Salta al contenuto principale
Passa alla visualizzazione normale.

CHRISTIAN CONOSCENTI

Landslide susceptibility zonation by exploiting GIS tools and two statistical methods: binary logistic regression and multivariate adaptive regression splines. A test in western Sicily (Italy)

  • Autori: Conoscenti, C; Ciaccio, M; Gómez Gutiérrez, Á; Rotigliano, E; Agnesi, V
  • Anno di pubblicazione: 2013
  • Tipologia: eedings
  • Parole Chiave: Landslide suceptibility; GIS; Binary logistic regression; multivariate adaptive regression splines; Sicily
  • OA Link: http://hdl.handle.net/10447/84147

Abstract

In the recent years advanced statistical methods and GIS tools have been frequently used to assess landslide susceptibility. The latter is estimated by establishing statistical relationships between landscape characteristics and spatial distribution of past slope-failures. These are mapped mainly by recognizing changes on slope morphology produced by gravity. Despite of this, most of the researches on landslide susceptibility do not consider that slope-failures modify topography and associate high probability of landsliding with topographic characteristics that differ from those that led slope-failures. In this research we analyzed landslide susceptibility in the basin of the Malvello river with two statistical methods: binary logistic regression and multivariate adaptive regression splines. The study area, which extends for 51 km2 in western Sicily (Italy), is characterized by large outcroppings of clays and marls and is severely affected by shallow landslides. Bedrock lithology and land use were included as predictive variables, in addition to a set of primary and secondary topographic attributes. The latter were derived from a digital elevation model where altitude of areas hosting landslides was interpolated from adjacent undisturbed portions of the slopes. We assume that these artificial surfaces represent the old topography more efficiently respect to the morphology of depletion and accumulation zones of landslides. Ten random samples, with the same number of positive and negative cases, were used to train and test the susceptibility models. Accuracy, as well as overfitting and robustness of the models, were evaluated by drawing receiver operating characteristic (ROC) curves and calculating the area under the ROC curve. In addition of comparing the performance of the statistical methods, the validation results allowed us to highlight advantages/drawbacks of reconstructing original topography of slope-failures when mapping landslide susceptibility.