3D hydrometeorological thresholds for early warning of rainfall-induced landslides in Campania (Italy): application to Partenio massif

Large mountainous areas of Campania (Italy) are frequently subject to rainfall-induced landslides, which sometimes cause heavy damage to buildings and infrastructure. Specifically, landslides are triggered on steep slopes covered with unsaturated air-fall pyroclastic deposits from eruptions of Vesuvius resting upon fractured limestone bedrock. The characteristics of these phenomena, their wide diffusion, and the difficulty of predicting their exact time and place of occurrence, which strongly depend on local factors, make the recourse to structural risk mitigation interventions rarely feasible. Hence, landslide early warning systems (LEWS) are the most effective way to mitigate the associated risk. Currently, the operating LEWS are based on empirical thresholds based only on precipitation information (e.g., intensity and duration of precipitation), but they give rise to numerous false and even some missed alarms. The inclusion of antecedent hydrologic information prior to rainfall events improves the predictive performance of hazard assessment tools and is here applied to the definition of hydrometeorological thresholds to be implemented in LEWS.

A novel methodology is proposed to define the hydrometeorological thresholds for large areas, considering the uncertainties linked to the spatial variability of geomorphological and meteorological factors. The proposed methodology is applied to the north-facing side (an area of approximately 80 km²) of the Partenio Mountains, a carbonate massif in Campania (Italy), frequently hit by rainfall-induced debris flows involving the pyroclastic deposits mantling the steep slopes.

As it often happens for geohazard inventories, the available dataset is too scarce to allow carrying out significant statistical analyses. Therefore, a 500-year long synthetic dataset of the hydrological response to precipitation of a reference slope with known geometry and a homogeneous soil layer with known properties is generated, providing the values of root-zone soil moisture and aquifer water level. Specifically, a stochastic NSRP rainfall generator is coupled with a previously developed physically based model of the flows through the unsaturated deposit and its hydraulic connection to a perched aquifer forming during the rainy season. The slope stability is evaluated under the infinite slope hypothesis, which allows the identification of landslide events. To define an operational LEWS for the whole study area, the effects on slope stability of the uncertainty related to the spatial variability of the slope morphological features, soil hydraulic and geotechnical properties is introduced. Similarly, the uncertainty of the meteorological and hydrological variables used for the definition of the 3D thresholds (rainfall depth, root-zone soil moisture and aquifer water level) is also considered, to mimic the effects of spatially variable quantities observed only in few sparse points. Consequently, the synthetic dataset is perturbed, superimosing Normal-distributed random fluctuations on the hydrometeorological variables and on the calculated values of the factor of safety.

The effect of uncertainty on the operational predictive performance shows the robustness of the hydrometeorological thresholds. Moreover, this result is confirmed by the application of the obtained thresholds to available data of occurred landslides, and measured rainfall and soil moisture in the north-facing part of Partenio Massif in the period 2002-2020.