Development of a non-parametric stationary synthetic rainfall generator for use in hourly water resource simulations

Abstract

This paper presents a new non-parametric, synthetic rainfall generator for use in hourly water resource simulations. Historic continuous precipitation time series are discretized into sequences of dry and wet events separated by an inter-event dry period at least equal to four hours. A first-order Markov Chain model is then used to generate synthetic sequences of alternating wet and dry events. Sequential events in the synthetic series are selected based on couplings of historic wet and dry events, using nearest neighbor and moving window methods. The new generator is used to generate synthetic sequences of rainfall for New York (NY), Syracuse (NY), and Miami (FL) using over 50 years of observations. Monthly precipitation differences (e.g., seasonality) are well represented in the synthetic series generated for all three cities. The synthetic New York results are also shown to reproduce realistic event sequences proved by a deep event-based analysis.