Botnst - 4/4/2006 10:25 PM
I'm a wacko apologist who has downloaded weather records from 1898 to the present from a dozen or so weather stations in my region. I've been doing this every month or so to look at the regional variation in temps and precip and wind vectors. Without any mind-melding from the liberal control structure in the sky, I personally filtered the data for extreme outliers (ignore > 2 SD +/-mean) and plotted the results.
There are several interesting ways to look at the data. The easiest is to plot the values over time. What you will see is an approximate sine wave which corresponds to the seasons. The amplitude is greater as one proceeds northward. Do you know why?
If you're really clever and take your time, you can remove that periodicity through calculating the periodicity that creates seasonality and subtracting that function from each value. This will result in a much less wavy plot. Or you can do like I did and say, F**k that" and look at the average highs and average lows and plot them. If you're especially ambitious you could even take the difference between those, square them, and divide it by the mean and sum the results. This gives you a measure of central tendency, the variance, over the range of the weather. Are you with me? Yes? If you have the ability to compute the variance, mean, and size of the data set, you can do some serious stats. I leave that for serious people.
Why not do the work yourself, plot it, and see what the actual data looks like rather than depending on some knuckleheaded scientist to tell you what it means. That way you wont be forced to accept on faith whether or not there is a trend in warming over any given time period for which records exist.
In problems like this we usually use a seasonality index table, with decimal values assigned to each time period which always sum to 1. Two columns make up the table - the year in question and the index value. In the particular problem you present, the weakness in the logic is going to be right there, because one cannot assume actual seasonality is not going to be static value for each - it could be affected by global warming itself. How one would compute the smoothing value itself to enter into the index table is an interesting problem in itself.
I don't know if atmospheric weather is the proper place to study global warming, given the differing theories on whether or not warm or cold weather can result from carbon dioxide induced warming due to the role played by the Gulf Stream. I think a better place to measure would be water temperatures in the Gulf of Mexico or some other shallow heat sink. If global warming is occuring, the water should be getting warmer, sooner, each year, with a easy to spot positive trend line over time. Anecdotely, we should see this demonstrated in more ferocious hurricanes, and also see hurricanes tracking further to the west/northwest due to an increase in overall kinetic energy.
You might want to try plotting your results on an Xbar & R chart, leaving the outliers in. Sometimes one finds all the outliers on one side or the other, which indicates a cause other than natural chance at work. In much of the work I do, we are trying to find whether or not outliers or inliers that are not plotting as a sine wave around Xbar are due to common chance or "special cause". In other words, a string of outliers should correspond to a volcanic eruption, for example, while ones occurring by chance should appear as occasional blimps.
Just about every data set of random measurements over time plot as a sine wave +/- 2SD about the mean. Shewhart said it is a natural property of random data sets, and I've seen it all my life. It is the behavior of the sine wave that indicates whether chance or action, especially human action, is at work.