Temperature Stats, 24z to 24z vs 09z to 09z

Mothman

Those of us with AWS (automatic weather stations) will see that daily stats are generally calculated on a midnight to midnight basis. This is not the standard practice with manual observations such as at the climatological stations of which I host one. These are taken once daily at 0900gmt.

The max and min temperatures are read and thermometers reset. The min reading is applied to day of observation and the max is "thrown back" to previous day.

For example, this morning, the min was -1.5C, and this is the min for today, 23rd, the max was 9.9C and this is the max for yesterday, the 22nd. The temperature at 9am was 1.7C, so when thermometers were reset, they both read 1.7C. This means that even if the temperature stayed above 5C tonight, then the min reading at 09gmt tomorrow will be 1.7C and that will go down as min for 24th.

So now you can see the limitations of the once a day reading. AWS are certainly accurate for calculating true mean temperatures, but for reasons for comparing with past then the standards that have applied for 100s of years have to continue.

But as Arctictree pointed out the AWS can come up with misleading stats as well.
A cold night where subzero temps happen before and after midnight, gets recorded on 2 days, whereas generally with the 09 to 09 obs, it will be one day. But anomalies occur with the 09 to 09 system as well, particularly at this time of year when sunrise is at its latest. The anomaly is when the 09z temp is sub zero. Even if the temp the following night stays above zero, the next day will have a sub zero min. This anomaly does occur near as often once we get into February as the sun is so much higher and has usually raised the temp above zero by 9am. This time of year, the sun is only just above horizon and it is nearly 10am before it makes an impact on the temperature.

I have AWS vs manual stats for days of air frost going back to 2000. The AWS being midnight to midnight, manual 09-09z. 2007 to 23rd Dec

The AWS usually records a few more days. Note how few air frosts there were in 2002.

I also have year mean temp stats for same period. 2007 to 23rd Dec
AWS Mean is the true average.
Manual, is average of mean max and mean min daily temps
AWS max min is calculated same as manual, but in this case, the max and min figures are from midnight to midnight days.
Apart from 2002 manual, there is relative consistency across the figures. It can be seen that the 09-09z figures reflect the true mean quite closely apart from 2002. I'll have to look closely at the 2002 figures closely and see if I can see if they are correct. part of the explanation may be in a not ideal placing of the AWS for part of the year.


It is possible to set up an AWS to record 09-09gmt days. This would solve the last column. It is done simply by setting the time on the station to midnight when it is 09gmt. but then the times of the all the max, mins etc would be wrong. Perhaps software like Weather Display allows the option of an off set day, but retaining real time?

Supercell

Some interesting stats there Mothman.
Your annual means are rising in almost perfect linear fashion, doesn't auger well for future winter snow if that trend continues (assuming winter month averages are trending the same way).
The frost stats for 2002 are really surprising, hard to imagine so few frosts all year, especially given the amount we have had this year in December alone.