Large-scale assessment of rainfall-induced landslide hazard based on hydrometeorological information: application to Partenio Massif (Italy)

Abstract. The definition of reliable tools for rainfall-induced landslide hazard assessment is often limited by the lack of long records of occurred landslides and relevant hydrometeorological variables. This is the case of the mountainous areas of Southern Apennines of Campania (Italy), diffusely covered by loose pyroclastic deposits laying upon limestone bedrock, and frequently subjected to rainfall-triggered shallow landslides. To get around this issue, a 500-year long synthetic dataset of the response to precipitation of a typical slope of the area has been generated, by means of a physically based model previously validated through experimental data. The obtained dataset, containing hourly values of soil moisture and suction, and of water level in an ephemeral aquifer developing in the uppermost fractured bedrock, has been used to assess slope stability through the calculation of the factor of safety. Based on the synthetic data, empirical thresholds for the prediction of landslide occurrence have been defined, either meteorological (i.e., based on rainfall intensity and duration) or hydrometeorological (i.e., coupling rainfall depth with antecedent root-zone soil moisture or aquifer water level). The results show that, where meteorological forcing and slope characteristics are perfectly known, hydrometeorological thresholds outperform the meteorological ones, and that a 3D threshold based on root-zone soil moisture, aquifer level, and rainfall depth, provides nearly unerring landslide predictions. The use of two antecedent hydrologic variables also allows identifying two different landslide triggering mechanisms, respectively typical of the beginning and of the end of the rainy season. To extend the application to large areas, the uncertainties linked to the spatial variability on slope geomorphologic characteristics and hydrometeorological variables were considered as random errors. Hence, foreseeing the application to the north-facing side of Partenio Massif (about 80 km2), the synthetic dataset has been perturbed, superimposing Normal-distributed random fluctuations to the calculated values of the factor of safety, and to the hydrometeorological variables used as landslide predictors. Although the uncertainty reduces the predictive skill of all the thresholds, the hydrometeorological ones show more robustness, with small numbers of both missed and false alarms. This result is confirmed by the application of the obtained thresholds to available data of landslides, rainfall and root-zone soil moisture for the period 2002–2020 in the area.