Main content area

An effective storage function model for an urban watershed in terms of hydrograph reproducibility and Akaike information criterion

Padiyedath Gopalan, Saritha, Kawamura, Akira, Takasaki, Tadakatsu, Amaguchi, Hideo, Azhikodan, Gubash
Journal of hydrology 2018 v.563 pp. 657-668
floods, hydrograph, hydrologic models, model validation, prediction, risk, runoff, system optimization, uncertainty, urbanization, watersheds, Japan
Rapid urbanization is considered to be an important factor that contributes to flood risk. Therefore, flood prediction in urban watersheds using appropriate runoff models is essential to avoid the harmful effects of floods. There are various storage function (SF) models such as Kimura, Prasad, Hoshi, and urban storage function (USF) models that have been widely used in different parts of the world as rainfall-runoff models in which the USF model was recently developed in Japan for the specific application in urban watersheds. However, the identification of an appropriate model remains challenging in the field of hydrology. Therefore, this study aims to identify an effective SF model for an urban watershed in terms of hydrograph reproducibility and from an Akaike information criterion (AIC) perspective. The SCE-UA global optimization method was used for the parameter optimization of each model with root mean square error (RMSE) as the objective function. The reproducibility of the hydrograph was evaluated using the performance evaluation criteria of RMSE, Nash-Sutcliffe efficiency (NSE), and other error functions of peak, volume, time to peak, lag time, and runoff coefficient. The results revealed that the higher values of NSE coupled with the lower values of RMSE and other error functions indicated that the hydrograph reproducibility of USF is the highest among the SF models. Furthermore, AIC and Akaike weight (AW) were used to identify the most effective model among all those based on the information criteria perspective. The USF model received the lowest AIC score and the highest AW during most of the events, which indicates that it is the most parsimonious model compared to the other SF models. Moreover, uncertainty characterization of the SF model parameters was also conducted to analyze the effect of each parameter on model performance.