Evaluating earthquake-triggered landslide hazard at the basin scale through gis in the Upper Sele river Valley

Abstract

To evaluate techniques for assessing earthquake-triggeredlandslide hazard in the Southern Apennines (Italy), a GIS-based analysis was used to modelseismically induced slope deformations. Geological, geotechnical, geomorphological and seismologicaldata were integrated into a standard earthquake slope stability model. The model assessed thelandslide potential that existed during the 1980 Irpinian earthquake in the Upper Sele river Valley.The standard Newmark displacement analysis, widely used for predicting the location of shallowunstable slopes, does not take into account errors and/or uncertainties in the input parameters.Therefore, a probabilistic Newmark displacement analysis technique has been used. Probabilistictechniques allow, e.g., an estimation of the probability that a slope will exceed a certain criticalvalue of Newmark displacement. In our probabilistic method, a Monte-Carlo based simulation modelis used in conjunction with a GIS. The random variability of geotechnical data is modelled by probabilitydensity functions (pdfs), while for the seismic input three different regression laws wereconsidered. Input probability distributions are sampled and the resulting values input into empiricalrelations for estimating Newmark displacement. The outcome is a map in which to each siteis related a spatial probability distribution for the expected displacement in response to seismic loading.Results of the experiments show a high grade of uncertainty in the application of the Newmarkanalysis both for the deterministic and probabilistic approach in a complex geological setting suchas the high Sele valley, quite common in the Southern Apennines. They show a strong dependence onthe reliability of the spatial data used in input, so that, when the model is used at basin scale,results are strongly influenced by local environmental condition (e.g., topography, lithology, groundwatercondition) and decrease the model performance.