Heat waves, generally defined as prolonged periods of extreme temperatures, have hit several areas of the world and have been observed more frequently in recent decades. Several proposals appeared in the literature for classifying heat waves, mainly related to expert-based and area-specific fixed thresholds or quantile-based approaches. Summer temperature patterns exhibit different stochastic processes, which can be determined through a latent variable that describes the membership of each observation to a normal or a high-temperature regime. We accommodate these characteristics by exploiting a Bayesian Markov-switching additive model for tail behavior modeling and aiming at the probabilistic classification of heat waves. We illustrate the proposal by analyzing the maximum daily temperatures in four locations of the Italian region Friuli Venezia Giulia. We use the model to describe past behavior and produce scenario-based projections.
