Sunday, 24 September 2017

Warped weather extremes

It usual to focus on weather which is the obvious extreme of the variable being considered.  By this I mean the highest maximum temperature or the lowest minimum temperature.  However today I have looked at variations on this theme with highest minimum temperature and lowest maximum temperature.

The catalyst for this was the overnight minimum this morning (24 September) being 12.4o C, which is the highest minimum temperature recorded for Carwoola in September since records started in 1993.  (There are of course issues about day-minima vs overnight minima, and it is possible the temperature will drop below 12,4 by midnight thus reducing my recorded minimum for the day, but it is still notable.  I shall return to this point below)  This follows the maximum on 23 September of 29.2o C which was definitely a new high temperature (or maximum maximum) for the month.  So at least there is some consistency.

This chart shows the maximum minimum for each month for all years over the period and for the last two years.  (I have assumed that things will return to more normal levels after today.)  I have added an arrow to point to the September 2017 value
 I restricted that chart to the long term (ie since 1993) average and the last two years.  I did also look at a scatterplot of maximum minimum (maximin) x year to see if there was a trend (anything that might give me an excuse to rubbish soon-to-be-former Senator Malcolm Roberts must be explored).  Despite the 4 warmest years being 2017 and 2011-13 the value of r2 was pretty miserable at 0.2614, so there is no significant trend.

Having looked at warping in that direction I thought I should in all fairness check the opposite, minimum maximum (minimax), temperature for September.
For both the last two years the line for minimax is well above the long term average.  I would like to say "Cop that, soon-to-be-former Senator Malcolm Roberts!" but unfortunately there is no significant trend in those data and of the last three years 2015 ranks 12th highest out of 26 readings, 2016 ranks 13th and 2017 is 21st. The value of r2 was 0.0009, which is about the lowest level of that statistic I have seen!  So no trend (which doesn't mean that the soon-to-be-former Senator Malcolm Roberts is not full of it).

In undertaking this unnatural analysis I did come across a few errors in the data which I have fixed up, so there might be some small changes in series compared to what I have used in the past.  This doesn't really prove Trewin's Law (any interesting statistic is probably a processing error) since some of the other interesting values turned out to be correct.

On the subject of Trewin's Law it is possible that the starting point for this post is a great example thereof, in that the overnight minimum might not be the minimum for the day (as explained earlier).  I have tried to look at overnight minima by only looking at the time period earlier than 9am.  For various reasons I only have 2 Septembers (2014 and 2015) with times recorded as well as dates and temperatures.  In 2014 the minimum daily temperature recorded was 7.2o C while the overnight low was 11.4o C.  Thus it is possible that there was a high, but unrecorded, overnight minimum in years such as 2004 and 2010, both of which recorded a daily maximum of 27o C.

Being optimistic, the previous recorder has advised that his minima were calculated in EXCEL for a day starting at 7:00am so it is likely that high overnight minima were captured for the day following.

He does also note "...had some curious minima not necessarily during the night. I think one day last month when it snowed we had a minimum of 1.2 degrees at 11.30.am."  The approach used interested me enough that I decided to have a crack at replicating it for August 2017.  I was able to do so, and compared results for the three processes: minimum over a calendar day; minimum between midnight and 9am; and minimum for a day starting at 7am.  My conclusions are:

  1. For many days they give the same value;
  2. Where differences arise it is often because the minimum occurs at 6am or 7am (a particular problem in August where that is the time of daybreak).
  3. On occasions there is a 47 hour difference between a minimum at 7am one day and the next minimum occurring at 6am in two days time.
  4. It is not possible without a lot of manual mucking about to automatically store the measures other than the calendar day approach.  It is particularly hard to incorporate the day starting at a specific time other than midnight.

My conclusion, giving great weight to points 1 and 4 above, is to stick to using the calendar day approach as my basic measure and to do ad-hoc analyses where needed to comment on things like a very warm night between two much cooler ones.  In discussing this with Frances I realised that 

  • For answering questions like "Was the month colder than normal?" using measures such as minimum temperature in the month or average daily minimum it probably doesn't matter greatly what measure is used: the shape of the graph will be pretty similar even though a few values might wobble up and down a bit.
  • For more complex questions such as "Is this extreme temperature a record?" a more careful approach is needed perhaps adopting a different set of metadata.
  • Most importantly I continue to regard the length of my data series too short to make any big statements about climate change!  (Other than, of course that soon-to-be-former Senator Malcolm Roberts is wrong.)

No comments: