## Some Oddities with cooling stations

Quick Note: Just so folks don’t get confused, I’m showing the short records, just to give you a brief look at why I move onto to long records. There is no take away from that except this: dont draw any conclusions from short records, So we look at them quickly to show the issue, and move on. Basically we are exploring the data, and in the end, I’ll add up some google earth code that makes the task of exploring much more fun. That’s basically the point of this blog: the tools, the programming, not so much the debate. Plenty of other places for that.

Now, that the whole analysis has been moved to raster, I took some time to play around with a question that has interested a couple of people. Cool stations. A while back when I was looking at ways of bounding uncertainties in the record I went on a hunt for the station that cooled the most and the station that warmed the most. A few weeks later Verity and Tonyb did a post on cooling stations. So, I started down the path again basically to understand how prevalent the ‘cooling stations” are and if there is anything special or unique about them. Now, the simplistic way to think about this problem is that in a warming world every place has to get warmer. Well, that’s just plain common sense. or is it? We can put this question differently. if the average goes up by say .8C in 110 year period how can ANY site see a negative trend? And if we find them, what does that mean? That’s an open question. So I started out down that path, and what I found makes me scratch my head. One way of looking at what I found is this: UHI may play a role. I’m just going to point out the issue or oddity I found and ponder on it. Its definitely not conclusive, but I did scratch my head and wonder about the significance of this. So this is notebook scriblings.

We start by simply looking at the distribution of all trends. For this exercise I’m not correcting for any autocorrelation, I’m basically on a fishing expeditions for the highest and lowest trends. Looks like this

That’s a monthly slope. And without testing that distribution you should be able to see a couple things: it’s peaky with long tails and the mean is going to be positive. Also note that you have some very large warming trends and cooling trends. Since the ends can be revealing, I sorted all the stations by trend and looked through them all. some 5000 charts.

The coolest: 40678349000 SANCTI SPIRIT 21.93 -79.45

Not much of a mystery there. In fact, going through the extremes you will find that nearly all of the extremes are these short records.

one of the warmest: 40371881001 ROBB RSAL 53.23 -116.97

Now, there is a whole separate issue with these short records, leave that aside. So Instead, I went looking at long records. Record that have over 1080 months of data from 1900 to 2009. Why 90 years? No real justification, I’m just exploring. Anyway, as we can expect the distribution shifts. Below see a distribution of the decadal trends for all long stations. While eliminating short runs changes the shape of the distributions and shifts it positive we STILL see stations that cool. random chance? or is there something different about them. Think on that, More tommorrow

OK, I just happened by, and I am too busy with other stuff to get caught up in the quagmire of climate change analysis/argument … but …

Am I missing something, or is it exactly as expected that (1) there would be a lot of variance in the trends observed at different stations, (2) the variance would be smaller for stations with longer records? (Look at the expression for the variance of the estimate of slope in linear regression — I haven’t, but I’m sure it goes as 1/sqrt(n) …) In the simple case where all the variance was due to measurement error, the variance of slopes would converge to 0, but in this case they should converge only to the true variability in temperature trend.

You say: “if the average goes up by say .8C in 110 year period how can ANY site see a negative trend? And if we find them, what does that mean? That’s an open question.”

Doesn’t it just mean that temperature trends are spatially variable? Didn’t we know that already?

1. yes the varience will be smaller the longer the record. Nothing shocking there.

2. Yes the trends vary in space. What seemed worthy of investigation was this: After 110 years of a warming signal ( I do believe in AGW) how many cooling

locations can still exist? And is there anything special about these places that make them more resistant to the underlying trend. One doesnt expect the warming to be uniform or monotonic. But just how variable is it? and is it something that we just put down to chance variation or is there something more there?

I suppose I could just shrug and say ” OF course after 110 or 160 years of gradual warming there will be spots that are persistently cooling.” But,

I’m suggesting a bit more curiousity is in order. ” I could, for example, look at the warming we see and shrug and say “natural variation”

Anyways, yes the short 20-30 spans will have the highest trends. So I was interested in just illustrating that to folks before going on to select the long series

oops, I mean standard deviation goes as 1/sqrt(n) — variance as 1/n.