A hybrid bayesian vine model for water level prediction

Abstract

Skillful hydrological prediction is highly valuable for optimizing water resources operation and planning. This study proposes a hybrid Bayesian vine copula (BVC) model for daily and monthly water level prediction. To achieve this goal, we first build a conditional vine copula (VC) prediction framework by utilizing the flexible vine copula as the baseline technique. Then, a Bayesian-based model averaging scheme was employed to derive better predictions by combining all of the candidate conditional VC models and meanwhile, quantify the uncertainty associated with these VC structures. The proposed BVC model was tested to predict the daily and monthly water levels at two stations in South China. The results demonstrate that the hybrid BVC approach can yield more reliable forecasts than the conditional VC model and traditional deterministic models such as the adaptive neuro-fuzzy inference system, according to the evaluation metrics.