Detection of changes in precipitation and runoff over eastern Canada and US using a Bayesian approach

Inference about the existence and characteristics of changes in mean level of hydrometeorological sequences that may be generated by climatic variabil ity is an important step before developing management rules in water resour ces systems. This paper presents a Bayesian approach, based on a single shi fting model, which can be used to study a change in the mean level of a set of independent normal random variables. Two different problems are conside red: the first is the detection of a change, while the second is the estima tion of the change-point and its amplitude under the assumption that a chan ge has occurred. This method is applied to precipitation and runoff data se ries over eastern Canada and U.S. during the twentieth century. The main re sults show an increase in the late sixties in the Eastern North American pr ecipitation. This supports conclusions drawn from a number of studies which identified the late sixties to early seventies as a period of possible cha nge.