Stochastic Analysis of Cold Spells

Abstract

The deterministic and stochastic structures of daily minimum air temperature time series are investigated in four historic records of at least 60 yr in southern Saskatchewan, Canada. Identification and removal of deterministic trends and periodicities is performed by linear regression and Fourier analysis. The residual series contains the stochastic structure of deviations from expected daily minimum temperatures and is analyzed in terms of abnormally cold periods below some temperature of interest?a partial duration series. The partial series defines a number of variables describing each spell: their number, timing and magnitude. These are modeled by probability distributions which provides good agreement with the observed data. The capability of the model to predict the risks of cold spells of particular characteristics is demonstrated and the use of a flexible definition of cold spells is illustrated.