Large-scale assessment of rainfall-induced landslides in pyroclastic soils of Campania (Italy): a synthetic hydrometeorological approach

The development of reliable tools for assessing rainfall-induced landslide hazard over large areas is often constrained by the limited availability of historical landslide inventories and high-quality rainfall data. This challenge is particularly evident in the pyroclastic soil deposits of Campania (southern Italy), where coarse-grained soils formed by air-fallen volcanic material exist in alternating layers. These deposits are frequently affected by rainfall-induced landslides, primarily triggered by intense rainfall, with antecedent soil moisture acting as a key preparatory factor.

In this study, the Partenio and Sarno Mountains, covering an area of approximately 500 km² and monitored by 23 rain gauges, were subdivided into three zones based on the probability distributions of rainfall event series. Events were separated using a minimum inter-event time of 24 hours with rainfall amounts lower than 2 mm. The zoning reflects the orographic control on rainstorms in the area and was defined using Kolmogorov-Smirnov tests. For each zone, the NRSP stochastic model of rainfall was calibrated based on observed rainfall data, and 500-year-long synthetic hourly rainfall time series were generated. These synthetic series were then used as input to a 1D model of the flow in the unsaturated soil deposit, to simulate the response to precipitation for a representative slope in each zone. The resulting time series of soil moisture and soil suction were employed to perform slope stability analyses, evaluating the factor of safety (FS) with the infinite slope model.

Using the synthetic dataset, empirical thresholds for landslide prediction were derived for each zone, including both meteorological thresholds (based on rainfall intensity and duration) and hydrometeorological thresholds (combining rainfall depth with antecedent root-zone soil moisture). The results indicate that hydrometeorological thresholds are more effective than meteorological thresholds when rainfall and slope properties are accurately known. Moreover, the inclusion of antecedent hydrological variables allows the identification of two distinctive landslide-triggering mechanisms typical of the initial and end phases of the rainy season.

To improve the reliability of the proposed approach, uncertainties associated with the spatial variability of geomorphological slope properties and hydrometeorological variables were explicitly considered. These uncertainties were modeled as normally distributed random errors, and the synthetic datasets of the representative slopes were accordingly perturbed. Accounting for uncertainty shows the robustness of the hydrometeorological thresholds, limiting both false alarms and missed events across all zones. This result was confirmed through validation against available landslide, rainfall, and root-zone soil moisture data for the period 1999-2025.

The proposed methodology provides a practical framework for incorporating uncertainty in hydrometeorological information into landslide hazard assessment over large areas. Furthermore, once the site-specific dominant hydrological processes and controlling variables are identified, the approach can be readily transferred to other regions affected by rainfall-induced landslides.