Monothetic Clustering of Hydrologic Catchment Characteristics
Abstract
Monothetic clustering (compared to polythetic methods) involves grouping observations together that share certain characteristics, providing enhanced interpretations of the cluster solutions because all members of each group have shared attributes. These shared characteristics and the final clusters are created using optimal recursive binary splits of the available observations on individual variables from the suite of available multivariate observations. Methods for selecting the number of clusters and the potential for handling missing observations are discussed. To illustrate the methods, clustering of fifteen hydrologic characteristics of n=151 gauged basins from Southeast Australia is considered. For each catchment, we calculate summary measures of the catchments physical attributes and their hydrologic functionality, obtaining six hydrological signatures calculated from 10-year continuous streamflow data and nine physical catchment metrics. The methods show which variables most clearly differentiate the clusters, allowing immediate conclusions about the variables that drive differences in the clusters. Profiles of cluster representatives are also used to understand the cluster solution. The status of the development of an R package to make the methods generally available will also be discussed.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H53E1460G
- Keywords:
-
- 1804 HYDROLOGY Catchment;
- 1914 INFORMATICS Data mining;
- 1984 INFORMATICS Statistical methods: Descriptive;
- 1874 HYDROLOGY Ungaged basins