“When standardizations are confined to the calibration interval only, pseudoproxy reconstructions performed with RegEM-Ridge suffer from warm biases and variance losses.”

From the paper, no, Letter:-

“We use a method (notes 9,10) adapted from the regularized expectation maximization algorithm11 (RegEM) for estimating missing data points in climate fields. RegEM is an iterative algorithm similar to principal-component analysis… We assess reconstruction skill using reduction-of-error (RE) and coefficient-of-efficiency (CE) scores as well as conventional correlation (r) scores.”

Note 10 is “Mann, M. E., Rutherford, S., Wahl, E. & Ammann, C. Robustness of proxy-based climate field reconstruction methods” and note 9 refers back to Rutherford.

Janama @ #156 “no Australian data” The Australian sector is on the cooler East side.

The link to the letter @ #109 has gone dead now, is that the norm after x days? (thankfully I’ve got it). Cannot get Methods without paying – all this seems like more varieties of data that should be publicly available and isn’t.

Of course, no code and only links to raw data (AVHRR data does not even seem to be uptodate) …

No,you didn’t screw it up. A 2 to 1 ratio of radii becomes a 4 to 1 ratio in area and the eye reacts to that. The square root takes it back to the correct relative relationship. Changing the 4 to a 2 reduces the size of the larger circles.

It does make a difference!

Did I screw up the R?

split.screen(c(1,2))
screen(1)
plot(sta.coord$x,sta.coord$y, pch=19, asp=1,col=c(“blue”,”red”)[1+(decmean.latex>0)],cex=4*abs(decmean.latex),xlab=”",ylab=”",axes=F,main =”Plot of Mean Station Trends (Radius)”)
map(anta, fill = F,add=T)
screen(2)
plot(sta.coord$x,sta.coord$y, pch=19, asp=1,col=c(“blue”,”red”)[1+(decmean.latex>0)],cex=2*sqrt(abs(decmean.latex)),xlab=”",ylab=”",axes=F,main =”Plot of Mean Station Trends (Area)”)
map(anta, fill = F,add=T)

Excellent point! This type of adjustment is usually done when plotting histograms with unequal width bars because in that case also the eye sees the area instead of the height of the bar.

I actually tried this (AFTER I had already posted the script). k = 2 works pretty well.

This same effect was noticeable when I tried plotting the results for the AWS as well as the stations on the same graph. I used squares for the AWS and circles for the staions. Even adjusting the relative sizes of the two symbols to match area didn’t help a lot because it got too crowded with a lot of overlapping.

Just regarding the plot, the eye (mine at least) compares the circles by area not radius. It might be better to use:

cex=k*sqrt(abs(decmean.latex))

for some choice of k.

Steve: I agree with this point; I’ve used this form of expression from time to time and like it a lot.

I follow it up and found a bug in “gnuplot”. If you run many linear plots for a prolonged time, it starts to give erroneous results. Annoying because if i can’t find a fix I’ll have to change the software.

The only chart I can find indicates a cooling trend but only goes up to about 1999 and no linear trend line.

That’s interesting, Martin. The Steig paper declares the Vostok trend as being +0.1C/decade over 1957-2006. Perhaps you have discovered a major fault in the paper? I think you should follow this up.