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

Abstract

Estimates of flood quantiles are necessary at various locations/hotspots in river basins for applications related to hydrological design and risk assessment of water resources systems. Desired flood quantiles could be estimated by frequency analysis of the available at-site data for locations having sufficient historical information on flows. However, for an ungauged or sparsely gauged location, the flood quantiles' prediction remains one of the challenging problems in hydrology. A variety of regional flood frequency analysis (RFFA) approaches are available for addressing this challenge. The approaches generally involve (i) regionalization, which refers to identification of a group of watersheds (i.e., region) resembling the target site's watershed in terms of attributes/characteristics influencing flood, and (ii) regional frequency analysis (RFA), which involves pooling of flood-related information from sites in the region and using it to estimate desired flood quantile(s) at the target site. Approaches to watershed regionalization for RFFA include those based on regression analysis (method of residuals), canonical correlation analysis (CCA), region of influence (ROI) and its extensions, hierarchical approach, and cluster analysis. Several articles and books provide an overview of the literature on regionalization of watersheds and issues related to RFA (e.g.,

Hydrological Similarity-Based Parameter Regionalization under Different Climate and Underlying Surfaces in Ungauged Basins

“…Uncertainty of hydrological model parameters is an important source of uncertainty in hydrological prediction. This uncertainty can be reduced by obtaining observed data required to determine the values of unknown parameters for hydrological models [2][3][4]. Parameter optimization algorithms are usually applied to basins with observed data to obtain simulations of runoff that are as close as possible to the observed runoff.…”

A Bayesian Fuzzy Clustering Approach for Design of Precipitation Gauge Network Using Merged Remote Sensing and Ground‐Based Precipitation Products

“…The proposed methodology could also be extended to design other networks monitoring precipitation (e.g., radar) and different hydrometeorological and hydrological variables. For example, to design a stream gauge network in a river basin, one could first use an effective regionalization methodology (e.g., Kiran & Srinivas, 2021) to delineate catchments in the basin into homogenous zones based on their hydrogeomorphological characteristics. Subsequently, ground‐based streamflow measurements (at existing gauges) and information on flows from satellite‐based products and/or rainfall‐runoff models could be utilized to determine flows at several ungauged locations (hotspots) on the stream network.…”

Multivariate Regional Frequency Analysis Using Conditional Extreme Values Approach

“…To facilitate this, the site‐specific attributes can be chosen such that they are available even for ungauged locations. In applications where the spatial extent of the study area is large, one can delineate the area into homogeneous regions using any effective procedure (e.g., Kiran & Srinivas, 2021b) and apply the proposed methodology to each of the delineated regions.…”