Weather generators (WG)that create synthetic daily weather data statistically similar to observed data involves precipitation simulation as critical component.Preserving statistical properties(e.g., spatial intermittence,low frequency variability)of precipitation is challenging.We propose a new framework for multisite precipitation generation consisting of three main components:a multi-site spatiotemporal field and corresponding temporal evolution based on empirical orthogonal function analysis (EOFA),principal component(PC)time series decomposed into intrinsic mode functions(IMFs)and subjected to Hilbert-Huang transform (HHT),and stochastic simulation(SS)achieved by assigning random phases for certain specific IMF.The proposed EOFA+HHT+SS model(EHS)is illustrated in a network of 12stations in Xiang River Basin,China,with MulGETS,a typical multisite Richardson type WG,as reference.The synthetic precipitation series were compared to records and evaluated with respect to basic statistics(mean and standard deviation of monthly precipitation),spatial dependence(spatial correlation and continuity ratio)and temporal dependence statistic(autocorrelation).EHS precipitation generator was found to have superior capability to conserve historical statistical properties of the observation,especially with respect to autocorrelation at various time lags and low frequency variability.EHS is an effective model for generating multi-site precipitation field and can be expected to generate plausible scenarios for impact study of climate change and climate variability.