A two-stage multisite and multivariate weather generator

Abstract

The spatial structure of climatic variables synthesized by a weather generator has considerable impact on the modeling of hydrological variability; however, in most cases, it needs computationally intensive work to reproduce multisite and/or multivariate cor-relations. This work proposed a two-stage weather generator (TSWG) to preserve intersite and intervariable correlations of daily precip-itation, maximum and minimum temperatures. The first stage generates daily precipitation and temperature for each site and for each variable with, but not limited to, the Richardson-type approach. The second stage rebuilds the multisite multivariate correlation using a distribution-free shuffle procedure. The TSWG was applied to a network of 15 stations in the Jing River catchment (Northwest China). It reproduced the statistical parameters and multisite and multivariate correlations well. Furthermore, indirect validation by hydrological modeling showed TSWG outputs could be used satisfactorily for simulating streamflow variability. As a distribution-free method, the correlation reconstruction method can be applied to variables with different probability distributions. The TSWG can efficiently reconstruct the correlation with one optimization for all stations and all variables, which is superior to most current methods operated once for one station pair and one variable. The TSWG provides an option for improved multisite and multivariate weather generation.