DEVELOPMENT OF A STATISTICAL FRAMEWORK FOR EVALUATING THE IDENTIFICATION OF SEISMIC EVENTS

THE PRIMARY OBJECTIVE OF THE DARPA NUCLEAR MONITORING PROGRAM IS TO MONITOR SMALL OR CLANDESTINE UNDERGROUND NUCLEAR TESTS. THIS REQUIRES THAT WE ARE ABLE TO DETECT, LOCATE AND IDENTIFY DECOUPLED 1 KT EXPLOSIONS (EQUIVALENT TO MAGNITUDE 2.5). REGIONAL IDENTIFICATION OF SMALL SEISMIC EVENTS IS A COMPLEX PROBLEM THAT HAS NOT BEEN SOLVED, ALTHOUGH A NUMBER OF PROMISING METHODS HAVE BEEN DEVELOPED. TO MONITOR NUCLEAR PROLIFERATION EFFECTIVELY, BOTH SCIENTIFICALLY AND POLITICALLY, IT IS CRUCIAL TO HAVE A STATISTICAL FRAMEWORK THAT (1) CAREFULLY TREATS THE RANDOM FLUCTUATIONS IN SEISMIC SIGNALS, INCLUDING CORRELATIONS AMONG THE FEATURES USED AS DISCRIMINANTS; (2) IS ABLE TO QUANTIFY CLASSIFICATION PROBABILITIES, INCLUDING FALSE ALARM RATES; (3) IS CAPABLE OF IDENTIFYING NUCLEAR EXPLOSIONS IN REGIONS FOR WHICH RELEVANT NUCLEAR EXPLOSION TRAINING DATA DO NOT EXIST; AND (4) PROVIDES A RIGOROUS FRAMEWORK IN WHICH TO REPORT AND JUSTIFY THE RESULTS. IN THIS PROSOSAL, WE DISCUSS MANY OF THE IMPORTANT ISSUES THAT MUST BE ADDRESSED, AND A PLAN TO COMBINE PROMISING STATISTICAL METHODS AND REGION-SPECIFIC KNOWLEDGE. WE PROPOSE TO DEVELOP A MULTIVARIATE STATISTICAL PACKAGE TO MEET THESE OBJECTIVES, AND TO IMPLEMENT IT AS A SUBSYSTEM OF THE IMS. ANTICIPATED BENEFITS: THIS EFFORT WILL PROVIDE A CRITICAL REVIEW OF APPLICABLE STATISTICAL METHODS, AND A SUMMARY OF RELEVANT REGION-SPECIFIC DISTRIBUTIONAL INFORMATION. IT WILL DEMONSTRATE THE APPLICABILITY OF CERTAIN DISCRIMINATION METHODS, AND DESCRIBE HOW THEY CAN BE INCORPORATED IN A STATISTICAL PACKAGE, WHICH WILL INCLUDE ESTIMATION OF CLASSIFICATION PROBABILITIES. THIS WORK WILL ALSO BENEFIT OTHER IDENTIFICATION SYSTEMS, EITHER BY ADDRESSING UNRESOLVED STATISTICAL ISSUES, OR BY PROVIDING A FRAMEWORK TO COMBINE PROMISING ELEMENTS OF THOSE SYSTEMS.