USA Divisions

We’ve occasionally looked at temperature time series from NCDC (National Climate Data Center) for USA48 (the 48 states of the continental USA excluding Alaska). But in addition to data for individual states, NCDC also offers data for each individual climate division within each state. The temperature data, for example, can be found here.

Segmenting USA48 by division rather than state will give us a much more detailed geographical picture of climate change in the mainland U.S. Let’s take a look at some of the things the data have to say.

First let’s see how the trend rate (from linear regression) breaks down by division. Here’s the trend rate for the entire time span (from 1895 to 2012.5), with red circles indicating warming trends, blue circles cooling trends, larger circles larger trends (click the graph for a larger, clearer view):

The strongest warming is in New England, the upper midwest, and the desert southwest. Much more of the country shows overall warming than cooling, but there are still a good many divisions with negative overall trend, especially in the southeast. The blue color is easier to see if you click the graph for a larger view, because the blue circles that are present are all rather small.

The U.S., like the globe, has not shown uniform warming (or cooling) throughout the entire time span. Here’s the same information, but trend rates are for the time span 1895-1910 only:

Note that many of the 1895-1910 trends are huge compared to 1895-2012.5 trends. But they’re not really that indicative of trend because the time span is so short (a mere 15 years) and divisional temperature time series are noisier than global, so the trend uncertainty is quite large. Some regions (including New England and the desert southwest) cooled substantially, others warmed, but these can’t properly be called trends — I’d call them fluctuations, or natural variation.

From 1910 to 1940 the map looks a lot different:

With only a few exceptions, the nation as a whole warmed during this time period. Although warming is strongest in the great plains and a (very) few spots show cooling, it’s a bit surprising how uniform the trend is across the country. Then, from 1940 to 1975, the situation reversed:

With few exceptions, most of the country cooled off. The eastern half shows consistently stronger cooling than the western.

Since 1975 we’ve seen near-universal warming, with only 4 out of 344 climate divisions giving negative trend rates:

People sometimes wonder what the “more recent” trends are. Again, with short time spans we can’t really consider them trends as much as fluctuations, but here for your enjoyment are the linear regression trend rates from 1990 to the present:

Over this brief span there’s been cooling in the Pacific northwest and the southeast (especially Florida), but most of the country has warmed, especially New England and the upper midwest. If we limit “recent” to just the data since 2000 (very brief!) we get this:

Again, some of the estimates are very large because of the brevity of the time span (a mere 12.5 years). Some regions have shown large variation, warming in New England, and cooling in the west.

We can also create anomaly maps for any given month, showing how much each climate division deviates from its long-term mean. I computed anomalies based on the entire time span of data, so the baseline period is 1895 to 2012.5. We’ve just experienced the hottest 6-month episode on record, so here are anomaly maps for the last six months to show where the heat has been concentrated:

Hot times have moved around the country, but been strongest overall in the upper midwest/Great Lakes/New England, while cooler times have been consistent in the Pacific northwest. Although the heat was consistently extreme for the nation as a whole, the most scorchingly hot extreme wasn’t the heat wave we’re in right now (covering the latter part of June and into July), but the amazing record-smashing “summer in spring” we had during March.

Last but not least, let’s look for recurrent patterns of time changes among the climate divisions. A natural way to do so is with principal component analysis. The strongest PC (“principal component”) is represented by this combination of time series records (red indicates positive coefficients, blue negative, larger circles indicate larger coefficients):

Essentially, this PC represents (is roughly proportional to) the average temperature over the eastern half of the country. The time series pattern looks like this (black is annual averages centered on Jan 1 so the final data point has a full 12 months, red is a lowess smooth, blue is the linear regression trend line):

We can see the overall pattern of warming, with rapid and consistent rise since 1975, and monster heat during the last 12 months.

The 2nd PC represents a different pattern:

With red in the west, blue in the east, in part this represents the east-west difference (west minus east). But there’s more to it than just that, because the red (positive) coefficients are consistently larger than the blue (negative). Here’s the time series pattern:

PC3 is essentially the south-minus-north pattern:

The time series for PC3 shows that the south has cooled relative to the north, or to put it another way, the north has warmed relative to the south:

PC4 is “coasts minus great plains”:

It has this time series pattern:

Higher-order PCs represent more complex geographical patterns, for instance the interesting map of PC6:

Those of you who are familiar with “orthogonal functions” will recognize that PCs like this one are really “empirical orthogonal functions” for the geographical pattern of temperature changes.

There’s a lot of detail in the set of temperature changes by climate division, and many interesting conclusions to be drawn. Perhaps some day I’ll give this the attention it deserves, but for now I’ll just note a few things:

  • Overall warming, especially since 1975.
  • Strongest warming in New England and the upper midwest, weakest in the Pacific northwest.
  • March 2012 was a monster.

