R-statistic based predictor variables selection and vine structure determination approach for stochastic streamflow generation considering temporal and spatial dependence

Abstract

Stochastic streamflow generation is crucial for water resources planning and management as well as water conservancy project design and operation. This study proposes an accurate, reliable and parsimonious approach for stochastic streamflow generation considering temporal and spatial dependence on the basis of regular vine copula model. The emphasis is on advancing an R-statistic based strategy of vine structure determination that divide the vine copula model construction into two independent parts and avoid continuous accumulation of uncertainty in the traditional Kendall's tau based method. Two study regions (the Upper Colorado River basin and Middle Yangtze River basin) with diverse hydrology regime and available data length are selected as case studies to showcase the performance of the proposed approach in practice. The results indicate better performance than two existing models in terms of streamflow estimation, and demonstrate that stochastic simulation series can preserve distribution and statistical characteristics of observed records. R-vine copula model constructed by the proposed approach is confirmed to possess low sensitivity to the number of predictor variables as well as good adaptability and robustness to streamflow series with diverse characteristics and abundances. The enhanced capability and performance stem from the accurate identification of predictor variables and characterization of complex and diverse dependence structures among different streamflow series, on the basis of a comprehensive and precise dependence measure, R-statistic.