Metodologías para la evaluación de peligrosidad a los deslizamientos inducidos por terremotos

Abstract

This thesis focuses on the use of stochastic modelling methodology to assess earthquake-triggered landslide hazard and susceptibility. Due to the fact that it is currently impossible to predict these phenomena early enough to take action, the most effective way to mitigate the risk of landslides is the hazard assessment line and the adoption of preventive policies. Landslides can be induced by various causes, which act as triggers. The most common are heavy rains and earthquakes, which can be considered triggers independently, although it has been proven that some earthquakes have triggered more landslides in areas previously affected by heavy rains. Therefore, although this thesis focuses on the assessment of landslide hazard associated with earthquakes, the effect of precipitation will also be taken into account. As with the majority of natural hazards, the study of landslides requires a mathematical model for the evaluation and analysis of the phenomenon's probability of occurrence in a particular region. In recent years, the scientific community has worked to find the model that best fits the reality of this phenomenon, but this is a difficult and complicated task mainly because landslides are complex phenomena that involve a large number of interacting parameters, such as terrain morphology, geology, precipitation levels, and the seismicity and tectonics of the region, among others. The central objective of this thesis is to investigate the mathematical modelling of landslide hazard at the regional scale. In order to evaluate the phenomenon, it is important not only to define a mathematical conceptual model, but also its computational implementation. The applications of stochastic and heuristic methods have increased in recent years, mainly due to the increased capabilities of computers and the widespread use of geospatial tools such as Geographic Information Systems (GIS). Currently, the application of statistical techniques together with a GIS provides important tools for creating interactive queries of the information, geospatial analysis, as well as the representation and visualization of geographic data, thereby supporting the hazard mapping and thus the decision-making in many fields, particularly those concerning natural hazards. This thesis combines these techniques and tools to make a contribution to natural hazard assessment. To reach the proposed objective, a comprehensive description of the techniques and tools available was compiled by analyzing and categorizing the different methods used in the scientific literature: heuristics, stochastics, artificial neural networks, etc. Since landslides, like other natural phenomena, are characterized by complex and sometimes chaotic behaviour, both linear and non-linear mathematical models should be used to assess landslide hazard. In this thesis, stochastic techniques were investigated, mainly those of logistic regression and neural networks, with focus on the assessment of landslide hazard in wide areas. These techniques were applied to the El Salvador earthquake scenario of 13 January 2001, which provides a significant amount of information to contrast the two models. The results of this study were compared with our predictions for both models, with the logistic regression technique yielding a value of 89.4 percent and the artificial neural networks model a value of 95.1 percent. Finally, earthquake-triggered landslide hazard models were developed for the Central American country of El Salvador, for the specific scenario of a past event. In addition, a probabilistic model of movement, in terms of peak acceleration (PGA) with a return period of 475 years, was developed. This movement, which is generally considered in the design of regulations for conventional structures, could be applicable in the prediction of future occurrences of this phenomenon. This should be useful for detecting dangerous areas for urban and territorial planning purposes.