Time Series Analysis for Predicting the Occurrences of Large Scale Earthquakes
Amei Amei, Wandong Fu, Chih-Hsiang Ho

Abstract
Earthquakes that have occurred worldwide during the period of 1896 to 2009 with magnitude of 8.0 or greater on the Richter scale are assumed to follow a Poisson process. Several autoregressive integrated moving average (ARIMA) models with different time steps are proposed to predict the occurrences of large scale earthquakes by fitting the model with a sequence of empirical recurrence rates (ERRs) time series. The last five or ten data points are used as prediction sets to check the predictive ability of the candidate models developed by the time series modeling techniques. For a full scale forecast, the best fitted model predicts a total number of 12 large scale earthquakes in the next 6 years worldwide. The application of ERR based ARIMA models to long-term earthquake prediction not only serves as a linking bridge between point processes and the classical time series but also extends the usage of statistical methods to a wide area of natural disaster predictions.

Full Text: PDF