Comparison of Logistic Regression and Random Forests techniques for shallow landslide susceptibility assessment in Giampilieri (NE Sicily, Italy)

摘要: Abstract The aim of this work is to define reliable susceptibility models for shallow landslides using Logistic Regression and Random Forests multivariate statistical techniques. study area, located in North-East Sicily, was hit on October 1st 2009 by a severe rainstorm (225 mm cumulative rainfall 7 h) which caused flash floods more than 1000 landslides. Several small villages, such as Giampilieri, were with 31 fatalities, 6 missing persons damage buildings transportation infrastructures. Landslides, mainly types earth debris translational slides evolving into flows, triggered steep slopes involved colluvium regolith materials cover the underlying metamorphic bedrock. has been carried out following steps: i) realization detailed event landslide inventory map through field surveys coupled observation high resolution aerial colour orthophoto; ii) identification source areas; iii) data preparation controlling factors descriptive statistics based bivariate method (Frequency Ratio) get an initial overview existing relationships between causative iv) choice criteria selection sizing mapping unit; v) implementation 5 techniques focused vi) evaluation influence sample size type sampling results performance models; vii) predictive capabilities ROC curve, AUC contingency tables; viii) comparison model obtained maps; ix) analysis temporal variation related input parameter changes. Models have demonstrated excellent capabilities. Land use wildfire variables found strong control occurrence very rapid