A Network Approach for Delineating Homogeneous Regions in Regional Flood Frequency Analysis

Abstract

Regional flood frequency analysis forms the basis for ascertaining design thresholds for extreme flow events for the purpose of resource management and design of hydraulic structures, especially at ungauged or partially gauged basins. A crucial step in this analysis is transferring available information from gauged sites to ungauged sites, which is achieved through delineation of homogeneous regions encompassing multiple catchment locations, followed by the formulation of a flood estimation model. While this process has been accomplished through a range of statistical homogenization alternatives, the present study offers a new approach rooted in the theory of complex networks, offering considerable advantages over what is traditionally followed. Data from 202 sites in Australia representing catchments of varying geographic, climatic, and vegetation attributes are used to assess the alternative proposed. The results are examined via (1) direct comparison of the location and number of homogeneous neighbors from network theory with results using canonical correlation analysis (CCA) and (2) assessing the accuracy of estimated flood quantiles by applying a common model that estimates flood quantiles using information from the two alternate groups of homogeneous sites (from network theory and CCA). Results show that network theory offers merit in delineating homogenous regions, with resulting design flood estimates showing improvements across different return periods compared to the CCA alternative used.

Regional hydrological frequency analysis at ungauged sites with random forest regression

“…Haddad et al (2012) showed that the ROI approach leads to more efficient and accurate flood quantile estimates compared to the fixes regions approach. Another such technique, Canonical Correlation Analysis (CCA), has been used for delineating homogenous regions in a number of studies ( See for instance Ouarda et al, 2000;Han et al, 2020). In the present study, CCA is used to delineate homogenous regions as Ouarda et al (2008) indicated that it leads to superior performances.…”

Fuzzy Ensemble Clustering Approach to Address Regionalization Uncertainties in Flood Frequency Analysis

“…Furthermore, each of the regionalization approaches (ROI, CCA, CA) considered in this study has several variants. One could attempt preparing ensembles corresponding to different variants of ROI (e.g., Cunderlik & Burn, 2006; Durocher et al., 2018; Formetta et al., 2018; Mostofi & Burn, 2019; Zrinji & Burn, 1996), CCA (e.g., Han et al., 2020; Ouali et al., 2016; Ouarda et al., 2000; Ribeiro‐Corréa et al., 1995; Shu & Ouarda, 2007), and CA (e.g., Basu & Srinivas, 2014, 2016; Cassalho et al., 2019; Farsadnia et al., 2014; Rao & Srinivas, 2006a; Wazneh et al., 2015) for investigating improvement in regions derived by application of FEC to those individual ensembles and their possible combinations.…”

Dimensionality reduction for regional flood frequency analysis: Linear versus nonlinear methods

“…Some methods include kernel‐based fuzzy c‐means cluster (FCM) and entropy‐based clustering (Basu & Srinivas, 2014, 2015). Han et al (2020) used a network approach for delineation on 202 catchments in Australia, while Jothiprakash et al (2021) applied FCM‐based regional flood frequency analysis to 43 watersheds of west‐flowing rivers in Kerala, India. Zalnezhad et al (2022) used adaptive neuro‐fuzzy inference system (ANFIS) using data from 181 gauged catchments in south‐eastern Australia for non‐linear regional flood frequency analysis.…”