Since the release of the Met Office Subset recently, I have been looking further at the data trying to evaluate what exactly is contained therein. One of the checks that one would like to do – determining how the global gridded data is calculated – is not really possible. The data set is self-admittedly incomplete and, anyway, a list of which stations are used and which aren’t is not provided. However, I decided to look at the grid cells information of the provided stations anyway to see what was going on.
Using another fairly extensive file of station information which I already possessed, ghcnd-stations.txt (which I believe to be downloaded from the previous incarnation of ClimateAudit.org), I did some checking of Met station information against the ghcn file. Using only those that I could easily match up in the two data sets, I found differences in coordinates and elevations of a number of stations. Although the heights (altitudes) need to be looked into further, because some of the differences can be quite large, the grid squares of most of the numerically different coordinates seemed to be unaffected by those differences. Except …
I have mentioned before that for some unknown reason, Met and Cru prefer to do the opposite of what one might normally expect for coding longitude values. East of Greenwich, their longitudes are negative and those west are positive – not what one would expect for drawing maps and not what one might generally find in other global reference venues. Unfortunately, this seems to have also caused some confusion for the Met Office.
From the Met data subset, the meta-information describing the following stations in the South Pacific was:
I plotted the locations in R (re-centering at the dateline – longitudes are measured from that coordinate). Grid cell boundaries are given in green. The graph on the left is the result:
The graph on the right is based on the coordinates gleaned from Wikipedia (yeah, I know…, but I trusted them anyway). The most obvious problem is that Nadi Airport from Fiji seems to have been relocated into the open ocean hundreds of kilometers from where a plane could safely land. But that is not the only error. Each of the four sites printed in blue on the right had the wrong sign for the longitude in the data. The net effect is that all four are NOT in the correct grid cell.
Is this important? I don’t know since we are not informed as to what is used in the calculation of the gridded data.
Could it have an effect on the global land series? Who knows? The southern portion of the tropical latitudes with wide ocean expanses and the absence of stations in a major part of central Africa is represented by relatively fewer grid cells and a sparser number of stations. As well, these latitude bands have a higher weight due to a larger area than the bands nearer the poles.
Either way, a little quality control would be a good first step toward producing some confidence in the whole process.
[Update December 28, 2009: The source that I used for Ono-I-Lau was http://en.wikipedia.org/wiki/Ono-i-Lau . I did not notice that text on that page stated
<blockquote>Ono-i-Lau is a volcanic and coral island in Fiji’s Lau archipelago.
One of the southernmost of the Lau Islands, it is located at 20.80° South and 178.75° East, and occupies an area of 7.9 square kilometers. It has a maximum elevation of 113 meters. It is 90 kilometers southsouthwest of Vatoa, the nearest island.</blockquote>
while the upper right corner on the same page gave the coordinates as 20°39′S, 178°44′W. I went with the text (confirmation bias? 😉 ).
Checking later, http://www.maplandia.com/fiji/airports/ono-i-lau-airport/ gives the information (and a Google map) as
airport name: Ono I Lau Airport
geographical location: 20° 38′ 59″ South, 178° 41′ 59″ West
so that Met appears to be correct on that one.]