Classification of Ontario watersheds based on physical attributes and streamflow series

•We investigate NLPCA and SOM for gauged/ungauged catchment classification.•We identify hydrologically homogenous clusters of basins using runoff indices.•We use classification based on PCA and runoff indices as benchmark classification.•NLPCA indicates reliable performance in gauged/ungauged catchment classification.•NLPCA can be used for ungauged basins classification prior to regionalization.

SummaryNonlinear cluster analysis techniques including Self Organizing Maps (SOMs), standard Non-Linear Principal Component Analysis (NLPCA) and Compact Non-Linear Principal Component Analysis (Compact-NLPCA) are investigated for the identification of hydrologically homogeneous clusters of watersheds across Ontario, Canada. The results of classification based on catchment attributes and streamflow series of Ontario watersheds are compared to those of two benchmarks: the standard Principal Component Analysis (PCA) and K-means classification based on recently proposed runoff signatures. The latter classified the 90 watersheds into four homogeneous groups used as a reference classification to evaluate the performance of the nonlinear clustering techniques. The similarity index between the first largest group of the reference classification and the one from the NLPCA based on streamflow, is about 0.58. For the Compact-NLPCA the similarity is about 0.56 and for the SOM it is about 0.52. Furthermore, those results remain slightly the same when the watersheds are classified based on watershed attributes – suggesting that the nonlinear classification methods can be robust tools for the classification of ungauged watersheds prior to regionalization. Distinct patterns of flow regime characteristics and specific dominant hydrological attributes are identified in the clusters obtained from the nonlinear classification techniques – indicating that the classifications are sound from the hydrological point of view.