Past IEM Features tagged: bias

List all tags


Time of Observation Bias

10 Aug 2015 05:45 AM
The long term climate record is mostly daily observations that do not represent calendar day summaries, but values representing 24 hour periods ending sometime during the mid morning or mid afternoon. While this time may not seem significant, there are very important climate implications. The featured chart displays the result of computing 24 hour summaries based on the hourly Des Moines temperature reports. If the values computed when the 24 hour period is split at midnight (actual calendar day summaries) is considered the 'truth', the bias can then be computed for the other 23 possible hours of the day that the temperature record can be split over. The result of this plot is known as a time of observation bias, with the largest biases for high temperature appearing in the mid afternoon. How can this be? Consider a hypothetical Sunday with each hourly temperature of 95 degrees, then the following day has each hourly temperature of 75 degrees. If you compute the 24 hour totals at midnight, the high on Sunday is 95 and on Monday is 75. But if you take the observation at 4 PM, the high reported on Sunday would again be 95 and the value reported at 4 PM on Monday would also be 95! The 24 hour period ending at 4 PM on Monday would contain those 8 hours of 95 degree reports from Sunday evening, thus the high is 95!

Voting:
Good: 15
Bad: 6
Abstain: 6

Tags:   climate   bias