Tag Archives: autocorrelation

A measure of weather variability

Double rainbow with Alexander band. Photo: en.wikipedia.org

Double rainbow with Alexander band. Photo: en.wikipedia.org

This morning, the weather was gloriously sunny with barely a cloud in sight, and the temperature around 12 degrees. Twenty five minutes later, it was cold and blustery. One of us commented that the weather in Canberra seemed more variable than it is in Perth. “More like Melbourne weather,” we both agreed, which set us thinking about a measure of weather variability. We could not find one published anywhere, but some sort of lagged autocorrelation would appear to do the trick. If the serial autocorrelation of temperature, lagged by several days were high in one location, and close to zero in another, then one would conclude that the first location has stable weather, and the second has more variable weather. In Perth, for instance, the best first-guess prediction of the weather tomorrow appears to be, “Same as today,” whereas in Melbourne one might be considered foolish not to take an umbrella with you, no matter how good the weather appeared. Presumably Perth has a high autocorrelation in temperature and rainfall even when the measurements are lagged by several days. The same might not be true in Melbourne.

The idea certainly seems worth pursuing, either as an academic exercise, or even as a method of describing weather patterns and predictability to the unwary tourist. And the autocorrelation could be calculated in a cannonical form, taking into account multi-variate data (temperature, wind speed, rainfall, etc.) instead of focussing on each variable independently.

Contributors: Angela O’Brien-Malone, Mark R. Diamond