A multitemporal kernel density estimation approach for new triggered landslides forecasting and susceptibility assessment.

In this work a new method for local landslide susceptibility evaluation and forecasting has been proposed, based on spatial statistics techniques and in particular on kernel density estimation. There are different methods existing in literature for this topic. One of the most used is the statistical bivariate method, based on the selection of different environmental factors. It calculates a susceptibility index that expresses how much each single factor weights and contributes in landslide hazard. The first limitation of the results obtained in this way is connected to the global character of the estimate. The second limitation is connected to the impossibility to have information on susceptibility from the interaction between landslides that are located close to each other, which are second order effects in landslides distribution. For these reasons this work proposes a new method that combines the bivariate statistical method with an approach based on kernel density estimation that was used and calibrated properly for landslides study. It was tested on a multitemporal landslides dataset located in Basilicata region (southern Italy) at Bosco Piccolovillage, where last landslide case history occurred on February-March 2005, inducing damages and collapses of about 80% of the buildings in the village. The test site has been useful in order to obtain a detailed landslide hazard zonation, more sensible to local variations of parameters, such as the spatial concentration and relationships landslide phenomena.