Statistical analysis of long-term displacement rate for definition of Early Warning thresholds applied to case studies of the Ruinon and Mont de La Saxe landslides

Large rockslides are characterized by complex spatial and temporal evolution, in addition to non-linear displacement trends and the significant effects of seasonal or occasional events on their behaviour. The displacement rate and the landslide evolution are intensely influenced by many factors like lithology, structural and hydrological settings, other than meteorological and climatic factors (e.g. snowmelt and rainfall). The relationship among these factors is evidently non-linear and site specific for each sector within the main landslide mass. Different Early Warning domains (EWD), characterized by different velocity regimes (slow to fast domains) and with different sensitivity to external perturbations (e.g. snowmelt and rainfall), have been identified in previous studies at the two sites. In this work, total displacement and displacement rate time series are extracted from Ground-based Interferometric synthetic aperture radar (GB-InSAR) surveys, monitoring of optical targets by total stations, a GPS network and multi-parametric borehole probes and from DEM difference (calculated by various algorithms) derived from terrestrial laser scanner surveys. The Mont de La Saxe rockslide (ca. 8 x 106 m3) is located in the Upper Aosta Valley (Crosta et al., 2014), and it has been intensely monitored since 2009 by the Valle D’Aosta Geological Survey and have been subdivided into 5 EWD. The Ruinon landslide (ca. 15 x 106 to 20 x 106 m3) is located in the Upper Valtellina (Lombardy region) and monitoring data are available starting since 2006 (provided by ARPA Lombardia) and subdivided into 8 EWD (Crosta et al., 2016). Both sites are located within an alpine deep-seated rock slope deformation characterized by dissimilar displacement velocity, from centimetres to meters per year, and which have undergone exceptional accelerations during specific events. We experiment the use of normal probability plots for the analysis of displacement rate data of each point belonging to different landslide sectors and recorded during almost ten years of monitoring. These analyses allow us to define: (i) values with a specific probability value expressed in terms of percentiles; (ii) values for which a specific change in behaviour is observed which could be associated to a specific type of triggering event (e.g. rainfall intensity, duration or amount; snowmelt amount). These values could be used to support the choice of threshold values for the management of Early Warning System, by considering also the minimization of false alarms. The analyses have been performed using data averaged with different time intervals so to study the effects of noise on the threshold values. Analyses of false alarm triggered by the choice of different threshold values (i.e. different percentiles) have been implemented and analysed. Finally, cross-correlation has been used to discriminate the different areas. This could represent an innovative approach to define velocity thresholds of Early Warning system and to analyse quantitative data derived from remote sensing monitoring and field surveys, by linking them to both spatial and temporal changes.