DISCRIMINANT ANALYSIS FOR DETERMINING HOMOGENEOUS SAMPLE POPULATION OF LANDSLIDE GROUPS IN HYDROLOGICAL TRIGGER ANALYSIS

M. Sorriso-Valvo (CNR-IRPI Cosenza, via G. Verdi, 248 - 87030 Roges di Rende, Italy)

Simple hydrological models for triggering of landsliding must conform the principle of stationarity for landsliding phenomena to which the model has to be applied. Non-stationarity landslides could be used, but the hydrological response model would become too complex for diffused landslides. Simple hydrological models reflect more realistically the prediction capability in terms of probability of reactivating a pre-defined or non pre-defined number of phenomena. Stationarity must be achieved in terms of homogeneity of behaviour of the sample group. It is necessary to assume that each group is composed of individuals not significantly different in terms of response to the hydrological trigger. As a first approach, given a population of diffused landslides, grouping can be achieved on the basis of the generally acknowledged principle that landslide morphology and slope geomorphology reflect landslide behaviour. Using discriminant analysis it is possible: to define groups whose characters are significantly different between them (significance level < = 5%); or to merge groups belonging to different environments (e.g., different rock composition). In the second case it is necessary to deal with absolutely no difference between groups (significance level > = 95%), but this condition is hard to find. Thus, calling Zg the percent of cases correctly classified for each group g it is tentatively suggested that if groups are more than 2, then they can be merged as far as their Zg < 2Zn (where Zn=1/n, i.e., the percent of cases correctly classified just by chance). As an example, the case study of a zone in northern Calabria is described, where a simple hydrological model could be applied to landslides of the same type (earth slide flow) affecting slopes in different sedimentary rocks.