Classification procedures in the context of PUB: ways forward?

Abstract. Limitations of cluster analysis as a procedure for classifying parameters of rainfall-runoff models are discussed, and a procedure is suggested by which such parameters could be estimated, using site characteristics together with a split record test as a measure of performance. It is suggested that geostatistical models may be a possible alternative to procedures based on cluster analysis, and that long-established principles of experimental design (replication, randomization) be used for comparing alternative PUB procedures.