Automatic detection of landslide events for risk management and early warning procedures

During the last century, the influence of anthropogenic activities on the environment has become more and more impactful. This trend can be attributed mainly to the substantial growth of the human population: the results involve increased urbanization, industrial and construction activities, and the realization of an ever-growing network of transport infrastructures. As a consequence, a sensible increase of the occurrence of natural hazards has been observed, with heavy socio-economic impacts on local communities. Among the different typologies of natural hazards, landslides are considered as one of the main threats to European soils. The term landslide includes a wide variety of phenomena and can be generally defined as a movement of a mass of rock, debris, or earth down slope, under the influence of gravity. The range of landslide events is particularly diversified, encompassing small rock falls as well as large movements of soil involving millions of cubic meters of material, and their evolution in terms of velocity spans over fourteen orders of magnitude. Starting from these considerations, the work here presented aims to introduce new methodologies for landslide risk management and early warning purposes. In particular, the main objective of this thesis involves the automatic analysis of monitoring data in order to identify the occurrence of potentially critical events and consequently disseminate appropriate alert messages. After a presentation of the main theoretical elements on which the work is based, the thesis explores the advantages of automatic monitoring systems based on Internet of Things technologies. The Chapter focuses especially on the improved performances of these techniques in term of ability to achieve high sampling frequencies and integrate different sensor typologies for a multi-parameter monitoring approach. The following chapters represent the core of the research project, focusing on the algorithms developed for the identification and assessment of potentially critical landslide events. Chapter 4 describes a criterion based on the analysis of displacement monitoring data to determine the presence of an accelerating trend. The algorithm follows a multi-level structure for the analysis in order to define the beginning of the acceleration phase, together with the dataset displaying this pattern. The following Chapter introduces a classification process intended to be applied to the outcome of the previous algorithm. The methodology relies on the evaluation of three different parameters, dependent of the dataset features and the results deriving from failure forecasting analyses, to assess the alert level for the identified event. Chapter 6 introduces two different procedures to define alert thresholds starting from available monitoring data. The first methodology is based on the forecasting model known as Inverse Velocity Method and integrates a normalization operation applied to monitoring outcomes, with the objective of providing a dimensionless parameter for the definition of generalized threshold values. The second algorithm introduces the concept of equivalent displacement, defined as the displacement generated in a time interval equal to the one showed by the potentially critical event. The process aims to compare the displacement produced by the event of interest with equivalent displacements referred to previously sampled data, to determine if the observed movement deviates significantly from the trend displayed by already available measurements. Finally, Chapter 7 entails the aspect related to the dissemination of relevant information obtained from the previously described methodologies. For this purpose, two different algorithms were realized exploiting the Matlab Report Generator™ toolbox. The first one is dedicated to the creation of a periodic monitoring report, including all relevant outcomes recorded over a predefined time period by instrumentation installed on site. The second one is a 1-page bulletin, generated in correspondence of the identification of a critical landslide event and delivered in near-real time for early warning applications.