Homogeneous region determination using linear and nonlinear techniques

Similarity between catchments in a region can be determined depending on catchment properties. This helps to understand the response behavior of the similar catchments more appropriately. Catchment classification plays a major role in the process of hydrological prediction in the case of ungauged catchments. The following categorization was carried out for 32 catchments of India. Principal Component Analysis (PCA) along with K-means clustering, was applied as linear classification; and Self-Organizing Map (SOM) and Kernel Principal Component Analysis (KPCA) were implemented as nonlinear classification methods on catchment attributes and daily streamflow time series. The classification established on streamflow signatures was taken as the reference classification. Results obtained from PCA, SOM, and KPCA were compared with results of the reference classification. The absence of discordant catchments from the clusters of SOM, based on catchment attributes, suggests homogeneity among SOM-derived clusters. Similarity index scores are 0.48 and 0.47, 0.46 and 0.42 ?for first, second, third and fourth clusters of SOM respectively with that of the reference classification technique. Nonlinear techniques with high similarity index values outperformed standard techniques. This study demonstrated the ability of classification based on catchment attributes to classify ungauged catchments.