Climate Data: Effects of Station Choice – Location

Apologies for the lack of posts recently.  Big things are afoot and it is still too early to give even a taster of them. I have been trying to keep up with developments elsewhere though and I’m watching with interest Jeff Id’s analysis of the CRU data using the GHCN dataset. His latest post is showing something very interesting.

First the location issues.  For anyone who has been following E.M.Smith’s analysis of the GHCN data that feeds into GIStemp, his assertion is that most, perhaps all of the temperature increase ascribed to anthropogenic global warming is due to the changing bias in thermometer location (see Figures 1 and 2). He said at the time: “This implies that the number of thermometers active in any particular period of time has a strong impact on the GAT [Global Average Temperature] in that time.”

Figure 1. Chart showing the distribution of temperature station records by latitude according to E.M.Smith’s lattitude band classification (link to original data).
Figure 2. Comparative numerical plots of data in Figure 1 charts: [a] linear scale, [b] log scale.

The defence of the climate modellers has always been that the locations don’t matter since all the data is converted to anomaly values – that is compared to the average temperature of a reference period (1951-80 for GISS and 1961-90 for GHCN and CRU).  The logic in this seems unimpeachable, nonetheless the more analysis that is done of any of the datasets, the more evidence there seems to be for some sort of bias.

Jeff Id’s analysis (CRU post #2) – specifically Figures 6 and 8, showed that, an unweighted average of station data showed little warming, but suggested that station choice and adjustment may be a factor in the producing the more familiar warming graphs. Analysis by Andrew Chantrill, using a different method but also without weighting the data series in any way is here.  When E.M.Smith looked at the most longlived thermometers in the temperature record (top 3000), there was a cooling trend by decade untill 1999 (Figure 2).  It would be interesting to analyse exactly which thermometers remained in this set after 1989 and understand in more detail the basis for the sudden jump in temperature.

Figure 2. Average temperatures calculated for data from the 3000 longest station records in the GHCN data as used by NASA GIStemp.

In Jeff’s most recent post (CRU post #3) the analysis has been extended by providing a weighting to the data by placing it in a grid so that where many data sets fall in a single grid cell they do not drown the signal from a single set alone in an adjacent cell.  An example of a grid used for climate data analysis is shown in Figure 3. Once grid weightings are introduced, some of the familiar global warming starts to appear in the resulting charts.

Figure 3. Equal area grid cell map. (Source: NASA GISS)

So if the unweighted data shows little warming, there are a number of ways the warming may become apparent (or is artifically introduced, depending on your viewpoint):

  • From changes in the reporting thermometers: changes in number and location of stations
  • Biases due to selection of stations from the available pool.
  • Biases in the way data is used from reporting stations – including adjustments, failure to use all of the available data, and infill of missing data. 
  • Differential weighting of data due to location

There are so many potiential biases and errors here. And we’re told that the Earth has warmed by 0.6-0.7 deg C. Is it any wonder so many computer hours are currently being spent putting this data though so many variations of analysis?

This entry was posted in Uncategorized. Bookmark the permalink.