Using NCDC’s Extremes Records website: http://www.ncdc.noaa...tremes/records/, I filtered out any tied records and only focused on new records. Additionally, I excluded all labeled POR sites that did not have a complete POR*.
*NOTE: These records are based on the historical daily observations archived in NCDC's
# Cooperative Summary of the Day data set and preliminary reports from Cooperative Observers
# and First Order National Weather Service stations, and as such are subject to change.
# The Period of Record (POR) represents the number of years with a minimum of 50% data
# completeness. All stations have a Period of Record of at least 30 years.
# When the 6-digit COOP ID does not exist for a station the 5-digit WBAN ID is used.
# Units are °F.
Here is what I uncovered:
Based on 1576 new Tmax records from 1 June – 18 August 2011, the frequency of most new records occurred 38.3% of the time as a one degree record:
New Record by degree F (see attachment)
There was one case in which the record was broken by 12°F:
DATE REC PREV PREV DATE STN ID LAT LON TYPE POR ST COUNTY NAME
7/12/2011 117 105 7/12/1998 418236 35.22 -100.25 COOP 45 TX WHEELER SHAMROCK 2
I assumed an expected re-occurrence of a new broken record based on the following:
If a weather station has 100 years of homogeneous observations, that is, no change to the station's location, instrumentation, or surroundings, a daily maximum or minimum record temperature could be expected to be broken on average about seven days during that year. Natural variation will cause this number to vary each year but should average about seven days over several years. Or for a given day, it would take 12,367 years for 10 new records to occur. During the first 10 years of record keeping approximately three new records will have occurred. However as the station's observations increase over time, it will become harder for new records to be established. This of course assumes that climate variation occurs as a normal probability distribution:
1 - (1 - 2/N)^n
The n is the number of weather elements under consideration and N is the number of years of available weather data. More records will be established quicker if more weather elements are being considered. Ref: Ralph Boas, Weatherwise (April 1980), pg. 60-64.
Essentially, if a record was broken earlier than this relationship, it would occur earlier than half the POR. For example, a station with a 20 year POR would expect a new record in 10 years (based on the above equation). If it occurred in 2010 and then again in 2011, that would be considered early. If it occurred last in 2001, then it would be on time, and if it occurred last in 2000, it would be considered late.
Of the 1576 new Tmax records, 497 occurred later than expected, 40 occurred on time (again, as expected by the above equation), and 1039 occurred early. The ~2:1 ratio of early record breaking over later record breaking is an interest relationship when studying climate change. I suspect that extreme droughts can easily explain this ratio (i.e., less clouds).
To test this theory, I took the same period (1 June – 18 August) in 2007 when there was no drought in Texas and the results are quite interesting:
There was a total of only 49 new broken Tmax records; 15 occurred late, 1 on time, and 33 early. Again the 2:1 ratio occurred but with much fewer broken records. Obviously drought increases record heat but not necessarily in an unexpected ratio (was the sample too small to be statistically significant?).
It would be interest to see if this ratio holds for other seasons and years.
Here's more from a recent Texas Study: