Comparison of different regionalization methods for streamflow drought characterization at ungauged catchments

The analysis of low flow indices is carried out here on the discharge data, recorded from 1923 to 2011, by 61 consistent hydrometric stations with at least 6 years of measures, located in Tuscany region (an area of about 23.000 km2), in Central Italy. The area is subdivided into different subregions using the L-moments method applied to the 7-day annual minima and to the Q70 annual series. The division into subregions is tested using discordancy and heterogeneity statistics. Different subdivisions into 1-3-4-5 regions, starting from previous studies on rainfall extreme values, are tested. For each river section the catchment area is identified and a new space is defined as a combination of the catchment geomorphologic and climatic characteristics. In this space a multivariate analysis (MA) based on natural logarithms of seven geomorphoclimatic characteristics (i.e. catchment area, longest flow paths; topographic mean slope; mean elevation; difference between maximum and minimum elevation; average value of Mean Annual Precipitation; mean soil permeability) is used to relate the duration and annual minimum indices of low flow to the investigated territory characteristics. Top-kriging interpolation technique, validated using a cross-validation procedure, is also applied. Top-kriging takes both the area and the nested nature of catchments into account and it is chosen for estimating stream flow related variables, because it is directly connected to the hydrographic network structure and geometric organization. The results of MA and top-kriging are compared with the ones of other interpolation techniques applied to different subdivisions in homogeneous regions, either deterministic, such as Inverse Distance Weighted, or geostatistical as Ordinary Kriging with an exponential semivariogram. The results are all validated using a jack-knife procedure. To compare the results, different error measurement (mean square error, mean relative error, etc.) are also assessed, to quantify the accuracy of the different techniques and to define the most suitable method for low flow regionalisation.