And you can always write better programs – use Perl::Critic to check your programs.

As I understand it “trend” is short for “linear trend”. In other words, a computed trend is simply a linear least squares fit to a time series over the period of interest. One hypothesises a linear model and ascertains whether the hypothesis is sustainable. I have some experience with climate data of various types, especially station data, and it is my constant and overwhelming experience that they do not change with time in an orderly fashion – which is reasonable enough given that they are generated by an assembly of chaotic systems. My work has convinced me that although station data may be remarkably linear wrt time over very varying scales, they are liable to exhibit step function characteristics at seemingly random intervals. This is remarkably simple to demonstrate graphically – GIFs can be provided if anyone is interested. Last night I looked at the Brenham data (stored in my download of GHCN data from about 2003). I have also looked at Lampasas. Both these sites have very clear regime changes, for example Lampasas at March 1957, magnitude -1.1 C approx, and again just before the data end, April 1998, magnitude about 1.5 C.

I’m happy to post the evidence as a few GIFs, if anyone asks.
OK, the data may not be wholly reliable, for observational reasons that continue (rightly) to exercise many contributors. I believe that this type of analysis may make it simpler to identify step changes that might be associated with instrumental or location changes.

Meanwhile, I counsel against basing too many inferences on data whose “trends” computed over an arbitrarily chosen period might simply be artefacts of observational misdeeds. I think it might be safer to fit linear trends only over periods where the data do not contain step changes The GIFs I’ve generated will show fairly conclusively where these periods are.

I do hope this arrives safely!

Robin

I have looked at regional spatial data all my career and I agree with your statement that the variability shown in your maps reflects other than high quality. The experienced eye is offended.

Your readers are being kind and apologetic in volunteering causes such as humidity and climate cycles (though these might have some effect in the final analysis). The main problem seems to be quality. For example, the change to red shades on adjusted 1980-2002 (your Map 3 on #37) arouses immediate suspicion that the adjustments changed the colours, not humidity or human settlement.

Couple of questions. Why use anomaly maps (see #44 Steve Mosher)? Is there not a better method of normalisation so that the starting point shows no anomaly? Best to start with a white map Number One?

Next, is the gridding/contouring package you use of a standard that your ore grade estimation geological friends would accept? There are geology packs that are pretty sophisticated. (Not being critical, I just don’t know the answer).

Observation. I have seen many maps of the USA 48 over the decades, for factors like the frequency of fatal road accidents, the age of children on leaving school, income levels…… Almost always there is that same vertical divide, as if the USA East was fundamentally different to the West. I can offer no more than the observation. (It’s like an ore body straddling two rather different rock types. The frequency of sampling can require more density in one rock type than the other).

Repeat. One of the major problems of interpretation remains the definition of stations unaffected versus affected by UHI and their subsequent use in adjustment. You can’t subtract a UHI station from a UHI station and expect the difference to show meaning. I have done some initial work locally and suspect here in Melbourne UHI started about 1920. Comparison stations within 20 km were heating 30 years later. Today they might have to be all considered affected. Geoff.

USHCN Survey Results based on 33% of the network

The southeast station’s showing little or no warming intrigued me. Referencing the CRN rating guide that Anthony uncovered, I suspect that very few southeast stations would achieve a rating of class 1 or 2 as there are tall trees everywhere here, and very few locations would meet the 5 degree to the horizon criteria required for class 2, let alone 3 degrees for class 1. Having excessive shade might explain a lack of warming, or even cooling. 50 or 100 years ago, some stations might have been in areas where it was clear cut. If you clear cut in the Southeast and wait 30 years, you get a wooded area with skinny, but very tall trees. If you aren’t actively clearing it, the trees are going to grow. You can see a lot of shade on the Georgia sites on Anthony’s surface stations.org, including pictures that appear to be mid day.

I just wanted to pop in and share that I have attempted a more difficult approach to the analysis rather then picking up the v2.min.z, v2.max.z or v2.ave.z documents.

Instead I tried the station by station data here: http://cdiac.ornl.gov/epubs/ndp/ushcn/usa_monthly.html

I then looked at the Tmin the Tmax and the Tave values in relation to the Tflags so that I could coordinate similar data together. I look for a “1O” flag to be associated with the monthly temperature before I include it in my analysis. The other issue is that some of the data banks go back as far as 1865; however, to try to keep everything on the same base line I have found that if I use the 1936 date that I can get a better line of consistency across all the sites.

The problem is, I believe I need a minimum of 90 years to have statistically significant data. In this case I went back and restarted my collection of sites based on a minimum of 90 years of data in which the temperatures had the “1O” or “1E” flags associated with the monthly values. This has drastically cut down on the available sites; however, except for certain areas like S. California or in the shadow of the industrial belts of the US, the data seems to have a fairly good overall quality.

I am envious of your tools as they appear to make spectacular graphics. However, with my old copy of Office and my Fathom 2 Statistical analysis tool I have found that I can get some pretty convincing data from the ushcn (formerly gchn) site. I would be curious as to what your graphics would look like if we applied a little filtering so that the data quality was a little more homogeneous and tracked not only the Tave; but, included Tmin and Tmax…

Dave Cooke