Estimation of alluvial sediments properties with a multilevel cluster analyses of spectral induced polarization data

In order to infer the sedimentological properties of saturated samples of alluvial deposits, a cluster analysis is applied on spectral induced polarization data (amplitude and phase resistivity spectra). The hierarchical tree of clusters is dissected and analyzed at different aggregation levels, comparing the electrical and sedimentological intra-cluster variability. A low number of clusters densely populated guarantees a perfect allocation into cluster on the base of a few electrical parameters but permits to obtain only a rough textural characterization of the sample, while increasing the number of clusters yields to reduce the variability intervals of the textural parameters and of the chemical properties of the saturation water within each cluster, thus reducing the uncertainty in their estimation. An excessively large number of clusters is ineffective since the uncertainty in the assignment of a sample to a specific cluster is too high. These results can be used as an indirect tool for sediments characterization and have been successfully tested on an independent sample. The cluster analysis allowed to use the whole spectral information to establish a link between geological and geophysical parameters, avoiding the use of one-to-one correlations, whose validity has not yet been confirmed for natural unconsolidated materials.