Spatially smooth regional estimation of the flood frequency curve (with uncertainty)

SummaryIdentification of the flood frequency curve in ungauged basins is usually performed by means of regional models based on the grouping of data recorded at various gauging stations. The present work aims at implementing a regional procedure that overcomes some of the limitations of the standard approaches and adds a clearer representation of the uncertainty components of the estimation.The information in the sample records is summarized in a set of sample L-moments, that become the variables to be regionalized. To transfer the information to ungauged basins we adopt a regional model for each of the L-moments, based on a comprehensive multiple regression approach. The independent variables of the regression are selected among a large number of geomorpholoclimatic catchment descriptors. Each model is calibrated on the entire dataset of stations using non-standard least-squares techniques accounting for the sample variability of L-moments, without resorting to any grouping procedure to create sub-regions. In this way, L-moments are allowed to vary smoothly from site to site in the descriptor space, following the variation of the descriptors selected in the regression models. This approach overcomes the subjectivity affecting the techniques for the definition and verification of the homogeneous regions. In addition, the method provides accurate confidence bands for the frequency curves estimated in ungauged basins.The procedure has been applied to a vast region in North-Western Italy (about 30,000 km2). Cross-validation techniques are used to assess the efficiency of this approach in reconstructing the flood frequency curves, demonstrating the feasibility and the robustness of the approach.

► Regional flood frequency analysis with emphasis on prediction uncertainty.
► L-moments are used to reconstruct the flood frequency curve.
► No homogeneous regions required.
► Optimal use of short-record data.
► The model allows the use of non-systematic data.