US Warmhole

Not all places on earth are experiencing global warming at the same rate. Let’s consider the U.S., the “lower 48 states.” Taking data from NOAA for the 344 climate divisions in this region, and computing the linear trend rate for each, we can see differences between different parts of the USA, with red dots for warming and blue for cooling, larger dots faster and smaller dots more slowly:

The most notable feature is that in the southeastern US there’s a region which hasn’t warmed nearly as much as the rest of the country. There are even places which have shown slight cooling, most notably in Alabama and Mississippi.

The part of the U.S. that has warmed far less (if at all) has been dubbed the “US warming hole.” I’ll call it the “warmhole” (just because I like the name). It covers a larger area than just Alabama and Mississippi. A number of papers over the last several years have investigated why this might be.

Meehl et al. (2012, J. Climate, 25, 6394) begin their abstract thus:

A linear trend calculated for observed annual mean surface air temperatures over the United States for the second-half of the twentieth century shows a slight cooling over the southeastern part of the country, the so-called warming hole, while temperatures over the rest of the country rose significantly.

The impression is that there is a small warming rate, especially during the second half of the 20th century. I’m not so sure that’s actually the case. Let’s look at the southeast, as defined by NOAA; here’s the data for mean temperature (note: I’ve converted from °F to °C, and these are temperature anomalies):

The blue line is a linear regression to all the data; its slope is a mere 0.004 °C/yr. But the red line shows a 2-piece fit, separating the data pre- and post-1958. The time was chosen by changepoint analysis. This fit is strongly statistically significant, and that’s taking both autocorrelation and the multiple testing problem into account. In my opinion, this means we shouldn’t think of “not warming since 1950” as much as we should consider the cause of the sudden drop in 1958, one which wan’t a one-year event but a lasting shift.

Some of the investigation centers around the seasonal pattern of the warmhole, and the trends in daily high temperature vs. daily low temperature. We can look at the data for daily mean, daily high, and daily low, both for the year-round average and for the four seasons (winter is D-J-F, spring M-A-M, summer J-J-A, autumn S-O-N). We’ve already seen the daily mean temperature year-round; here is the same for different seasons:

The most likely shift time is 1958 during winter and spring, 1961 in summer, and 1950 during autumn. However, the spring and autumn shift times don’t achieve statistical significance (at 95% confidence), not even close, while winter and summer do.

Most notable is the size of the apparent shift. For summer it’s a sudden drop of about -0.63°C, and the estimated size is about the same for spring (-0.56°C) and autumn (-0.58°C), but they don’t reach statistical significance due to the higher variance of those seasons’ temperature. But for winter, the drop in 1958 is a whopping -2.01°C, about three times as large as for any other season.

Looking at daily high temperature, the story is much the same. Here’s the data for annual means:

and here’s the breakdown by season:

The annual shift time estimate is 1958, as it is for winter and spring but for summer it’s again 1961, with autumn happening in 1947. Again, the spring and autumn shifts are not statistically significant (not even close); for summer it just makes 95% confidence, and for winter the significance is beyond doubt.

And, as for mean temperature, high temperature shows a much larger drop in winter (-2.02°C), with summer more modest (-0.76°C).

The overnight low also shows much the same behavior. Here’s the annual mean:

and here are the seasonal changes:

Yet again, significant in winter and summer, not in spring or fall. The winter shift is 1958, the summer shift is 1961, and winter drops a lot (-2.01°C) while summer drops far less (-0.52°C).

What I haven’t done is look in greater detail at the geographic distribution of changes; I simply took the “southeast” as defined by NOAA. But I do think that other authors haven’t considered the possibility that the “warmhole” is a consequence, not of a trend change, but of a sudden shift. In fact, not only has the research I’ve seen emphasize trend changes rather than value changes, some have focused on how 10-year trends change over time, but my study suggests that 10 years is such a brief time span that the natural fluctuations of 10-year trends are too large to provide maximum information.

Wherefore the warmhole? There seem to be many ideas.

Meehl et al. conclude that the warmhole is likely related to SST (sea surface temperature) anomalies in the Pacific and Atlantic oceans, particularly the tropical Pacific. Its seasonal differences may be due to different phenomena related to the formation of more persistent “ridges” in the west and “troughs” in the east, which increases advection of cold air from the northeast to the southeast region during winter, but in summer primarily causes greater precipitation and cloudiness.

Yu et al. (2014, Scientific Reports, 4, 6929) focus on high temperature during summer, attributing the warmhole to the aerosol indirect effect and precipitable water vapor. In summer, short-wave cloud forcing dominates, partly offset by the greenhouse effect of increased water vapor, while long-wave cloud forcing “can warm both winter Tmax and Tmin.”

