HadUKP precipitation, nothing to see here

Posted: July 7, 2015 by tchannon in Analysis, weather

This little work rather counters the headless chickens preparing for a French cooking pot.

Image

Figure 1, CRU/Hadley/Met Office precipitation series starting 1766, just one in the HadUKP series. Good news, there is nothing more than weather noise in any of the 11 region series. All bundled in this PDF. (2MB)
For number watchers, Jan 2014 came 11th wettest.

This note is on their web page

We are currently planning a project to merge the HadUKP series with the England and Wales Rainfall series described above. The outcome of this project will be a single historical rainfall series for the UK.
http://hadobs.metoffice.com/hadukp/

I suppose that makes sense yet neither series is IMO satisfactory on geography. The UK has a variety of weather regimes. Merging regimes has the effect of mixing evidence where average is not very useful. Is there a better solution?

Couple of days ago I was in passing reminded of the HadUKP data, took a shufty at the data files, looks easy. Apart from needing to change the way I was handling multiple series in one set, worked immediately, no core code change.

The intent is an honest look where the variation is moved into the z-score domain and without the perfectly normal variation during a year. Perhaps counter-intuitively this if anything enhances seasonal change.

In the rankings I include the actual figure.

High Rank Date Z Actual
1 1903.79 3.32831 218.1
2 1947.21 3.1805 177.5
3 1912.63 2.99229 192.9

Here, March is normally a drier time of year, not in 1947.

As you can see a side effect is disputing official rankings where normal seasonal change is not taken into account… who cares if it rains during early winter, that’s what it does, now how exceptional is it? A second point in this is breaking the meteorological poor practice of hard edging seasons, as the real world of course does. Timing varies.

Original article showing these plots, has references, here
Various Talkshop articles have appeared eg. here and the technique works on rain, sun and temperature monthly.

Articles like this tend to be blog traffic killers, as-if there is any way to show ordinary, null, and have any real interest. The blog can stand some of this.

Post by Tim

Comments
  1. A great article. I’d just recently said that one of the very few things that had been changing was rain – and then I remembered last time I looked there had been no change.

    I guess from this data that it was “no change”.

  2. tchannon says:

    My worry SS is getting this stuff wrong, misleading people.

    A very difficult area with no real solution is the CDF stuff I headline. A density computation is heuristicso I am not even going to try automating that. (the PDF here is click, answer Y to delete all files, and wait)

    Awkward tails are usual with rain leaving a fair case for saying I am being silly. An instance is in there (and in the 1910 series), an extreme almost off scale (*). I don’t have a problem with this, the IMO inherent 1/f nature of rain make that likely, occasionally. In a way it is nice that the supposed averaging of many independent 1/f sources tends to Gaussian, what I am seeing.

    I can think of a few monstrous rain events I’ve experienced, tend to rather local.

    * East Scotland 1995. One in 1910 series, East Anglia. Both the drier east coast.

  3. Kevin Hearle says:

    Achieve all the data before they homogenise and ruin any ability to analyse what really happened.

  4. tchannon says:

    There will be more I could do to the data and how the results are presented but the law of diminishing returns has kicks in. Is enough added to warrant my time?

    There is remaining variance variation, an artefact of the annual variation, weather is more variable at some times of the year. Monthly has approximately the same shape as the annual variation. If this was adjusted it would have a small effect, inevitably changing rankings slightly. I’ve concluded this is of dubious worth and especially when I am trying to keep the work simple.

    Additionally the underlying data violates Nyquist and Shannon; the data contains artefacts of poor mathematics which make month data wrong.

    Changes to the presentation is probably much easier to justify but also takes a lot of time to do.. I did start to put together a cover page, particularly an area map until I realised it isn’t a great deal of help: the data site page map does not cover all the data regions, some are combinations. For example, some data regions include Wales; Wales is not marked.

    I haven’t included the value of the last datapoint as a number, started to do this then it occurred to me this is number chasing. Good enough can be seen from the new detail thumb but it the value is a new extreme it will be in the table of values. If someone really needs to know they only have to ask for the file of rankings. (it’s saved to disk automatically)

  5. A. Ames says:

    tchannon:

    Sorry for the delay in responding. You and followers might be interested in the following
    of relevance to rainfall analysis:

    http://wattsupwiththat.com/2015/07/01/a-way-to-calculate-effective-n/
    http://nvlpubs.nist.gov/nistpubs/jres/099/jresv99n4p377_A1b.pdf

    The NIST article shows that different regions in the States have different Hurst coefficients,
    so merging data series across the UK likely destroys information useful for prediction.

    The question of what a statistically significant difference might look like is a fascinating one.