A novel dynamic rockfall susceptibility model including precipitation, temperature and snowmelt predictors: a case study in Aosta Valley (northern Italy)

Abstract The overarching goal of the study was development a potentially dynamic rockfall susceptibility model by including climate predictors. work is based on previously defined critical thresholds relating three indices — effective water inputs (EWI), wet-dry cycles (WD) and freeze–thaw (FT) occurrence. pilot area located in Aosta Valley region (Italian Western Alps). settings were optimized through stepwise procedure, carried out means generalized additive models (GAM). Predictors included topographic, climatic additional snow-related variables. As predictors, mean annual threshold exceedance frequency calculated for each index. All developed an automatic penalization statistically non-significant variables (i.e. shrinkage). initial set without considering potential inventory bias. Secondly, “visibility mask” produced to limit modelling domain according event census procedures. Thirdly, GAMs functional relationships analysed verify physical plausibility Finally, reduce concurvity, principal component analysis (PCA) predictors out. Key findings as follows: (i) ignoring bias led excellent performance but physically implausible outputs; (ii) selection non-rockfall points inside managing influence (iii) inclusion resulted improvement interpretability associated maps, being EWI, WD maximum cumulated snow melting most important plausible predictors; (iv) PCA strategy can efficiently concurvity.