anomalies.
2. The record high for July 2026 (76.89°F) eclipsed July 1936 (76.77°F) by 0.12°F. I haven't researched the CIs for each, but it's possible it's still a statistical tie.
3. As I've shared elsewhere, record temperatures are less significant for climate than trends and larger scale distributions of temperatures. A stronger case can be made that extreme heat is getting worse in the US by looking at data in different ways. Below I'll show what I think can do a better job. And here I'll stick to CONUS, since that's what the current record is concerning.

As I discuss elsewhere, the frequency of daily Tmax temperatures greater than 95°F are increasing. Clearly there's a bump frequency during the 1930s, but these occur a bit more frequently now (see more discussion on this by Zeke Hausfather at ClimateBrink). But here we're concerned with monthly values, so I downloaded the nClimDiv monthly Tavg from NOAA and then sorted them by rank. Of the 1579 months in CONUS since 1895, I selected the top 5% (n=79) and plotted the distribution of the top 5% of monthly Tavg temperatures. I then binned these in 5-year increments and plotted the frequency of the top 5% of monthly Tavg values.

There are some limitation here, since we're less than 2 years into 2025-2029, and the count could be significantly affected by how I bin the values. But this shows 6 of the top 5% of Tavg occurred between 1935-1939, and that was tied in 2000-2004 and in 2010-2014, then exceeded in 2020-2024 with 10. If I do the same thing with Monthly Tmax, the 5-year bin with the highest count (7) is still 1935-1939 but 2000-2004 and 2020-2024 both tie 1930-1934 for second place (6), and third place (5) belongs to 2005-2009 and 2015-2019. We're less than 2 years into 2025-2029, so we don't know where that will end up.

If I plot the top 10% of Tmax temperatures (n=158), then 2020-2024 ties 1935-1939 for the top year with 10 of the top 10% of Tmax temperatures each.

Likewise, the Dust Bowl was a problem for about a decade or less, while AGW is strongly affecting long term trends. So if we look at larger time frames, the 21st century becomes more significant. So 26 of the top 5% of Tmax values show up in 2000-2024 and 21 show up in 1925-1949. Likewise, 7 of the top 1% of Tmax temperatures (n=16) show up in 2000-2024 while only 5 show up in 1925-1944, even though the top 2 monthly Tmax records are in 1936 and 1934. In fact, of the top 1%, 8 show up in 100 years of the 20th century and 8 show up in the 25+ years in the 21st century.

My results for Tavg are robust to my choice of binning. Here's how the data look with 25-year, 10-year, 2-year, and 1-year bins (note that the dataset begins in 1895, but I started these in 1900, since none of the top 5% show up before 1900). 

No matter how you look at it, There's a spike in the top 5% of monthly Tavg temperatures during the late 1930s that is eclipsed by spikes in the 21st century. These four graphs I think do a better job of showing that the July 2026 monthly record is more than just a fluke. It's a symptom of the fact that CONUS is warming.