An earthquake is a natural phenomenon that occurs as a consequence of Earth's internal dynamics. It originates deep under the planet's surface and cannot be predicted with our current knowledge. Thesis addresses a network analysis approach to acquire new knowledge about the characteristics and development of seismic activity over time. We implement and compare various network models based on time and space interactions between earthquakes and on the assumption of self-similarity in seismic activity in a selected geographic area. Using networks that are generated in multiple consecutive time windows, we extract a feature set and present its changes over time in a time series. Finally, ARIMA model for time series prediction is used to verify if it is possible to predict characteristics of seismic activity in the future.
Analysis of generated networks and time series shows that the majority of used network models produce relevant networks that return reliable and predictable response in a time series. However, using ARIMA model to predict new data points in a time series turns out to be insufficient, especially in time intervals when the strong earthquake occurs.
|
