On the Value of River Network Information in Regional Frequency Analysis

• DOI: 10.1175/jhm-d-20-0053.1
• Publication Date: 2020-11-25T18:17:24Z
• Authors: Kichul Jung; Taha B. M. J. Ouarda; Prashanth R. Marpu
• Keywords: Quantile; Jackknife resampling

Abstract Regional frequency analysis (RFA) is widely used in the design of hydraulic structures at locations where streamflow records are not available. RFA estimates depend on precise delineation homogenous regions for accurate information transfer. This study proposes new physiographical variables based river network features and tests their potential to improve accuracy hydrological feature estimates. Information about types both definition estimation process. Data from 105 basins arid semiarid United States were our analysis. Artificial neural ensemble models canonical correlation produce flood quantile estimates, which validated through tenfold cross jackknife validations. We conducted model performance statistical indices, such as Nash–Sutcliffe efficiency, root-mean-square error, relative mean absolute bias. Among various combinations variables, a with 10 produced best performance. Further, 49, 36, 20 networks classified dendritic, pinnate, trellis networks, respectively. The classification appeared provide superior compared without classification. results indicated that including proposed combination could types. finding has considerable implications structure design.