Multi-Site High-Frequency Stochastic Precipitation Generator Using Censored Skew-Symmetric Distributions

Abstract

Due to improved measuring instruments, an accurate stochastic weather generator for high-frequency precipitation data is now possible. However, high-frequency precipitation data are more zero-inflated, skewed, and heavy-tailed than common (hourly or daily) precipitation data. Therefore, classical methods that either model precipitation occurrence independently of their intensity or assume that the precipitation follows a censored meta-Gaussian process may not be appropriate. In this work, we propose a new multi-site precipitation generator that uses a censored non-Gaussian vector autoregression model, where the random errors follow skew-symmetric distributions. The proposed stochastic precipitation generator not only drives both the occurrence and intensity of the rainfall events simultaneously using only a single model, but it also provides nice physical and statistical interpretations. We apply this stochastic generator to 30-second precipitation data obtained from a dense gauge network in Lausanne, Switzerland, and we show that our proposed model can provide as accurate predictions as the long short-term memory (LSTM) model but with uncertainties and more interpretable results.