Data mining of atmospheric parameters associated with coastal earthquakes

Earthquakes are natural hazards that pose a serious threat to society and the environment. A single earthquake can claim thousands of lives, cause damages for billions of dollars, destroy natural landmarks and render large territories uninhabitable. Studying earthquakes and the processes that govern their occurrence, is of fundamental importance to protect lives, properties and the environment. Recent studies have shown that anomalous changes in land, ocean and atmospheric parameters occur prior to earthquakes. The present dissertation introduces an innovative methodology and its implementation to identify anomalous changes in atmospheric parameters associated with large coastal earthquakes. Possible geophysical mechanisms are discussed in view of the close interaction between the lithosphere, the hydrosphere and the atmosphere. The proposed methodology is a multi strategy data mining approach which combines wavelet transformations, evolutionary algorithms, and statistical analysis of atmospheric data to analyze possible precursory signals.; One dimensional wavelet transformations and statistical tests are employed to identify significant singularities in the data, which may correspond to anomalous peaks due to the earthquake preparatory processes. Evolutionary algorithms and other localized search strategies are used to analyze the spatial and temporal continuity of the anomalies detected over a large area (about 2000 km2), to discriminate signals that are most likely associated with earthquakes from those due to other, mostly atmospheric, phenomena. Only statistically significant singularities occurring within a very short time of each other, and which tract a rigorous geometrical path related to the geological properties of the epicentral area, are considered to be associated with a seismic event.; A program called CQuake was developed to implement and validate the proposed methodology. CQuake is a fully automated, real time semi-operational system, developed to study precursory signals associated with earthquakes. CQuake can be used for the retrospective analysis of past earthquakes, and for detecting early warning information about impending events. Using CQuake more than 300 earthquakes have been analyzed. In the case of coastal earthquakes with magnitude larger than 5.0, prominent anomalies are found up to two weeks prior to the main event. In case of earthquakes occurring away from the coast, no strong anomaly is detected. The identified anomalies provide a potentially reliable mean to mitigate earthquake risks in the future, and can be used to develop a fully operational forecasting system.