NEURAL NETWORK AIDED EVALUATION OF LANDSLIDE SUSCEPTIBILITY IN SOUTHERN ITALY

Landslide hazard mapping is often performed through the identification and analysis of hillslopeinstability factors. In heuristic approaches, these factors are rated by the attribution ofscores based on the assumed role played by each of them in controlling the development of asliding process. The objective of this research is to forecast landslide susceptibility through theapplication of Artificial Neural Networks. In particular, given the availability of past eventsdata, we mainly focused on the Calabria region (Italy). Vectors of eight hillslope factors(features) were considered for each considered event in this area (lithology, permeability, slopeangle, vegetation cover in terms of type and density, land use, yearly rainfall and yearlytemperature range). We collected 106 vectors and each one was labeled with its landslidesusceptibility, which is assumed to be the output variable. Subsequently a set of these labeledvectors (examples) was used to train an artificial neural network belonging to the category ofMulti-Layer Perceptron (MLP) to evaluate landslide susceptibility. Then the neural networkpredictions were verified on the vectors not used in the training (validation set), i.e. in previouslyunseen locations. The comparison between the expected output and the artificial neuralnetwork output showed satisfactory results, reporting a prediction discrepancy of less than4.3%. This is an encouraging preliminary approach towards a systematic introduction ofartificial neural network in landslide hazard assessment and mapping in the considered area.