دانلود رایگان مقاله انگلیسی کاهش تاخیر فرآیند برآورد پارامترهای زلزله در زمان واقعی - کاربردهای درخت KD برای اخطار زلزله - الزویر 2018

Abstract

Earthquake parameter estimations using nearest neighbor searching among a large database of observations can lead to reliable prediction results. However, in the real-time application of Earthquake Early Warning (EEW) systems, the accurate prediction using a large database is penalized by a significant delay in the  processing time. We propose to use a multidimensional binary search tree (KD tree) data structure to organize large seismic databases to reduce the processing time in nearest neighbor search for predictions. We evaluated the performance of KD tree on the Gutenberg Algorithm, a database-searching algorithm for EEW. We constructed an offline test to predict peak ground motions using a database with feature sets of waveform filter-bank characteristics, and compare the results with the observed seismic parameters. We concluded that large database provides more accurate predictions of the ground motion information, such as peak ground acceleration, velocity, and displacement (PGA, PGV, PGD), than source parameters, such as hypocenter distance. Application of the KD tree search to organize the database reduced the average searching process by 85% time cost of the exhaustive method, allowing the method to be feasible for real-time implementation. The algorithm is straightforward and the results will reduce the overall time of warning delivery for EEW.

 Discussion and Conclusion

 In this study, we evaluated the viability of earthquake fingerprint searching methods for EEW, using database structure to reduce searching time for large databases. Specifically, we evaluated the GbA as an example of the EEW fingerprint search algorithm. We found that database size is a critical factor in providing  reliable predictions of ground motion (PGA, PGV, PGD) and source parameters (magnitude and hypocenter distance) for EEW. We also present the KD tree approach to reduce the searching time, so that large database searching is feasible for real-time implementations in EEW. By empirical validation, we demonstrated that the searching time using KD tree can be approximately 85% less than the exhaustive approach for the GbA EEW earthquake database. (Strauss et al, 2017) has studied extensively on the cost-benefit effects of a warning system in the United States; the study has shown that the number of injuries from earthquakes can be reduced by more than 50% if EEW can provide timely and accurate alerts. One of the potential applications of the database searching method is to directly estimate peak ground motions from the observed ground motions for any given site in real-time seismology application such as EEW; it avoids the multi-step modeling errors that could be accumulated through source parameter estimation and the ground motion attenuation relationship, since the final errors can lead to significant uncertainties in the final shaking information. Ideally, the goal of EEW is to serve as an alarm for severe ground shaking in real-time rather than source characterization. The fingerprint searching methodology could also be extended to tackle other challenges in EEW, such as event detection (i.e. earthquake/noise discrimination).