Caratterizzazione, modellazione predittiva e studio della variabilità locale e regionale delle proprietà idrologiche dei depositi di versante

Landslides are one of the widespread natural hazards and among these ones there are shallow landslides: landslides that mainly involve the soils (slope deposits, SD) that unconformably cover the geological substratum. A method of prevention that is currently promoted by scientific research is the realization of shallow landslide susceptibility maps which require the knowledge, among different factor, of the engineering-geological properties of the soils. Among the latter, hydrological properties and in particular hydraulic conductivity (K) of the SD, has a relevant weight being the major factor controlling the distribution and movement of the water in the subsoil. So, the requirement of characterizing of the engineering-geological properties of the SD, for which literature is still rather limited, is clear. Specifically, this thesis has as its first objective the characterization of the hydrological properties of the SD, focusing mainly on K. Another important aspect is represented by the fact that the engineering-geological properties of the SD (including K) are not currently well-known due to the high cost (both in terms of time and money) associated with data collection, laboratory and in situ tests. In this regard, this PhD thesis addresses two other aspects. Firstly, to investigate the predictive capacity of different techniques to estimate K from indirect methods. Finally, the local and regional variability of K of the SD was analyzed in order to obtain continuous maps of K. During the first 15 months, an intense field survey was carried out in the study area which has an extension of 420 km2, by means of 150 sampling sites and 720 hydraulic conductivity in situ tests (Ktests) and the collection of 146 samples for laboratory tests. Concerning first topic of this thesis, some of the engineering-geological properties (% gravel, % sand, % fines and Atterberg limits) show a tendency to distribute in different ways in relation to the lithology of the geological substratum. As regards K of the SD, the uncertainty of the Ktests was first estimated (coefficient of variation = 2 – 3%). All Ktests have been divided into 4 horizons according to the depth within which the measurements were performed. A high relationship was observed between log K and the depth of Ktests (R – Pearson = – 0.79). Some Ktests have been performed within shallow landslides it has been observed that K measured inside the landslide is about 4 times lower than that K measured inside landslides. Then, the correlations between K and engineering-geological properties of SD were realized through bi and multivariate statistical analysis: an overall weak correlation among log K and these properties is exhibited. Within the same textural classes and same grain size curves, it emerged that the effect of the lithology of the geological substratum can be considered negligible with respect to the K of the SD. Once the engineering-geological characterization of the SD was completed, another research topic was to evaluate and quantify the predictive efficacy of indirect methods to estimate K. This process was carried out by applying 31 empirical correlations (or pedo-functions, PTF) which are present in the literature and they showed very poor accuracy to predict K of SD. So it was decided to apply two methods for K prediction: multilinear regression and artificial neural networks. Through multilinear regression a new PTF was obtained which proved to be highly effective in predicting K (R2 = 0.82), as well as valid and robust from a statistical point of view. Similar results have been obtained through the implementation of artificial neural networks (R2 = 0.85 – 0.86). So, multilinear regression and neural networks prove to be quite efficient methods in predicting K of the SD. The last topic of this thesis was the analysis of the spatial distribution of K at site and regional scale. For each site the variability of K (range and interquartile range of log K is equal to 2.0 and 0.8 respectively). The spatial analysis at regional scale was performed with two different approaches: in the first case the Ktests were divided into four horizons according to the depth within which the Ktest was performed, in the second approach the entire dataset was previously normalized for the effect of depth. For each of the two approaches, exploratory geostatistical analysis were performed in order to verify the assumptions of normality, stationary and absence of trend, necessary for the correct execution of geostatistical methods. The algorithms of the Ordinary Kriging, Inverse Distance Weighted and Empirical Bayesian Kriging have been implemented which have allowed to generate the first maps of the continuous values of log K at regional scale in the study area. The different algorithms have provided similar results in terms of log K values (for the different approaches considered) obtaining accuracies (NRMSE = 10 – 20%) which are in good agreement with other examples of the literature.