Fair-ish and Balanced-ish
Monday, October 13, 2003
The Radiosonde Network: Trends, Accuracy and Outliers
In the study of the Earth's climate history, good data is absolutely essential. And one method of data quality control is to compare data-sets from very different techniques. If they match, then it's a good sign. If not...
However, like all things in life, it just isn't that simple. For example, are the two data sets measuring the same phenomena (in many cases, they are measuring slightly different phenomena)? What are the margins of error on both data-sets (when large margins of error are involved, a spurious correlation can often be observed)? There are many more problems, but I'm sure that you get the picture.
Now, onto the actually topic of this post: the radiosonde network. A radiosonde is essentially a weather balloon. Various stations around the world have been releasing them for years now, and collecting lots of valuable data on the earth's atmosphere and how it changes over time.
But, what happens when we alter the data-set slightly? Fortunately this has already been done for us by James Angell, a NOAA researcher (Effect of exclusion of anomalous tropical stations on temperature trends from a 63-station radiosonde network, and comparison with other analyses, Journal of Climate 16(13) page 2288).
Angell took a 63 radiostation network and looked for outliers (stations which give very different readings of tropopause temperatures relative to the average). He identified 9 and removed them from the data-set.
Now what happened to the trends in the data; in the Tropics the new data-set showed a warming trend in the Tropopause (there used to be a cooling trend), the previously observed warming in the Troposphere increased substantially to 0.13K/decade. When one looks at the global as a whole, rather than the Tropics, the trends decrease but there 54 station network shows greater warming than the 64 station network. These trends are from 1958 – 2000.
Interestingly, when Angell compared the 54 station data-set to the 63 station data-set, from 1979 to 2000, the differences where substantially less. This is indicative of the problem fading over time (perhaps better experimental techniques?).