    Expect further changes.

  • 15 responses to “USA Divisions

    1. That’s a great visual illustration of PCA components as orthogonal functions! I’m going to have to bookmark this for teaching purposes.

    2. It would very nice to see an animation of the “maps”, showing (perhaps) 20 year trends…

    3. …would be very nice to see an animation of the “maps”…

      Seconded, but all the same these are terrific.

    4. Tamino, you should apply a cutoff for SNR of each orthogonal mode. This way you could denoise the data.

    5. Just to add an absolute congratulations on the graphics. Tufte would be proud! An extremely effective method of display.

    6. Regarding the 1940 – 1975 cooling: some studies note a correlation with more than 1000 atomic bomb tests over that period – many explosions were atmospheric, injecting dust into the stratosphere. The practical effect was a mini-nuclear winter.

      • K.a.r.S.t.e.N

        I am not aware of any study which makes a convincing case with regard to these tests. The nuclear winter theory requires surface burning smoke to be effective. The main culprit for the post-WWII temperature dip is anthropogenic sulphate aerosols. On top of that, black carbon aerosol emissions over the US peaked in the 1920’s already which presumably imposed an additional cooling effect until the 1970’s. In contrast, the rapid rise in US-BC emissions in the first quarter of the last century might have enhanced the warming between 1910-40.

        It is once again a terrific and enlightening analysis from Tamino. Not only becomes the (south)eastern US aerosol-induced cooling between 1940-75 apparent, but also the higher subsequent warming signal over the same region (albeit with warm patches over the southwest as well) seems to evolve out of the noise (1975-2012). Otherwise, my awfully “under-skilled” eyes suggest that the first PC nicely separates the elevated western parts of the US (Rockies and to their west) from the rather plain terrain in the central and eastern parts of the country (Great Plains all the way to the Atlantic Ocean).

        • I think it stems from this.

        • K.a.r.S.t.e.N

          Thanks for the link JCH! Now I kinda remember where I came across this claim. It got published in the “famous” Journal of Atmospheric and Solar-Terrestrial Physics: Fujii 2011

          To be honest, for this to become a plausible substitute for the “sulphate-emission-theory”, the author got to work a bit harder to make his case. In my point of view, he goes beyond what’s in the actual literature he bases his claims on, namely Turco et al. Science 1983. Without city-buring and wild-fires (caused by the bombardments), the effects are negligibly small, particularly over the southern hemisphere. To his credit, he get’s the AMO cause-and-effect-relationship right.

        • I find the proximity of WW2 and the temp drop fascinating. It was a global combustion event. Each 75mm halftrack in my father’s platoon on Iwo Jima fired approximately 12,000 shells in 36 days. The island is volcanic. Throw in Naval bombardment, aerial bombardment, all the artillery, the tanks, etc., surely that equals a volcano! But no.

          I spent a couple of working out estimates of cities burned, ships sank, explosives detonated, snow darkened on the Russian front, etc., and then I found that article. Still not enough!

        • JCH, the same thought occurred to me, though I didn’t try to work up an analysis. One of the things stopping me was the scope of the war itself–the mention of the ‘Russian snow darkened’ hints that you made a considerable effort to deal with it.

    7. A really nice method for visualization when there is a number of comparisons to give context. However, going beyond three axes on the PCA is beyond what I can understand.

    8. Don Fontaine

      How do you determine the significance of the PC eigenvalues? I currently only have arbitrary, “explains XX% of variance”, type of understanding.

    9. Are the size of the circles in the anomaly maps proportional to the absolution numerical deviation from the mean, or the standard deviation (sigma-levels) from the mean?

      [Response: Absolute numerical deviation.]

    10. If you look at the regional temperatures you’ll see:
      no significant warming in the South, Ohio Valley and the South East since 1895;
      no significant warming since 1986 in the North West, the Northern Rockies and plains and theWest;
      no significant warming since 1994 in the South West;
      no significant warming since 1998 in the Upper Mid West and the North East.

      Explain please.

      [Response: “Since 1998” … really?

      Hmmm … Shall I point out the much stronger and far more numerous significant warming episodes for regions, the nation as a whole, and the globe, then insist that you “Explain please”? … Should I muse about the much higher level of noise, and of natural variation, in regional temperature than in national or global? … Could I mention that the lack of “significant” warming isn’t the same as the lack of warming … Might I point out that climate science actually predicts regional and temporal differences in temperature change …

      Nahhh … there’s a much simpler, and completely correct, “explanation” for your list. It’s called “cherry picking.”

      I suspect that’s the best you can do.]