Since 1895, the conterminous USA (lower 48 states) has warmed significantly:
The red line is a modified lowess smooth, on a time scale which mimics 20-year moving averages.
I decided to look for patterns in the geographical distribution of temperature change, using principal component analysis (PCA). The warming pattern of the 1st principal component (1st PC) matches the nationwide warming pattern excellently:
The warming pattern is on the right; on the left is a map showing how each of the 344 climate divisions match the 1st PC. Red dots means a match, with bigger dots meaning better match. The interesting thing is that the match is outstanding for the eastern half of the country, but not so great for the western half.
Does that mean that the western USA hasn’t matched the long-term trend shown by the national average? In fact, no.
PCA identifies patterns that are common to many of the climate divisions. The fact that the western half of the country doesn’t match so well isn’t because the western trend differs from the 1st PC, it’s because the fluctuations diverge. The eastern half of the country tends to show the same month-to-month fluctuations (these are all monthly data from NOAA, although I’ve plotted yeary averages for greater clarity), but the western half, in spite of showing a very similar trend, tends to show different fluctuations.
I’m primarily interested in the longer-term stuff, not the short-term month-to-month variations, so I decided to take a different tack. Instead of doing PCA on the base data (monthly averages), I first smoothed the monthly data for each climate division with a modified lowess smooth, using a time scale which mimics 10-year moving averages. Then I computed PCA on those values, to seek patterns which tended to be common among the “roughly 10-year moving averages” rather than the monthly data. This, I believe, will tell us more about how the trend patterns tend to group together, rather than putting so much emphasis on how the fluctuation patterns group together.
With that strategy, I got this for the 1st PC:
The 1st PC still matches the nationwide trend well. But now the geographical distribution of how it matches (the map on the left) is very different. Instead of an east-vs-west difference, it’s much more uniform across the country, demonstrating that the USA has participated in the nationwide warming more uniformly.
I do note that the match is considerably weaker in a region of the southeast, centered on Alabama and Mississippi. This is the location of the “warmhole” which I blogged about here. It appears that the strategy, to focus more on trends than fluctuations by using PCA on long-term rather than short-term data, has worked. It appears to have identified the most prominent pattern of deviation from the nationwide average behavior, i.e. the “warmhole.”
If we look at the 2nd PC, find something quite interesting:
There’s the warmhole again; the red dots show where the local pattern matches PC#2, while the blue dots show where it “anti-matches.” Looking at the time series pattern, we see that the warmhole so identified shows a sudden drop in the 1950s. This is in accord with what I found in my earlier post about the warmhole, in which (looking at it closely) the data suggested a sudden drop in 1958. Bear in mind that in the PCA, the time series pattern represents the difference between the “red-zone” pattern and the “blue-zone” pattern, so the warmhole cooled significantly in the late 1950s relative to most of the rest of the country (especially the north and the west).
Also of interest are patterns for different seasons. I’ll show just one now, the 1st PC for summer (June-July-August) temperature:
This illustrates that summer in the USA showed quite an outburst of hot temperatures during the 1930s, reaching a peak comparable to present temperatures. Other seasons show no such 1930s outburst; Autumn, for instance, shows strong warming over the last 20 years but nothing of the kind during the 1930s:
There are many other interesting things, including other seasons (winter and spring) and other principal components with interesting geographical patterns. I hope to post about those soon.
This blog is made possible by readers like you; join others by donating at My Wee Dragon.