Cluster Analysis as Applied to Regional Flood Frequency

Abstract

A statistical technique for delineating groups of stations to be considered a region for regional flood frequency analysis is presented. The technique, which utilizes cluster analysis, allows the inclusion of a diversity of factors which might be considered to be of relevance when seeking stations in differing basins having similar extreme flow characteristics. The method incorporates both a basin similarity measure, imbedded in the clustering algorithm, and a regional homogeneity measure used to evaluate station partitionings obtained from the clustering algorithm. The result is that groups of stations are identified that can be considered sufficiently homogeneous to effect an efficient spatial data transfer. An application of the methodology to rivers in southern Manitoba, Canada, is presented to illustrate pertinent aspects of the procedure.