Rock mass geomechanical properties to improve rockfall susceptibility assessment: a case study in Valchiavenna (SO)

Abstract

The overarching goal of the study is to develop a rockfall susceptibility map for Valchiavenna (SO), located in the Italian Central Alps. The approach was two-fold: the first part of the work consisted of developing geomechanical maps, which are relevant to rock mass instability, whilst the second part was aimed to the implementation of the obtained geomechanical maps as predictors in a statistically based rockfall susceptibility model. The chosen target variables, collected in an available geomechanical field surveys database, were Joint Volumetric Count (Jv), the equivalent hydraulic conductivity (Keq), and weathering index (Wi). The available dataset was updated with several new geomechanical surveys, whose locations were chosen through the application of the Spatial Simulated Annealing algorithm. Based on this updated and homogenised dataset, the target properties were regionalized using different deterministic, geostatistical and regression techniques, comparing performance and error metrics resulting from a leave-one-out cross-validation procedure. Regionalization results of the target variables showed different reliability degrees. To improve the hydrogeological processes understanding on another spatial scale, an infiltration density map was prepared, based on field-mapped elements prone to infiltration-Rockfall susceptibility modelling was performed using Generalized Additive Models (GAM), along with the more commonly used topographic predictors. Model performance is assessed using both non-spatial and spatial k-fold cross-validations to estimate the area under the receiver operating characteristic curve (AUROC). Predictor smoothing functions and deviance explained were analysed in order to assess the influence of the geomechanical predictors on the model. The geological-geomorphological plausibility of the susceptibility map including geomechanical predictors was assessed by a comparison with the only topography-based susceptibility map. Model results showed reliable rockfall discrimination capabilities (mean AUROC>0.7). Rockfall data for model training and testing were extracted from the IFFI (Inventario dei Fenomeni Franosi in Italia) inventory and updated with additional field-mapped rockfalls. A potential inventory bias in the IFFI inventory was observed by comparing performance and predictors behaviour of models built with and without the additional rockfalls.