IDENTIFICATION OF HYDROLOGICAL NEIGHBORHOODS USING CANONICAL CORRELATION-ANALYSIS

Determination of homogeneous flood frequency regions is a necessary st ep in non-Bayesian flood regionalization procedures. A homogeneous reg ion is defined as a group of drainage basins whose standardized hood f requency curves are similar, to within a certain margin of sampling er ror. An alternative to using homogeneous regions is to define a hydrol ogical neighborhood, or region of influence, for each of the drainage basins. In this paper, we propose a method for identifying the hydrolo gical neighborhood of a drainage basin based on canonical correlation analysis. Under the normality assumption, the proposed technique leads to the identification of a neighborhood having an ellipsoidal contour in a space of canonical variables. A data set consisting of 55 catchm ents from the Province of Quebec (Canada) and the jack-knife sampling procedure are used to assess the proposed approach.