| description abstract | In recent years, the stationary assumption of hydroclimatic variables as well as their temporal and spatial dependence structure have been challenged due to anthropogenic and climate changes. This study developed a stepwise and dynamic C-vine copula–based conditional model (SDCVC) to incorporate the nonstationarity into a monthly streamflow prediction, which depicted the predictor–predictand association on the basis of monthly streamflow and rainfall series of upstream and downstream stations in the Yangtze River basin of China. The model consists of (1) nonstationary modeling of the margins and nonstationary temporal and spatial dependence structure by the generalized additive models with location, scale, and shape (GAMLSS) and C-vine copula incorporating climate-related indexes as covariates during training time frames; and (2) a four-dimensional C-vine copula–based conditional quantile function to generate the simulated series during validation time frames. Three kinds of nonstationary models corresponding to different degrees were investigated to show the impact of nonstationarity on the streamflow forecasting. The proposed SDCVC model considering highest degree of nonstationarity outperformed the other two nonstationary models in terms of the performance metrics, because the SDCVC model not only described the dynamic change of time-varying connections between parameter and the large-scale climate forcings, but the stepwise strategy, by selecting the optimum time horizon, helped increase forecasting accuracy. Furthermore, the SDCVC model is superior, to some extent, to classical data-driven approaches [support vector regression (SVR) and adaptive-network-based fuzzy inference system (ANFIS)] in terms of the performance metrics. | |
