The question arises: how might temperature variation have changed? We all know (if you don’t already, then read this blog!) that things like temperature show a combination of trend and fluctuation. We almost always focus on the trend, because that’s what shows the most obvious changes over time. But those fluctuations … the “noise” that we add to “signal” to get “data” … the “variation” we add to “average” to get the “weather” … are they changing too? Or are they just doing the same old same old kind of fluctuation they’ve been doing all along? This is a very different question than we usually hear about in discussion of climate change, not about a change in the average temperature, but whether or not the fluctuations have somehow changed.
They’re prominent in today’s news on two continents. The great plains of the USA are shivering through some of their coldest temperatures on record as the “polar vortex” invades from the north. Meanwhile, Australians suffer through their hottest month and worst heat waves ever, hell on earth for a place already known for it’s heat.
But mainly, it’s that polar vortex thing. Some have suggested that the rapid warming of the Arctic compared to the rest of the world, plus the dramatic decline of Arctic sea ice, have changed things in a fundamental way. It has thrown a monkey wrench into the jet stream, and during winter it can cause the polar vortex to fragment, part of it diving southward and bringing the deep freeze with it.
As a first glance, I’ll look at monthly average temperature for a specific region, in the northwest corner of Maine. It’s one of the grid boxes for the Berkeley Earth Surface Temperature gridded data set, and here’s the temperature anomaly throughout time, for the winter months (Dec/Jan/Feb) only:
The solid red line is a smooth fit. It too fluctuates rapidly (probably too much, but it certainly captures what’s going on with the signal. But it captures too much of the noise, so we shouldn’t use it to estimate the trend.
But we can use the residuals — what’s left over after subtracting that smooth — to estimate how the noise may have changed over time:
It certainly looks like they’ve tended to get bigger over time. But “looks like” is poor statistical analysis. Let’s look at the squared residuals, which will all be positive:
That doesn’t just “look like” it’s increasing over time, statistical analysis backs up that conclusion.
You might suspect that I used the gridded data from Berkeley Earth because I intended to do all the grids, not just northwest Maine. Right you are.
I took just the data since 1900 (I’m not so interested in what happened before then), fit a straight line, then analyzed the residuals to look for a change in variance. Here’s a map, with red dots where the variance seems to be going up, blue where it’s going down, larger dots for faster changes:
The greatest change is happening in a ring around the Arctic circle. Imagine that.
By no means is this a sufficiently rigorous analysis. It would be better, I think, with daily rather than monthly data (and I think Berkeley Earth provides a gridded data set for that too). A better smoothing method could be settled on, a better way to define what’s the probably “noise,” and a better way to define how things truly vary.
But at least it’s a start. If we take a close-up at the results near the USA, we see this:
The eastern USA has, overall, seen an increase in variance, while much of the western USA has seen a decrease. What will the future will bring?
This blog is made possible by readers like you; join others by donating at My Wee Dragon.