Multi-site, multivariate weather generator using maximum entropy bootstrap

Weather generators are increasingly becoming viable alternate models to assess the effects of future climate change scenarios on water resources systems. In this study, a new multisite, multivariate maximum entropy bootstrap weather generator (MEBWG) is proposed for generating daily weather variables, which has the ability to mimic both, spatial and temporal dependence structure in addition to other historical statistics. The maximum entropy bootstrap (MEB) involves two main steps: (1) random sampling from the empirical cumulative distribution function with endpoints selected to allow limited extrapolation and (2) reordering of the random series to respect the rank ordering of the original time series (temporal dependence structure). To capture the multi-collinear structure between the weather variables and between the sites, we combine orthogonal linear transformation with MEB. Daily weather data, which include precipitation, maximum temperature and minimum temperature from 27 years of record from the Upper Thames River Basin in Ontario, Canada, are used to analyze the ability of MEBWG based weather generator. Results indicate that the statistics from the synthetic replicates were not significantly different from the observed data and the model is able to preserve the 27 CLIMDEX indices very well. The MEBWG model shows better performance in terms of extrapolation and computational efficiency when compared to multisite, multivariate K-nearest neighbour model.