Comparison of Method of Residuals and Cluster Analysis for Flood Regionalization

Abstract

The method of residuals was recently used by the U.S. Geological Survey (USGS) to delineate seven flood regions for the state of Kentucky. An alternative approach is to use cluster analysis in conjunction with important statistical properties of the maximum annual flood‐peak series. Applying the FASTCLUS clustering procedure of the Statistical Analysis System, five cluster regions are identified using a similar data base as the USGS study. Flood regions delineated under these two methods of flood regionalization are then compared by examining trends in important hydrological characteristics within each of the regions, and through a discriminant analysis based upon watershed physical attributes. Results show that, although cluster regions are in no way similar to those defined by the method of residuals nor coincident with geographical boundaries, they are more distinguishable and better discriminated in terms of the hydrological characteristics controlling flood response than the USGS regions. Furthermore, standard errors associated with the regression equations relating the log‐Pearson, Type‐III 50‐year flood estimate to watershed physical attributes are comparable under the two methods of regionalization.