Abstract

Stochastic weather generators are commonly used to simulate time series of daily weather, especially minimum (Tmin) and maximum (Tmax) temperature and amount of precipitation. Recently, generalized linear models (GLM) have been proposed as a convenient approach to fitting weather generators. One ... Show moreStochastic weather generators are commonly used to simulate time series of daily weather, especially minimum (Tmin) and maximum (Tmax) temperature and amount of precipitation. Recently, generalized linear models (GLM) have been proposed as a convenient approach to fitting weather generators. One limitation of weather generators is a marked tendency to underestimate the observed interannual variance in monthly, seasonal, or annual total precipitation and mean temperature, termed the ‘overdispersion’ phenomenon. In this study, aggregated statistics, consisting of seasonal total precipitation and mean Tmin and Tmax, are introduced as additional covariates into the GLM weather generator. With an appropriate degree of smoothing of these aggregated statistics, this approach is shown to virtually eliminate overdispersion when applied to 2 sites, Pergamino and Pilar, in the Argentine Pampas. The addition of these covariates does not distort the performance of the weather generator in other respects, such as annual cycles in the probability of precipitation and in the mean Tmin and Tmax. For seasonal total precipitation, the reduction in overdispersion is partially attributable to a corresponding reduction in the overdispersion of the frequency of precipitation occurrence, as well as to apparent temporal trends or ‘regime’ shifts. For seasonal mean Tmin and Tmax, the reduction in overdispersion is largely due to temporal trends on an interannual time scale. Show less