Mascioli et al. (2017, Environmental Res. Lett., 12, 034008) consider a wider geographic area and its diverse history. They emphasize the role of the North Atlantic Oscillation (NAO) and Pacific Decadal Oscillation (PDO), but also emphasize the complexity of the situation, that the warmole “reflects both anthropogenic aerosol forcing and internal climate variability, but the dominant drivers vary by season, region, and time period.”

Most recently, Partridge et al. (2018, GRL, doi: 10.1002/2017GL076463) define the geographic limits of the warmhole more precisely, and (largely in agreement with what I see) point to a regime shift in 1958 where annual maximum (Tmax) and minimum (Tmin) temperatures decreased by 0.46°C and 0.83°C respectively.” They also identify slightly different regions for winter/spring effect as opposed to summer/autumn, and suggest that the two seasons’ behaviors have different causes. They point to the Meridional Circulation Index (MCI), the AMO, and the PDO as strongly related to winter temperatures but don’t explain the summer temperature pattern. What I find most plausible is their “evidence that the jet-stream exhibited a shift in the late 1950’s coincident with the start of the warming hole, resulting in a greater tendency of northerly winds to bring cool air to the southern U.S.

I don’t see that any consensus has yet emerged, particularly about the causes, so I expect there’s still more to be learned about this. But I definitely credit Partridge et al. for identifying the regime shift in 1958, and for delineating the location of the warmhole more precisely, including the differences in its location during different seasons. In any case, this is science in action. Different researchers propose different ideas, and until we have tested competing theories strongly enough we’ll have to wait for real confidence in any explanation of the details of how and why U.S. temperature history shows such interesting differences in different regions.

This blog is made possible by readers like you; join others by donating at My Wee Dragon.


25 responses to “US Warmhole

  1. I grew up in the humid Southeast (before air conditioning). There were nights when it was difficult to sleep because it simply did not cool off enough after the Sun went down. Living in Florida, which was also very humid was not the same due to reliable ocean breezes. I’ve spent time in the high desert where it gets blisteringly hot during the day but one goes diving for a coat shortly after sunset.

    Is the SE not showing the effects of global warming because high humidity levels were already trapping heat at night?

    Has anyone looked at 1900 to current year temperatures broken into daytime/nighttime buckets? Is there a larger impact on nighttime temperatures than daytime, overall?

    Are there more deniers in the SE because they are not personally experiencing global temperature increases, unlike other parts of the country?

    • Are there more deniers in the SE because they are not personally experiencing global temperature increases, unlike other parts of the country?

      Probably more historical-social. The South had slaves, skewed conservative as a result, and said conservatism means less acceptance of new, ideology-challenging science.

      I doubt the observed temperature increase thus far is really enough to change many people’s minds. We already tend to filter out information that disagrees with us, so it probably falls below the level needed to overcome that.

      • People who live where lakes no longer freeze with thick ice (or don’t freeze at all) and where spring is coming earlier certainly report a change in climate. Whether they believe the changes are caused by humans or not they are being confronted with visible change.

        In the West we now have year around fire seasons and massive tree die off. In the upper Midwest fishing shacks go out on the ice later and come off sooner. Or don’t go out at all. In New England sugaring is dying out.

        In the SE if there is little change to date then people are not being confronted with the same sort of undeniable evidence.

      • I live in SC, having lived for 20+ in Georgia, and I think there is something to Bob’s speculation. It’s undeniable that there is a conservative bent here, but I do think that the relatively muted net climatic change may still have some effect on people’s perceptions.

  2. But I do think that other authors haven’t considered the possibility that the “warmhole” is a consequence, not of a trend change, but of a sudden shift.

    Thank you for pointing out this rather large shift or natural variation which goes to show that we do not fully understand the full range of natural variation available on this planet.
    Was 2.1 C a 3 sigma shift?
    Then we had a 5 sigma shift in the global sea ice last year with nary a viable reason put forward for a regime change.
    Either we have much larger error bars than we realise or someone has mucked up the measurements.
    CO2 certainly did not do a regime shift at these spots.

    • Sceptical Wombat

      CO2 certainly did not do a regime shift at these spots.
      How do you know? Is it not possible that somewhere a tipping point was passed which caused a discrete change in some feature of the climate (for instance the jet stream) which in turn resulted in the drop in temperature in the SE. I’m not saying that CO2 did cause it, just that there is premature to say that it did not

      [Response: Ice core CO2 data precedes direct measurement of atmospheric CO2, and some partly overlap (e.g. Law Dome in Antarctica). It suggests there was no regime shift.]

  3. Thanks for all you work on this and most especially for making it readable to those of us with only some science/statistics background. I’m still trying to figure out the best place to live as the climate changes :-).

  4. These plots are outrageous, because of the downward jump in the middle. Ok, a lot of bewildering comes from those unrelated two red trendlines, which are unphysical. I would rather have preferred a change point analysis with piecewise linear,but contigious approximation.
    From first look, I would guess a problem with data acquisition, a change in measuring baseline or the like. So similar shifts occur somewhere else?

