Quakefinder: A Scalable Data Mining System for Detecting Earthquakes from Space

W e present an application of novel massively parallel datamining techniques to highly precise inference of important physical processes from remote sensing imagery. Specifically, we have developed and applyed a system, Quakefinder, that automatically detects and measures tectonic activity in the earth’s crust by examination of satellite data. W e have used Quakef inder to automatically map the direction and magni tude of g round displacements due to the 1992 Landers earthquake in Southern California, over a spatial region of several hundred square kilometers, at a resolution of 10 meters, to a (sub-pixel) precision of 1 meter. This is the first calculation that has ever been able to extract a rea-mapped information about 2D tectonic processes at this level of detail. W e outline the ar-b:+^..C __-^ -P LLn..-,.-c--l--_.A.^-L----1 ..-^LUlbCLblllt: “I cut5 yu~l‘auuuta SyaLanl, ucucu up”” a combinat ion of techniques drawn from the fields of statistical inference, massively parallel comput ing and global optimization. W e confirm the overall correctness of the procedure by compar ison of our results with known locations of targeted faults obtained by careful and t ime-consuming field measurements. The system also performs knowledge discovery by indicating novel unexpla ined tectonic activity away from the primary faults that has never before been observed. W e conclude by discussing the future potential of this data mining system in the broad context of studying subtle spatio-temporal processes within massive image