There is often a large debate over whether we should keep updating the 30-year reference period every 10 years (the World Meteorological Organisation does it every 30 years, the Met Office traditionally does it every 10 years but has partially held onto 1961-90 in the recent update).

My view is that it depends on what analysis we're doing and that there is often plenty of room for argument. When we're comparing current weather (or the past month or year's weather) with the "average", i.e. what the public are used to as being "average" conditions, I think repeatedly shifting the reference period forward is a good idea, because it is the most representative of the average that they're used to. For instance if we compare January 2010 with the 1961-90 average we're using a reference period that ends before a fair number of people were even born!

But I think when we're doing an analysis of long-term climatology, it is often better to stick with one reference period. For the period 1993-2009 my Cleadon weather records are always compared with the estimated 1971-2000 averages, and I don't think I'll be updating to 1981-2010 anytime soon. It would have the effect of masking any long-term changes in Cleadon's climate. In fact I think a much longer reference period (say 1951-2000 or even 1901-2000) would be most ideal for this kind of analysis, but such reference periods are harder to get data for than the 30-year means.

I suspect that these sort of considerations might be behind the Met Office's inconsistency in updating to 1971-2000, e.g. the monthly assessments are clearly in the former category, but one could argue that, for instance, the CET diagnostics are more the latter type and so are arguably better served by a reference period that is more representative of the longer-term (and 61-90 is more so than 71-00).