    [Response: The trend with discontinuity fits so much better than a continuous piecewise linear fit (even allowing for the extra degree of freedom for the discontinuity), that’s strong evidence of its reality. And in this case, it doesn’t require a sudden change in global (or even continental) energy, just a sudden re-distribution. I consider the most likely explanation to be sudden change in the jet stream, bringing consistently colder air, especially in winter. But I don’t know nearly enough about the physics behind it to consider my opinion well-informed.

    I too considered some data problem. But if you look at individual stations, there’s so much consistency from one place to others nearby that the data-acquisition expanation becomes far less plausible.]

  5. An interesting thing I learned in the last month is the marked and rapid (*) effect living things — notably vegetation, plankton, and microbiota — can have upon climate, especially regional climate. I have references if anyone is interested in more details.

    But, related to this post, is the very recent paper, R. E. Alter, H. C. Douglas, J. M. Winter, E. A. B. Eltahir, “Twentieth Century Regional Climate Change During the Summer in the Central United States Attributed to Agricultural Intensification”, Now the region is a tad northwest of the “warming hole” described here (see below), but it does pose the counterfactual of what might the temperature there look like if it weren’t for the corn.

    (*) “Rapid”, at least in geologic terms.

  6. Statistically speaking, this is a really interesting problem, as generally, continuity is assumed for datasets. However, there are cases where it is inappropriate and harmful, although evidence for that inpropriety generally comes from outside the dataset.

    In a political context, for example, Pres George W Bush’s approval numbers are an instructive case. They change abruptly upwards on 11 September 2001. A penalized smoothing spline fit has them heading north in advance of the 9/11 event, and, speaking of non-physical realizations, that would mean the spline prediction was non-causal. The same data are used to show a break at the decision to invade Iraq.

    There is, in fact, an R package which supports this all, I’ve learned: ShapeChange which does changepoint determination and fitting using shape-restricted regression splines.

  7. On the trail of a cause, circa 2012: Leibensperger, et al, “Climatic effects of 1950–2050 changes in US anthropogenic
    aerosols – Part 2: Climate response
    ”, Atmos. Chem. Phys., 12, 3349–3362, 2012.

  8. Thanks, Tamino, on two counts.

    One, it’s an interesting question, one which I’ve been aware of for a while, and have wondered about. (But I hadn’t seen the papers on the topic.)

    Two, as a resident of South Carolina, I’m on the fringes of said ‘warmhole’ and consider climate trends a practical matter!

    A question: I notice that you never actually gave the linear trends for the modern warming era (post ’58, mutatis mutandis). What is that annual rate, anyway? By eyeball, the annual mean temperature anomaly looks to be about 0.25 C/decade, which is pretty darn quick compared to the global figure, but it’s hard to be very precise.

    [Response: Good eye; it’s 0.23 C/decade.]

  9. Lately we have another excursion into misunderstanding of adjustments by Mr Paul Homewood leading to a completely unchecked ideological report by Mr Delingpole (he never does any other sort) which wound up in Breitbart.

    I have checked vs Berkeley Earth and clearly whatever Homewood did wrong it is something HE did wrong, not an error in the science.

    This is perhaps, hot enough to get attention at hotwhopper and I emailed Sou but I surely haven’t sorted the error yet and suspect we may need a keener view of how the mistake occurs (apart from the desire to support the fantasy reality can be made to go away). I suspect a serious misreading of statistics, one place, one month, two years separated by a long period of time, large variation in the data… the ingredients seem to be there.

    Berkeley shows the temperature adjustments are not a commie plot :-)


    • I would guess the reported changes are due to Time of observation bias .
      I am on a cell phone so can not link.
      Victor V has a good post on the issue at sk sc including reference to relevant papers.

      [Response: One of the reasons I chose NOAA data is that it already includes correction for TOB (and other factors).]

  10. The same thing happened in the Netherlands in 1950. A sudden drop in temperature of the same magnitude. It had dramatic effects for summers. In the 25 years before 1950 there were several heatwaves and in the 25 years after it no heatwaves at all. The consensus of climate scientist was that it was caused by a different way to measure temperature. So not a real drop just a artefact. Many temperature data has been corrected since then. Maybe this happened in the US Warmhole just as well?

  11. 1958 is suspicious in that it lines up with the first international geophysical year. I kind of wonder if there was some sort of regional change in measurement methodology as a result of that, which is somehow not captured by the usual homogenization techniques.

  12. Maybe I”m just trying to divert attention from the “Consequences” discussion, but I took another quick look at this post tonight.

    The geographical distribution of the “cooling’ spot (top map) is intriguing. Ancient memories think of this as a region heavily influenced by air masses moving off the Gulf of Mexico.

    What patterns are seen in sea surface temperatures in the Gulf of Mexico?

  13. Russell Seitz (@RussellSeitz)

    DocSnow failed to look past the first V in VVUWT ? Deeply shocked am I.