Avaliação do modelo recorrente de regressão de aceleração de pico do terreno com base em dados de uma mineradora do interior de São Paulo, Brasil

Abstract

Seismicity is intrinsic to the existence of our planet, so the understanding of seismology is essential for humanity. Ground motion models (GMM) seek to estimate the vibration at a given point from seismic source data and its distance. Among the many applications, one example is the design of earthquake reinforced structures. In the seismological literature, GMMs that consider a linear multiple regression model from the logarithm of peak ground acceleration (PGA) are recurrent, being estimated from the logarithms of seismic power and distance from the source. The objective of this study was to evaluate statistical aspects of this approach, a perspective commonly neglected in the specific literature. Five (05) models were developed: two of simple linear regression (PGA as a function of power and distance separately) and three of multiple linear regression (with data with and without logarithmic transformation). The results suggested that, for the considered data (one year period of a microseismic monitoring system of an open pit in São Paulo - Brazil), the assumptions of normality and homoscedasticity of the residues were not respected by any model. The adjusted determination coefficients did not exceed 16%, a low value, but possibly explained by: (i) lack of distinction of seismic sources in the database, (ii) heterogeneity of the geology around the pit, (iii) failures in the microseismic monitoring system and (iv) limitations in the statistical approach. Therefore, such model widely used in Seismology, for the employed database, was not statistically valid. A non-linear approach is suggested as an alternative to solve the issue.