Assessing the unexpectedness of a very large observed rainfall event in the metropolitan region of Belo Horizonte, Brazil

Exceptionally large realizations of hydrometeorological variables summarize our knowledge on the underlying processes under extreme conditions. However, the formal probabilistic modeling of record-breaking precipitation and flooding events has received limited attention from researchers and practitioners, particularly with respect to estimation of the future behavior of the variable of interest once a large magnitude event is observed. This paper discusses the use of the two-parameter Rayleigh distribution for describing the dynamics of record-breaking precipitation events, as well as predicting the distributions of future realizations of the process, on the basis of the theory of records. The main motivation for this study was an unprecedented precipitation event which accumulated 186 mm in 3 h-almost twice the previous record-and caused severe damage in the city of Belo Horizonte, in the Brazilian state of Minas Gerais. Our results suggested that, before observing this large magnitude event, the evolution of record-breaking precipitation amounts is properly captured by the model. However, this exceptional event was severely underestimated during prediction-the model estimate was more than 80 mm smaller than the observed rainfall amount. On the other hand, the inclusion of this event in inference entailed a lack of fit of the model and strongly disrupted the dynamics of previous record events, which further highlighted the difficulties of dealing with such unexpected large observations of hydrometeorological variables even under theoretically sound mathematical frameworks.