Monthly Runoff Regime Regionalization Through Dissimilarity-Based Methods

Abstract

A number of procedures can be cited in the literature to perform stream flow prediction in an ungauged basin. Stream flow functions as flow duration curve and flood frequency curves can be obtained by hydrological or statistical models. Also flow regime curves are needed for water resources assessment: they are complex (non monotonic) functions and require special care in the parameterization. Here we propose a dissimilarity-based regionalization model to estimate this particular feature of the stream flow process, as the monthly flow regime. The proposed regional statistical frame work is based on the measure of the dissimilarity (sometimes also referred to as distance) between all the possible pairs of flow regimes available in the region. Each regime is considered as a whole hydrological object and the distance between each pair of regime curves is computed through a suitable metric in a non-parametric way. Dissimilarity values then compose a distance matrix which characterizes the variability of the regime shapes in the region of interest. The prediction of regimes in ungauged basins is obtained by creating corresponding distance matrices of basin features taken among geographic, geomorphologic and climatic attributes, usually referred to as descriptors. Suitable basin descriptors are those whose distance matrices are reasonably correlated to the flow regime distance matrix. This choice allows us to use complex descriptors, like the rainfall regime curve. Identification of the suitable descriptors is performed through an unsupervised procedure based on multiple regressions on distance matrices. Once identified the relations, the candidate descriptors of the ungauged basin can be used to select the most similar gauged basins to use as neighbours for estimation of the required runoff regime. The procedure is applied to a set of 118 basins located in northwestern Italy. The performance of the regional estimation is assessed by means of a cross-validation procedure and through comparison with other parametric regional approaches. In most of the cases, the distance-based model produces better estimates of flow regimes than the “standard” procedure, using only few catchment descriptors, with the advantage of demonstrating the role of complex basin features, as for instance the rainfall regime curve.