Analysis of seismicity of central Luzon by pattern recognition

The distribution of large shallow earthquakes (magnitude ?5.5) in central Luzon, Philippines, related to the underthrusting of the Eurasian plate, is examined by the technique of pattern recognition. The objective of the study is (1) to forecast restricted zones of seismic activity, (2) to determine geological features which, in combination, are associated with the earthquakes in the environment of subduction, and (3) to determine whether more insight into regional seismicity can be gained by pattern recognition studies than by conventional seismicity studies. It appears from this analysis that large earthquakes occur (a) in a zone between the shallower portion of Manila Trench on its landward side and the belt of volcanoes and (b) on large strike-slip faults in the region such as Verde Island Passage fault and the Philippine fault. In these zones, large earthquakes occur at locations identified by large topographic relief in the vicinity and a high degree of fractured crust. In contrast to characteristics in California and central Asia, the intersection of faults nearby is not a major characteristic for large earthquakes in and near central Luzon. Topography and degree of fractured crust are important characteristics in all areas studied so far. Several earthquakes which are known to have occurred were considered unknown, and control experiments were carried out to determine whether the locations of these earthquakes would be identified as having potential for large earthquakes. The results of control experiments indicate that the pattern recognition technique is fairly successful in spatial forecasting if the known historical record is assumed almost complete for various major zones (85% in this case). It fails to forecast if information is totally lacking about earthquake activity on a tectonic feature. In the present study, a control experiment was able to correctly identify the locations of two of the largest earthquakes known to have occurred in the area. It was better able to identify locations of large earthquakes than a simple contouring of known earthquake locations would suggest.