Uncertainty analysis of 3D hydrometeorological thresholds for rainfall-induced landslides forecasting

Rainfall-induced landslides are widespread geohazards, often characterized by shallow and fast movements. Their occurrence is not easily predictable. Particularly, southern Apennines of Campania (Italy), widely covered by loose pyroclastic deposits laying upon limestone bedrock, are often subjected to massive shallow landslides after intense and long precipitation. The operational early warning systems for rainfall-induced landslides (LEWS) usually rely on empirical thresholds based only on the precipitation information (e.g., intensity and duration of rainfall event), which give rise to false and missed alarms. The reliability of landslide prediction would benefit from the inclusion of hydrological information about the state of the slope prior to rainfall events. In fact, in the last decade, novel hydrometeorological thresholds that mix hydrologic predisposing factors and the features of rainfall events have been developed for landslide forecasting. Specifically, adding information linked to major hydrological processes occurring in the slope improves the performance of LEWS.

The study refers to landslide-prone areas nearby the town of Cervinara, on the slopes of Partenio Massif, representative of a geomorphological setting typical of wide areas of Campania (Italy). Firstly, to obtain a significant data series for statistical analyses, a 1000-year hourly synthetic dataset, mimicking the hydrological response of the slope to meteorological forcing, was generated. Specifically, a stochastic NSRP rainfall model was coupled with a physically based model of the unsaturated flow through the soil cover, hydraulically connected to a linear reservoir simulating a perched aquifer which develops in the uppermost part of the bedrock during the wet season. Both the models had been previously calibrated and validated based on field monitoring data. The synthetic dataset of the slope cover response to precipitation is obtained in terms of soil suction and water content, and perched aquifer water level. The stability of the slopes is assessed under the infinite slope hypothesis, allowing the identification of the occurrence of landslides. The results highlight how novel approaches in the definition of thresholds, considering the 3D hydrometeorological space (i.e., root zone soil moisture, aquifer water level and rainfall event depth), can significantly improve their predictive performance, compared to the common bidimensional thresholds based either on meteorological or hydrometeorological variables.

Moreover, in real practical applications for landslide forecasting, it is not always possible to implicitly assume a perfect knowledge of the variables to be measured for defining the thresholds, especially for a wide area. In fact, both the hydrological and meteorological variables are affected by significant uncertainty, mainly owing to spatial variability. Similarly, the calculated factor of safety, based on the simulated soil moisture and pressure and the assumed soil physical parameters, can be affected by uncertainty, as slope morphological characteristics and soil hydraulic and geotechnical properties are also variable in space. Thus, in this respect, the effects of the uncertainty of slope geomorphological characteristics, as well as of soil hydraulic and geotechnical properties, embedded as probabilistic variables, have been investigated on the obtained 3D hydrometeorological thresholds and on the corresponding predictive performance.