Future multivariate weather generation by combining Bartlett-Lewis and vine copula models

Abstract

The assessment of future extremes is hindered by the lack of long time series. Weather generators can alleviate this problem, but easily become more complex when generating multiple variables. In this study, a weather generator combining Bartlett-Lewis models and vine copulas is presented. The combination of these models allows for the stochastic and physically coherent generation of longer time series with statistics similar to those of the time series used as input. This model chain has already been assessed on the basis of historical observations, but never on the basis of future simulations. However, the model chain could have practical value for extending climate simulations, which should be investigated. Combining recent versions of the Bartlett-Lewis model (for the generation of precipitation) and vine copulas (for the generation of temperature and evaporation), the model was applied for two time series of historical observations and one time series simulated by the RCA4 RCM for the years 2071-2100. For the future simulations, the weather generator performed comparably as for the historical observations for the statistical moments and the correlation. The results for the extremes were more ambiguous, but still provided valuable information. The adequate performance for the statistical moments and the correlation, combined with the continuous development of both Bartlett-Lewis models and vine copulas, indicates that the weather generator might be of use for the characterization of extreme events under climate change.