Main content area

A case study for ANN-based rainfall–runoff model considering antecedent soil moisture conditions in Imha Dam watershed, Korea

Kang, Boosik, Ku, Young Hun, Kim, Young Do
Environmental earth sciences 2015 v.74 no.2 pp. 1261-1272
case studies, data collection, decision making, engineering, flood control, hydrologic models, neural networks, planning, prediction, soil water, spatial data, storage, watersheds, Korean Peninsula
Prediction of hydrological water balances in dam reservoir water storage on long-term continuous rainfall–runoff estimation is highly important for accurate operational decision making related to sustainable water management planning, flood control and supply planning, etc. The physically based engineering model has been used conventionally, but requires a vast amount of hydrological and geographical data set for model construction. However, the data-driven model requires relatively low computational burden and shows reasonable accuracy depending on the feasible design of the predictor set. In this study, the dam inflow prediction using physically based model and data-driven model was compared. The antecedent soil moisture conditions were utilized effectively for training artificial neural network of continuous reservoir inflow modeling. The results indicated that the data-driven and physically based models had R ² values of 0.61 and 0.68, and Nash–Sutcliffe efficiency values of 0.60 and 0.66, respectively, which shows reasonable performance. Even though the dam inflow prediction results using the physically based model showed a relative superiority than the data-driven model, the difference was not high enough to diminish the advantages of the data-driven model. The data-driven model could be an effective alternative in areas of limited availability in hydrologic observations and geospatial data sets.