Bayesian change-point analysis of heat spell occurrences in Montreal, Canada

Positive/upward shifts in the rate of occurrence of heat spells can considerably impact socioeconomic sectors. Particularly, populous urban areas and centers of regional socioeconomic activities are more vulnerable to the enhanced activity of heat spells. In this study, 24 time series of annual counts of summer-season (June–August) heat spells are derived from homogenized records of daily minimum and maximum temperatures (i.e. Tmin and Tmax) observed at McTavish station, located in the center of Montreal (Canada), over the period 1896–1991. Twelve of these time series, which fulfill the assumption of the Poisson process for heat spell occurrences, are examined for abrupt changes in the rate of occurrences using hierarchical Bayesian change-point approach. In these analyses, a heat spell is defined as an extreme climate event with Tmin and Tmax simultaneously above selected thresholds and a duration ≥ 1-day. The results of the Bayesian change-point analyses suggest structural inhomogeneities within the heat spell observations, i.e. the results do not support abrupt changes for all time series of annual counts of heat spells; this may not have been possible to detect by studying heat spells defined on the basis of just a single combination of Tmin and Tmax thresholds. Furthermore, the overall results of the Bayesian change-point analyses and those of commonly employed nonparametric trend detection and estimation techniques, when applied to change-point free smaller samples, suggest that there is inadequate evidence in favor of increased activity of heat spells in Montreal during the third last and second last decades (i.e. 1970s and 1980s) of the 20th century, which are the most recent decades of the observation period analyzed. Copyright © 2006 Royal Meteorological Society