[Vision2020] US temperature station trends

[Paul Rumelhart]

Vision2020 peeps that haven't been turned off by the perennial global 
warming debate,

I have been playing around with data from the U.S. Historical 
Climatology Network, specifically looking at their V2 dataset of 
averages, both raw and adjusted.

I ran some numbers on the data to find out what the trend-lines were for 
the individual stations.  I used a simple least squares linear 
regression function to convert the data points into a linear trend line 
in slope/intercept format.  The intercept is meaningless for what I'm 
looking at, but the slope should show the increase or decrease in 
temperature per year. 

When I ran the numbers for the U of I's station, I get the following:

raw data: -0.00612
 adjusted: -0.00504

Since the data is plotted as degrees Fahrenheit vs. year, then this 
should mean that the raw data shows a 0.612 F decrease in temperature 
per century, and an adjusted figure of a 0.504 F decrease in temperature 
per century.

I had assumed that Moscow was in the minority as far as trend-lines go, 
given that from everything I've seen the mean global temperature of the 
Earth is increasing (I just disagree on why).  So I looked at the 
adjusted data for all of the stations in the file, and counted the 
number of positive vs. negative trend lines.  This is what I found:

Adjusted data:
    Positive:  1035
    Negative:  183

This seemed reasonable; I would expect more stations to be going up than 
down.  I was surprised it was so many stations showing a positive trend, 
but that's what I see in the data.

Since the trend line for Moscow changed once the adjustments were made, 
I ran the linear regressions for the raw data to see if any others had 
changed:

Raw data:
  Positive:  671
  Negative:  547

This really surprised me.  Apparently, their adjustment process caused 
at least 364 (547 - 183) of the 1218 stations (29.9% of them) to change 
sign as far as their trend line was concerned.

Now I'm really curious what their adjustment process involves, which I 
will be looking into soon.

Also, let me make it clear that it's been a long time since my last 
statistics course.  I think I'm doing the linear regressions correctly, 
but I could be doing them wrong.  I'm also using the simplest form of 
data fitting, I'm sure there are more advanced algorithms to find a 
better fit to the data for linear trend lines.  I'm not even sure that I 
should be looking for linear trend lines, since I'm not sure the data 
behaves linearly.  However, it was a fun project. 

If anyone wants to do their own analysis, I used the following files at 
the following URL:

ftp://ftp.ncdc.noaa.gov/pub/data/ushcn/v2/monthly/

9641C_201008_F52.avg.gz - adjusted monthly averages as of August 2010
9641C_201008_raw.avg.gz - unadjusted monthly averages as of August 2010
ushcn-v2-stations.txt - list of stations and their metadata.

I'd be happy to provide my source code if desired (as long as you don't 
laugh too loudly), but anyone wanting to replicate this should probably 
do their own analysis independent of mine.

Any thoughts?

Paul