Applications of Generalized Likelihood Ratio Tests to Seismic Event Identification
Final rept. 25 May 1990-25 Aug 1993
MISSION RESEARCH CORP SANTA BARBARA CA
Pagination or Media Count:
A summary of the work performed under this project is provided. It contains brief descriptions of statistical methods developed and data analysis performed during the early stages of this contract to improve yield estimation. Preliminary results of a study of discriminant transportability for Eurasia are also summarized. The main body of the report, however, focuses on development of statistical methods and their applications to seismic event identification problems relevant to monitoring a CTBT or NPT. Recently developed classification and outlier detection methods are applied to the problem of seismic event identification, particularly for regional events. The methods, based on the generalized likelihood ratio GLR, have considerable flexibility to rigorously treat a mixture of continuous and discrete features, normal or non-normal statistics, equal or unequal covariance matrices, and different distribution functions for different event types. Statistical procedures are also presented to test the validity of appropriate assumptions and, if necessary, transform the data to simplify implementation of the outlier and classification methods to ensure that the necessary assumptions hold. Methods are also presented and applied to select the optimal subset of discriminants and to treat missing data values robustly. This collection of methods provides a robust statistical framework to identify seismic events and to estimate classification probabilities rigorously under extremely general conditions.
- Nuclear Weapons