Variation in temperatures

 I have recently changed my calculation of the long term averages in my assessment of the temperature anomaly from a series starting in January 1993 which included data from (or based on) BoM data at the Airport site and my Weather Station (WS), to a series starting in January 2019 solely from my WS.  This was done to avoid differences due to the change in site,

I knew there were differences in maximum temperatures due to the cooling effects of sea breezes on the BoM data and had made adjustments sing regression models to compensate for that.  However recently I had decided that there might also be less significant, but still regular differences in the minimum series.  Rather than fiddle around trying to adjust the minima as well - giving a rather artificial set of data from 1993 - 2018 - I decided that a 6 year series would give a "good enough" basis for noting if a month was generally warmer/colder than normal (which is my aim in calculating the anomaly).

In working on the September data I noticed that the average temperatures were all over the place, as shown by the dashed lines in this chart.

This made me wonder if:

• September was an unusually variable month; and/or
• If the variability was higher in the WS series than in the previous longer series.

So I calculated the average daily maximum and minimum temperatures for each month and the Standard Error (SE) of those averages.  I then calculated the Relative Standard Error by expressing the SE as a % of the average.  That gave the following chart.

There are several interesting points in the chart.

1. Minima are subject to much greater variation than maxima;
2. September is variable but not as variable as July/August;
3. The WS series for minima is less variable that the hybrid series (possibly to be expected as it is much shorter so less chance of outliers);
4. The two maximum series are very similar (suggesting that my adjustments were satisfactory); and
5. Overall, the differences in the maximum series suggest the recent years are warmer than the earlier years (but there isn't a long enough WS series to make much of this).

I don't see any need for change as a result of this investigation.  I will keep the original (longer) series but use the WS-only series for calculating the anomaly.  If I find a rainy period I might investigate adjusting the old minimum series as I did the maxima but it is a fair bit of effort.