Uso de métodos probabilísticos na análise de estabilidade de taludes. São Luís, Maranhão, Brasil Use of probabilistic methods in slope stability analysis. São Luís, Maranhão, Brasil

Resumen

The present research aimed to generate and compare the susceptibility maps for the occurrence of shallow landslides in terms of probability of failure on the Sao Luis Island, obtained by applying the FOSM (First Order Second Moment) and Monte Carlo probabilistic methods. The differences found between the two approaches are analyzed. The methodology consisted of three main steps. The first step is related to the organization of a database of environmental information in a georeferenced format using the Geographic Information Systems platform SPRING. These data corresponded to the topography and the soil map, from which it was possible to identify the geotechnical parameters necessary for the analysis of slope stability in relation to each pedological class. The spatial information was exported from the SPRING platform in raster format and served as input data for the computational routines of stability analysis. The second step refers to the computational implementation of the probabilistic methods associated with the infinite slope stability model. The FOSM method was implemented using the Matlab program, while the FORTRAN language was used for the Monte Carlo method. The independent variables of the slope stability model considered to be random variables were cohesion and friction angle. Initially, it was considered that both variables had a normal distribution for the application of the FOSM and Monte Carlo methods, in order to compare the results of the different probabilistic approaches. Additionally, the Monte Carlo method was also simulated using a lognormal distribution to model cohesion, in a more coherent manner with the actual statistical behavior of that variable, while the friction angle remained modeled by a Gaussian distribution. Different numbers of simulations were proposed for the Monte Carlo Method, in order to verify the variation of the rupture probability in these scenarios. The final stage corresponded to the creation of maps for each configuration of the analysis and the consequent evaluation of the observed variations in terms of probability of failure. In general, the probabilities of failure obtained for both probabilistic methods are similar in the case where both independent variables are normally distributed. When cohesion is represented by the lognormal distribution in the Monte Carlo method, null values were obtained for the probability of failure, even for high numbers of simulations. In conclusion, it can be said that the change in the representation of the cohesion distribution and the quality of topographic information affect the results of probability of failure. The use of topographic information with higher spatial resolution would define more accurate susceptibility measures and would better reflect the action of cohesion in the form of lognormal distribution in the results.