The present research work focuses on the regionalisation of hydrometric information (i.e. transferring empirical information on streamflow regime from neighbouring catchments to the catchment of interest), which is widely used for retrieving accurate estimates of hydrological design variables (e.g. flood flows, mean annual streamflow, low-flow indices, etc.) in ungauged or scarcely gauged basins. The literature reports on several statistical regionalisation methods, which are characterised by

more »

... fferent ways of accounting for hydrological similarity between catchments and spatial correlation (or cross-correlation, or intersite correlation) among the hydrological observations collected at different streamgauges. This Thesis aims at deepening our understanding on the added value and impacts of catchment similarity and spatial correlation on the prediction of flood quantiles and flow-duration curves in ungauged river cross-sections by presenting the results of a threefold study. First, we consider the reference procedure for design flood estimation in Triveneto, North-eastern Italy, which assumes the entire study area to be a single hydrologically homogeneous region. Our analyses, based on an updated database of annual maximum floods, confirm the outcomes of previous studies, that is Triveneto cannot be regarded as homogeneous in terms of flood frequency regime; our study also highlights the need for an update of the reference procedure for design flood estimation in the study area. To this aim, we show that a focused-pooling approach, which delineates homogeneous pooling-groups of sites for any given target site by referring to selected geomorphoclimatic descriptors which are particularly relevant for describing regional flood frequency, leads to regional samples characterised by significantly improved homogeneity and, therefore, more reliable design flood estimates. Although focused pooling is capable of properly exploiting catchment similarity, the general approach does not consider the effects associated with spatial correlation among streamflow series. Nevertheless, all regional datasets of annual sequences of flood flows present some degree of cross-correlation between observed series. Its effects on the accuracy of regional prediction are not well studied yet. Therefore, the second part of our study addresses this important issue by considering two regionalisation procedures that do i Abstract consider intersite correlation explicitly, although in two radically different ways. These are the Generalized Least Squares, GLS, and a geostatistical method (i.e. Top-kriging, TK). Recent studies show that TK outperforms GLS for predicting empirical flood quantiles, but they also speculate that the presence of intersite correlation might affect the accuracy of these methods in predicting true flood quantiles. To better understand this aspect, we applied GLS and TK for predicting flood quantiles in a homogeneous pooling-group of sites in Triveneto under different cross-correlation scenarios through a Monte Carlo simulation experiment. Our analyses clearly show that, for both methods, an increasing degree of spatial correlation among the flood sequences results in an increasing masking-effect on the true flooding potential. Morever, we confirm that TK significantly outperforms GLS when they both assume flood quantiles to scale with drainage area alone, yet, we clearly point out that both methodologies (GLS and TK) significantly improve their accuracy and reliability when flood quantiles are regressed against several catchment descriptors, leading to rather similar overall prediction performances. In the third and last part of our study, we compare regression methods and geostatistical methods for predicting flow-duration curves in a large and heterogeneous study region, the Danube river basin. In particular, we show that multi-regression models are not a viable regionalisation procedure, while geostatistical models provide much more accurate predictions of flow-duration curves over large and hydrologically heterogeneous study areas. In summary, all the analyses confirmed the added value for statistical regionalisation of properly handling hydrological heterogeneity, also highlighting the pivotal role played by intersite correlation in observed streamflow time-series. ii