Combining clustering and classification for the regionalization of environmental model parameters: Application to floodplain mapping in data-scarce regions. (March 2020)

• Record Type: Journal Article
• Authors: Jafarzadegan, Keighobad; Merwade, Venkatesh; Moradkhani, Hamid
• Is Part Of: Environmental modelling & software. Volume 125(2020)
• Publication Date: 2020-03
• Subjects: Regionalization -- Environmental models -- Data-scarce regions -- Classification -- Floodplain mapping -- Machine learning; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.104613
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Prediction in data-scarce regions is one of the challenging issues in environmental problems. In hydrology, this issue is commonly addressed by utilizing regression or similarity-based regionalization techniques. The core of similarity-based regionalization techniques is a physical/climatic similarity metric that is typically predetermined from the knowledge about the physics of the problem and study area. The purpose of this paper is to: (1) reduce the subjectivity that exists in the selection of the physical/climatic similarity metric by establishing a systematic approach, and (2) propose a generic similarity-based regionalization framework that estimates the parameters of environmental models in data-scarce regions. The efficacy of the proposed framework is evaluated for the regionalization of a statistical model that creates probabilistic floodplain maps in data-scarce regions. Results show a trained Support Vector Machine (SVM) with ten basin descriptors and accuracy of 86% is an appropriate physical/climatic similarity metric that creates reliable floodplain maps in the Arkansas-White-Red region. Highlights: This framework generalizes the regionalization to a wide range of environmental models. This framework reduces the subjectivity that exists in regionalization problems. A new version of the hierarchical clustering algorithm is proposed for basin classification. Basins are clustered based on their behavioral similarity rather than attribute similarity.