Friday, May 29, 2009

Iowans can't measure the temperature



Belle Plaine and Toledo are 18 miles apart with only 30 ft difference in elevation. The daily temperatures vary by as much as 20 degrees, and does so far more often than would be expected by cold fronts being between the cities. The red curve is a 365-day running average of the difference between the average daily temperatures. That curve for decades was over 2 deg F , with Belle Plaine hotter. If this temperature
difference was real, it should have led to decades long thunderstorms and winds. Noted the sudden onset of bad data in about 1964 I thought it might be a station move, but the station record doesn't show it moving then. It stopped again in 1998, although the noise didn't go back to 1950s levels.

The small green line in 1965 is 100 years of global warming. Clearly the noise for one year is greater than that. There is no way we can nudge global warming out of signal this bad.

Here is why. When you make a measurement of anything, and you decide to adjust it for an error, you are saying that the error in the raw measurement is the value of your adjustment. That becomes the error bar. Consider this from a site dedicated to explaining error analysis.

The accuracy of a measurement is how close the measurement is to the true value of the quantity being measured. The accuracy of measurements is often reduced by systematic errors, which are difficult to detect even for experienced research workers.Source

Now that was a statement about linear systematic errors. What we have in the temperature stream is nonlinear systematic error. If you make a 2 deg correction for this period of time, you can't claim that the accuracy of the data is less than 2 degrees. This has profound implications

Over the past century global warming is said to have warmed the world by 1.1 deg F, yet they are making corrections of +/-2 degrees to the data stream. That means then that one can't possibly, say that the world has warmed by 1.1 deg F. Statistically it is lunacy to say that the world has warmed by 1.1 +/- 2 degrees, which means it might have cooled by 1 degree or it might have warmed by 3 degrees. The data doesn't allow us to tell.

Before I make the final comment, I want to let everyone know that the data for the chart above came from Dave's favorite site.here. It is a government site.

One final thing, in the comments some have criticized me for using CO2science previously as a source for my yearly data. I compared CO2 science, which is raw with the final edited data for Belle Plaine. You can see below that the differences are minor and thus the criticism is simply that of people stretching to find anything wrong with what I am saying. What am I saying? The data is crap.

6 comments:

  1. Like in the Clinton-Morrison couplet I also downloaded this data and went through and ran a difference. The mean difference was -0.93deg with a 95% confidence on the mean of +/-0.19deg.

    This is an example of the importance of the Central Limit Theorem in statistics.

    Again, as in the other data set, there are disturbing differences in the data set. I believe this is called error. Error occurs in a lot of data sets.

    Again, thankfully, the climate models are verified by using gridded averages on a continental scale. The fact that there's a truckload of data here allows one to ascertain how "dramatically different" the stations are on average. In this case it is less than a degree on average for the 1948 to 2005 time frame with a confidence on that mean calculated of 0.2.

    Again, the reality is that data collected in the real world, in the "field" is not perfect. That's why its important to look at the data on a larger scale. The importance of blocking the data in these ways, gridding the averages and taking continental scale data become important.

    But this speaks to a larger topic in statistics. In any given distribution there's a spread of data. The importance of the central limit theorem is that with sufficient data the data will take a normal distribution, and the calculation of the confidence interval on the mean scales with the inverse of the square root of the number of samples. More samples more confidence around the mean.

    But, as I stated before, I'm neither a scientist nor a statistician. But anyone can take data and pick it apart at a very fine level. The power of data comes in large amounts of it, not in picking individual data points.

    ReplyDelete
  2. Correction: I noted that some of the data didn't line up (no small feat to find the misalignments in over 20,000 lines of data), so when I reran the calculation I find a mean difference of -0.90deg with a 95% confidence on the mean of +/- 0.074deg which is a bit better.

    ReplyDelete
  3. Another Correction: I noted that in this data set as well there's a few missing blocks so the numbers didn't align either. So I had to go through the >20,000 lines of data and track down the misalignments and recalculated. This round I found a mean difference of 0.70degress with a 95% confidence interval on the mean of +/- 0.062deg. Again a bit better.

    I will have to double check this but it looks like the dates line up much better overall.

    ReplyDelete
  4. Sorry, that second correction was supposed to be related to the Iowa Illinois set. My bad.

    So here's the score:

    Toledo - BellePlain mean difference = -0.9deg with a 95% confidence on the mean of +/- 0.07deg

    Iowa - Illinois sites mean difference = 0.7deg with a 95% confidence on the mean of +/- 0.06deg

    ReplyDelete
  5. This post is really a piece of work. Thank you for this informative content.

    ReplyDelete
  6. Sorry, that second correction was supposed to be related to the Iowa Illinois set.

    ReplyDelete