Integrated monitoring of a landslide sequence at the Poggio Baldi Natural Laboratory (Italy) and perspectives for ANN-based learning

The increasing frequency of landslides, driven by intense meteorological events, demands the development of functional, reliable, and cost-effective monitoring strategies. In this context, natural laboratories equipped with multi-sensor infrastructures represent essential facilities for testing the integration of multi-platform and multi-temporal remote sensing data for landslide hazard assessment.  This study presents the monitoring framework and its application to a complex landslide sequence that occurred in March 2025 at the Poggio Baldi Natural Laboratory (Northern Apennines, Italy). Triggered by an exceptional rainfall event characterised by 117.8 mm of cumulative precipitation in 24 hours, the sequence involved a two-stage failure process: an initial 30,000 m³ earth flow followed, approximately 48 hours later, by a 35,000 m³ rockslide.

The monitoring infrastructure enabled a multi-scale characterization of the entire rainfall event, documenting the activity of the entire slope and assessing the rock failure activity from the main scarp. At the cliff scale, the permanent ground-based monitoring network — integrating optical and thermal cameras, acoustic sensors, and meteorological stations — captured the kinematic evolution of both failure phases. Digital Image Correlation (DIC) applied to optical and thermal sequences allowed high-frequency quantification of the earth flow displacement field, which reached peak velocities of 100 cm/h. Thermal infrared analysis identified pre-failure anomalies, likely related to localised soil saturation and initial surface deformation during nighttime. For the rockslide, acoustic monitoring enabled a three-phase reconstruction of the collapse dynamics, while motion-triggered optical systems detected a significant increase in rockfall frequency as a clear precursor to the main failure. Post-event characterization was achieved through high-resolution UAV photogrammetry for volumetric quantification and Ground-Based Interferometric Arc-SAR (GB-InSAR) monitoring, which documented the transition from active displacement to slow-moving residual deformation and highlighted the slope's sensitivity to subsequent rainfall events. Satellite imagery from Sentinel-2 and PlanetScope provided detection of the slope response to the meteorological trigger, identifying surface changes in the immediate aftermath of the rainfall, limited to the upper slope.

The continuous monitoring results also highlight its importance in constituting large training datasets suitable for the development of nowcasting and near-forecasting strategies. Furthermore, the multi-year continuous datasets collected from 2021 at Poggio Baldi, combining high-frequency meteorological records with detailed rockfall inventories, are currently being exploited to train deep learning models based on Neural Networks approaches. These models aim to capture the complex, non-linear relationships between meteoclimatic drivers and slope response, with the ultimate goal of developing predictive tools for rockfall occurrence at the cliff scale. The presented monitoring system based on continuous sensors and periodic surveys proved to be a cost-effective framework able to provide robust and scalable solutions for landslide monitoring