Applicazione di analisi statistica multivariata, Rete Neurale Artificiale e metodo euristico per la valutazione della suscettibilità da sinkhole nella piana di S. Vittorino (RI)

San Vittorino plain is located on the northeastern boundary of the Lazio region. It is probably the area within the Lazio Region with the highest density of sink- holes and that presents the highest risk because of the presence of concentrated important structures and infrastructure. The study area is mainly characterized by carbonate forma- tions, with developed karst processes and wherefore tectonic units, defined by a different paleogeographic evolution and separated by regional importance structural elements, con- verge. The sinkhole susceptibility evaluation in the plain was carried out through the application of a multivariate statistical analysis, a heuristic method and a procedure Artificial Neural Networks. The subsidence phenomena in the San Vittorino plain are mainly related to deep piping processes, that caused ground- water movement through faults and fractures that displace the carbonate bedrock. The upward migration of deep seated fluids caused the mobilization and erosion of the shallower continental deposits to which the effects of the dissolution process related to the presence of high content of H2S and CO2 are added. These processes significantly increase during seismic events or high intensity rainfall events. The performance of the predictive models was evaluated using ROC curves. The results show that the Artificial Neural Networks procedure provides a more reliable accuracy; instead, the heuristic model by using bivariate and multivariable analysis have greatly a limited accuracy. Obviously, the heuristic method fails to provide the expected results of the performance because the phenomenon is not known, but the artificial neural networks even if successful to interpret very well the complex phenomena not disclose the relationships between the dependent and independent variables.