Most Extremely Hotter-than-usual Month

In a recent post I showed that in June the northwest had its most extremely hot month on record.

To compute how extremely hot each month was, I took the difference between the month’s temperature and the average for that month throughout the year, returning temperature anomaly. Then I divided the anomaly by the standard deviation for that month to compensate for the fact that winter months show more variation naturally — a large departure from average isn’t so unusual in winter as it is in summer, so such a departure isn’t really so “extreme.” For the northwest, June was the most extreme out of all 1,446 months on record (using data from NOAA).

It’s easy to imagine my being accused of “cherry-picking” for choosing the northwest, now. After all, we just had an extreme heat wave there — what a time and place to pick! It made me wonder, what about looking at all of the U.S. states? I could almost hear the protests that “Oh yeah? Well state (insert state here) had its most extremely hot month way back in (insert year here).”

So I did exactly that. For each of the 48 states in the conterminous U.S. I computed which month was the most extremely hot, after computing anomaly (to allow for the seasonal cycle) and scaling by the month’s standard deviation (to allow for the greater natural variation in winter months).

Here’s a plot of how many states had their most extremely hot month in each year since 1895:


Clearly there have been some years that brought extreme heat. In fact, back in 1961 no fewer than six states had their most extremely hotter-than-normal month. But 2012 topped even that, with seven states reaching the pinnacle of hotter-than-usual.

More noteworthy is the fact that even though most years no state reached its peak, every year of the 2010s at least one state did. So I also tallied how many states had their most extremely hot month by decade. The result is quite telling:


In 2nd place is the decade of the 1960s, with seven states experiencing their most extremely hotter-than-usual month. But in 1st place, by a huge margin, is the 2010s. This decade we’ve already seen eighteen states reach the extreme. Truly, here in the U.S. the 2010s have beaten all those other decades by a wide margin.

And the decade of the 2010s isn’t even over yet. It’s barely more than half over.


A reader wondered what it would look like to compute the most extremely colder-than-usual month for each state. Here it is, by year:


And here, by decade:


19 responses to “Most Extremely Hotter-than-usual Month

  1. Timothy (likes zebras)

    What would also be interesting would be to compute the opposite metric. The most extremely colder-than-usual month.

    Naively I would expect the same sort of chart, but flipped with the majority of the most extremely colder-than-usual months in the earlier years and decades. However, if global warming is also associated with an increase in the variation around the increasing mean, then it is possible that recent decades will have a few extreme cold months.

    [Response: Fascinating. See the update at the end of the post.]

    • Timothy (likes zebras)

      Thank you for doing that. Intriguing results.

      • Thanks also.

        With: “And the decade of the 2010s isn’t even over yet. It’s barely more than half over,” maybe the contrarian modelers are hoping that the system will revert past the mean and pull a bunch of cold records out of the air.

  2. Clearly some sort of commie weather modification experimentation going on in 1960-61.

    • Quite possibly… :-)

      The winter of 1960 was memorable enough, apparently, to inspire this 50th anniversary review from a NOAA analyst:

      February and March 2010 marked the 50th anniversary of one of the most severe periods of winter weather on record in the western Carolinas and extreme northeast Georgia. Beginning in February 1960 a series of winter storms and a prolonged period of below normal temperatures disrupted commerce, schools, and life in general across the region.

      The severe winter weather was not confined to the southern Appalachians and nearby areas. The entire eastern United States was affected. According to narratives at the time (Ludlum 1960a,b), the month of March 1960 had more records for snow and cold in the eastern states than any previous March. March 1960 is still the coldest March on record at Asheville, Charlotte, and Greenville-Spartanburg.

      Didn’t find anything like that for the hot records in ’61, but apparently the affected area was the west coast, somewhat similarly to this year, as you can infer from this story–essentially a list of this year’s June heat records:

  3. Dick Veldkamp

    Thank you for yet another interesting post!

    For good measure (and in line with your usual practice I think), the bars in the ‘colder-than-usual’ graphs should probably be in blue. ;-)

  4. skeptictmac57

    Well, now that the data are becoming so convincing (undeniable?) , have you noticed how many contrarians are now claiming ” Nobody (ever?) is saying that global warming isn’t happening (as though this was some strawman argument that hardly existed), we just object to the solutions that are being crammed down our throat (as though all that much is being done). And if you look, you still see the same old denier arguments EVERYWHERE, while they now push this intellectually dishonest line.

  5. There must be a nonparametric test you can do on this.

  6. I count from the graph 34 successes (new hi’s) for all previous decades 1900 and after for a rate of 34/600 or .06 (my count could be off by a couple as I was eyeballing the graph). So far in the 2010s we have 18. Assuming no more record hi’s get set in any new state for the rest of the decade that leads to an observed probability of .36 as an absolute min for this decade. A quick binomial test in R to see if 18 successes in 50 tries is to be expected from the previous distribution [binom.test(18,50,p=.06)] leads to rejection at parts per billions (p-value = 2.833e-10). This test is conservative, of course, as any additional successes simply make the 2010 decade even more extreme.

    As for the lo’s by decade, historically we see (again if I am eyeballing right) a rate of .08 successes (new lo’s) per decade. The same test for 0 in 50 tries [binom.test(0,50,p=.08)] is also significant but only at the p=.03 level. Of course in this case the result is NOT conservative but rather a maximum difference as additional successes will tend to make the distributions more equal.

    [Response: A caution: the events aren’t independent. When one state hits its most extreme, nearby states are likely to as well.]

    • Re. independence comment…True. But that would also be true in the 1900-2009 base period as well.

  7. Is there a (correct) way to compensate for the fact that under a relatively stable climate, times in the past are more likely to have “hottest” (or “coldest”) months? Alternately, could you accompany this with a graph of, for instance, the states with months which exceed the period average by more than two SDs (showing how often in the past the same extremes were approached but later exceeded)?

    [Response: This analysis doesn’t track all times when a record was set, it tallies the times when the highest value in the entire time span occurred. So it doesn’t need the compensation you mention.]

  8. On these graphs, each state gets exactly one point: during which year did that state reach its highest temperature for the entire 1895-2015 period? It’d be interesting to see the graph of the number of states which hit highest-temperature-to-date records in *each year*. To what extent are we seeing an increase in record-breaking-to-date heat events over the past couple of decades, compared with the 20th C?

  9. What is clear is that “cold” gave up in the 90’s.

  10. Chris O'Neill

    Can anyone explain what is going on with GISTemp since 1998?

    According to , the trend since 1998 is quite statistically significant, but this was not the case before the update including the June anomaly was released a few days ago. There appears to be a lot more changed than just the addition of the June anomaly.

    The skepticalscience calculator also says the GISTemp trend since 2000 is statistically significant.

    [Response: Starting this month they switched from using the ERSST version 3b data set for sea surface temperature, to using the updated ERSST version 4. I’ll post about the new temperature data soon.]