Synthetic generation of hydrologic time series based on nonparametric random generation

Abstract

Synthetic hydrologic time series can be used to quantify the uncertainty of a water resources system. Conventional parametric models, such as autoregressive moving average or Markovian models, assume that the variable under consideration is Gaussian. This assumption, however, is a shortcoming of parametric models and motivates the development of nonparametric approaches. Nonparametric models based on a kernel function have an innate low-order structure and are restricted to highly persistent variables. This study presented a seminonparametric (SNP) model that takes advantage of both parametric and nonparametric models to generate monthly precipitation and temperature in the Conchos River Basin in Mexico. By adopting a consistent and robust scheme from the Markovian model and a nonparametric mechanism to generate a distribution-free random, component, the SNP model reliably reproduced sample properties such as mean, variance, correlation, and multimodality in the probability density function.