Application of clustering methods for hydrological Regionalization using the camels-br database

Abstract

The catchments parameters regionalization is crucial for streamflow prediction in ungauged basins model parameterization and watershed development and management. To overcome the limitation of reduced amount of hydrological data the Catchment Attributes and MEteorology for Large-sample Studies – Brazil (CAMELS – BR) was produced and made publicly available. Limited application of clustering methods in catchment analysis in Brazil particularly using the CAMELS-BR dataset highlights a research gap in the literature. This study presents a robust catchment clustering methodology that incorporates multiple clustering methods and addresses their divergences applied to the CAMELS-BR dataset. The methodology introduced in this study involves a multi-method clustering approach that combines the K-means Partitioning Around Medoids (PAM) and Fuzzy C-means (FCM) techniques. The literature has not explored the establishment of a consensus among clustering methods for classification unlike the methodology proposed in this study which emphasizes deriving a classification based on collective agreement among multiple methods rather than relying solely on individual performance metrics. The hydrological clustering conducted in this study shows a low level of agreement with the hydrographic regions defined by ANA.