Skip to main navigation Skip to search Skip to main content

Ensemble artificial neural network and generalized additive model for data-scarce regional frequency analysis in design flood estimation

  • Konkuk University
  • Korea Environment Institute

Research output: Contribution to journal › Article › peer-review

Abstract

Designing reliable hydraulic structures is crucial to mitigate the increasing risk of floods. However, small streams are left ungauged or poorly gauged owing to insufficient data and require indirect estimation methods. This study aimed to examine the application of regional frequency analysis (RFA) in the context of data-scarce conditions prevalent in small streams in South Korea. Hydrological processes are inherently nonlinear, requiring models that can effectively capture these nonlinear relationships. Thus, we used an ensemble artificial neural network (EANN) and a generalized additive model (GAM) and incorporated canonical correlation analysis (CCA) to enhance the model performance. Moreover, when using predictive models, it is important to identify the most appropriate and applicable variables from all available variables. Therefore, we also conducted the Morris sensitivity analysis to evaluate the input factors and contribute to the development of an RFA model with improved efficiency, thereby also addressing the existing scarcity of sensitivity analyses. Design flood estimates were derived for 67 watersheds in South Korea. Both EANN and GAM demonstrated satisfactory model performance, achieving high Nash–Sutcliffe efficiency values after CCA. The CCA-GAM outperformed the EANN by exhibiting a lower error and bias. Furthermore, the Morris sensitivity analysis revealed that the river basin area was the most influential variable in estimating design floods and that the mean river basin slope had the least influence. These results indicate that reliable flood discharge estimations can be achieved for ungauged basins using limited observational data through model training using observed regional data.

Original languageEnglish
Article number134886
JournalJournal of Hydrology
Volume667
DOIs
StatePublished - Mar 2026

Keywords

  • 68T07
  • 86A05
  • Design flood estimation
  • Ensemble artificial neural network
  • Generalized additive model
  • Sensitivity analysis

Fingerprint

Dive into the research topics of 'Ensemble artificial neural network and generalized additive model for data-scarce regional frequency analysis in design flood estimation'. Together they form a unique fingerprint.

Cite this