Rockfalls are among the most frequent and least predictable landslide processes in mountain areas. They are sudden, they happen at small scale, and they are spatially scattered, so forecasting them is still hard. The main difficulty is the lack of quantitative, continuous observations of the kinematic precursors that come before failure. This thesis tackles that problem at the Poggio Baldi Landslide Natural Lab (Santa Sofia, Northern Apennines, Italy), a rock slope that has been repeatedly reactivated, most recently in 2010 and March 2025. The work follows four objectives: Infrastructure, Characterization, Detection and Prediction. First, the site was turned from a target of episodic surveys into a permanent, automated multi-sensor observatory. It combines optical time-lapse and high-resolution cameras, UAV and TLS-based 3D surveys, ground-based radar (GB-InSAR, Doppler and ArcSAR), acoustic, thermal and weather monitoring. Second, a geomechanical susceptibility index (Q-SIF) was developed and combined with failure-probability estimates (Pf) to show where failures are most likely and how often they occur. Third, automated Change Detection workflows produced the first daily rockfall inventory of the slope: 171 events over 395 days (May 2022 – May 2023). This inventory served as ground truth. It confirmed the predictive validity of the Q-SIF model: failed cells had higher susceptibility, and susceptibility correlated strongly with rockfall count (R² ≈ 0.78). It also allowed rockfall activity to be compared with lithology and rainfall. Comparing sensors showed that spatial and temporal resolution decide what can be detected. A targeted ultra-high-resolution campaign in the high-hazard zones (≈1.8 cm/px, May–July 2024) recorded 956 events. It also captured, for the first time at the site, centimetre-scale pre-failure kinematics: progressive material loss along a fracture before a block collapsed, a 21.24 cm planar slide, and a 10.68 cm interlayer opening (dilatancy) that came before detachment. These results do not yet make up a predictive model. They do show that the infrastructure and the analytical workflow can isolate and measure the high-resolution signals that future early-warning algorithms need. The permanent laboratory is offered as a replicable model for moving from reactive to proactive rockfall monitoring.

Multi-sensor monitoring for the analyses and prediction of rockfalls at the Poggio Baldi Landslide Natural Lab

SANTICCHIA, GIACOMO
2026

Abstract

Rockfalls are among the most frequent and least predictable landslide processes in mountain areas. They are sudden, they happen at small scale, and they are spatially scattered, so forecasting them is still hard. The main difficulty is the lack of quantitative, continuous observations of the kinematic precursors that come before failure. This thesis tackles that problem at the Poggio Baldi Landslide Natural Lab (Santa Sofia, Northern Apennines, Italy), a rock slope that has been repeatedly reactivated, most recently in 2010 and March 2025. The work follows four objectives: Infrastructure, Characterization, Detection and Prediction. First, the site was turned from a target of episodic surveys into a permanent, automated multi-sensor observatory. It combines optical time-lapse and high-resolution cameras, UAV and TLS-based 3D surveys, ground-based radar (GB-InSAR, Doppler and ArcSAR), acoustic, thermal and weather monitoring. Second, a geomechanical susceptibility index (Q-SIF) was developed and combined with failure-probability estimates (Pf) to show where failures are most likely and how often they occur. Third, automated Change Detection workflows produced the first daily rockfall inventory of the slope: 171 events over 395 days (May 2022 – May 2023). This inventory served as ground truth. It confirmed the predictive validity of the Q-SIF model: failed cells had higher susceptibility, and susceptibility correlated strongly with rockfall count (R² ≈ 0.78). It also allowed rockfall activity to be compared with lithology and rainfall. Comparing sensors showed that spatial and temporal resolution decide what can be detected. A targeted ultra-high-resolution campaign in the high-hazard zones (≈1.8 cm/px, May–July 2024) recorded 956 events. It also captured, for the first time at the site, centimetre-scale pre-failure kinematics: progressive material loss along a fracture before a block collapsed, a 21.24 cm planar slide, and a 10.68 cm interlayer opening (dilatancy) that came before detachment. These results do not yet make up a predictive model. They do show that the infrastructure and the analytical workflow can isolate and measure the high-resolution signals that future early-warning algorithms need. The permanent laboratory is offered as a replicable model for moving from reactive to proactive rockfall monitoring.
16-mar-2026
Inglese
MAZZANTI, PAOLO
Università degli Studi di Roma "La Sapienza"
113
File in questo prodotto:
File Dimensione Formato  
Tesi_dottorato_Santicchia.pdf

accesso aperto

Licenza: Creative Commons
Dimensione 6.56 MB
Formato Adobe PDF
6.56 MB Adobe PDF Visualizza/Apri

I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.14242/385411
Il codice NBN di questa tesi è URN:NBN:IT:UNIROMA1-385411