Stochastic generation of daily precipitation considering diverse model complexity and climates

Abstract

Stochastic models are powerful for precipitation generation; however, their performance toward diverse climate conditions should be fully addressed before application. With data from 191 stations in China, this study investigated the applicability of some commonly used single-site precipitation occurrence and amount models. The data represent diverse climate conditions in terms of mean monthly wet days of 1–21.5 days and mean annual precipitation amount of 15.5–2631.3 mm. The stochastic models represent multiple model complexity including four occurrence models (the first-, second-, and third-order Markov Chain and semi-empirical distribution) and five amount models (the 2-parameter Gamma and Weibull distribution, the 3-parameter skewed normal and mixed exponential distribution, and 21-parameter semi-empirical distribution). Overall, the models do not perform better with increasing model complexity, especially for occurrence models. The first-order Markov chain and the mixed exponential distribution outperform the other models for occurrence and amount simulation, respectively. However, model performance varies with climate conditions. The semi-empirical distribution and the mixed exponential distribution are respectively recommended for occurrence and amount generation in dry climates, while the first-order Markov chain for occurrence simulation and the skewed normal distribution for amount generation are better for wet climate conditions. These results provide useful information for stochastic precipitation generation in most regions of the world.