Photos placed in horizontal position with even amount of white space between photos and header Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL SAND NO XXXXP Kristin Phillips-Alonge, Hunter Knox, Curtis Ober SPE5 Far-field Waveform Predictions Using Full Waveform Inversion
2 Introduction Source time functions (STF) estimates for SPE2, SPE3, and SPE4prime used for prediction of STF for SPE5 SPE science questions addressed: – Using new physics-based numerical methods to characterize explosive source from observable far-field signals. – What is the impact of damage, shot size, depth of emplacement, and geology on signals and how to take changes into account in predictions and modeling SPE End Goals Improved physics-based, predictable models for the source time function improves our nuclear monitoring capabilities by allowing us to better estimate far-field yield and other properties of the explosive source and allows for differentiation from other possible seismogenic sources.
3 Full waveform inversion for Source Time Functions Code: based on Discontinuous Galerkin Method Physics: Acoustic and Elastic 2D Source Inversions: based on first four geophones from each geophone array 3D Source Inversions: using first four geophones from all arrays, to be presented at SSA Time integration: fourth-order Runge-Kutta Peak frequency: 20Hz Fluxes: Lax Friedrichs and Riemann Boundary conditions: Free surface at the top of mesh, non-affine elements fit to topography A B
4 Source inversions: Shots and physics 1 comp. Acoustic 1 comp. Elastic 2 comp. Acoustic 2 comp. Elastic 3 comp. Acoustic 3 comp Elastic SPE2 ✔✔ SPE3 ✔✔ SPE4’ ✔✔✔✔✔ In progress Yield used for SPE5:5080 kg TNT equivalent
5 Do small velocity perturbations impact inversion results? 1.Shallow velocity data based on AWD surveys along arrays (“AWD”) 2. Based on 3D tomography (Leiph Preston, SNL), 2 km spatial resolution (“Leiph”) 3.Based on ambient noise (LLNL, Pitarka et al. 2015, Wagoner, 2014) (“LLNL”) 4.Simple homogenous half space with granitic properties (“Granite”) A BA B A B A B
6 Comparison of Acoustic results for different models: Station fits Vertical component Late time fits affected by: - 2D fit to 3D data - Algorithm preferentially fits highest amplitudes
7 Comparison of Acoustic results for different models: Time samples AWD Granite Leiph LLNL time 0.5 sec time 1.0 sec time 1.5 sec Vertical particle velocity profiles
8 SPE4prime Elastic source inversions, 1 comp (Z only), 2 comp (Z and R) Station fits, vertical velocity AWD model
9 SPE4prime Elastic Source Time Functions AWD model
10 SPE4prime Comparison of Acoustic and Elastic physics, Z comp. only AWD Leiph Granite LLNL Elastic Acoustic Elastic Acoustic AWD model Vertical particle velocity profiles, 1.0 sec
11 STFs for SPE2, SPE3 to SPE4prime, predictions for SPE5 v4 v3 v2 v1 AWD model
12 Predictions for SPE5 v1 v2 v3 v4 v1 v2 v3 v4 AWD model
13 v1v2 v3 v4 Vertical particle velocity profiles for different SPE5 source time functions AWD model
14 Conclusions Future work: - More runs in 3D, medium inversions to increase complexity and accuracy of velocity model -Possibility of using more advanced physics: e.g. Anelastic (including attenuation), including anisotropy - Velocity model appears to impact the amplitude of the source time functions but not the overall shape. The main impact on the source time functions is the shot, stations and components fit and the physics used. -Post-SPE5 assessment will allow for narrowing down the best method and model to use for future predictions Post-SPE-5 prediction assessment -Compare observed and predicted waveforms -Invert data for actual source -Make improvements to models and methods to obtain predicted source time functions and waveforms closer to actual observed SPE5 observations Next steps for SPE-6, phase II DAG, possible improvements -Predictions for SPE-6 will benefit from lessons learned after SPE5 predictions. -Phase II DAG measurements will benefit from high quality surficial velocity data measurements from Seismic Hammer experiments -Good data available for SPE4prime, would have helped to have all three velocity components for geophone arrays for previous shots.
15 Pitarka, A., R.J. Mellors, W.R. Walter, S. Ezzedine, O. Vorobiev, T. Antoun, J.L. Wagoner, E.M. Matzel, S.R. Ford, A.J. Rodgers, L. Glenn, & M. Pasyanos (2015). Analysis of Ground Motion from An Underground Chemical Explosion. Bulletin of the Seismological Society of America, Vol. 105, No. 5, pp., Oct. 2015, doi: / Wagoner, J.L. (2014). Working toward a site-specific geomodel, Nevada National Security Site, RMR2014 Review of Monitoring Research for Ground-based Nuclear Explosion Monitoring Technologies, Albuquerque, New Mexico, 18 June 2014 References
Backup Slides 16
17 Acoustic Source Time Functions, AWD model
18 Source time functions – Acoustic (3 comp) vs Elastic, AWD model, SPE4prime