The Statistical Analysis of the Earthquake Hazard for Turkey by Generalized Linear Models

Abstract

In this paper, 4863 earthquake data of magnitude 4.0 and greater from 1900 to 2014 are statistically analyzed for the earthquake hazard in Turkey. The magnitude-frequency relationship in earthquake risk analysis is often performed by Gutenberg-Richter model. With the use of this model, information about earthquake potential of any region can be obtained by previous data and by estimating parameters such as return periods and possibilities of their occurrence. In this study, the relationship between earthquake numbers and magnitudes is modelled with the Generalized Linear Models as an alternative to Gutenberg-Richter model. Generalized Poisson Regression model and Generalized Negative Binomial Regression models as Generalized Linear Models are utilized in the study. Generalized Poisson Regression model is found as the best model when considering the dispersion parameters and model selection criteria. Exceeding probabilities and return periods are calculated for the selected years depending on yearly average occurrence number of earthquakes estimated with the Gutenberg-Richter and Generalized Poisson Regression models. According to the results, Generalized Poisson Regression model can be employed for seismic risk modelling in Turkey.