QUICK DESIGN GUIDE Progress in Signal-Based Bayesian ... · (copy and paste the link into your web...
Transcript of QUICK DESIGN GUIDE Progress in Signal-Based Bayesian ... · (copy and paste the link into your web...
QUICK DESIGN GUIDE (--THIS SECTION DOES NOT PRINT--)
This PowerPoint 2007 template produces a 42”x90” professional poster. It will save you valuable time placing titles, subtitles, text, and graphics. Use it to create your presentation. Then send it to PosterPresentations.com for premium quality, same day affordable printing. We provide a series of online tutorials that will guide you through the poster design process and answer your poster production questions. View our online tutorials at: http://bit.ly/Poster_creation_help (copy and paste the link into your web browser). For assistance and to order your printed poster call
PosterPresentations.com at 1.866.649.3004
Object Placeholders
Use the placeholders provided below to add new elements to your poster: Drag a placeholder onto the poster area, size it, and click it to edit. Section Header placeholder Use section headers to separate topics or concepts within your presentation. Text placeholder Move this preformatted text placeholder to the poster to add a new body of text. Picture placeholder Move this graphic placeholder onto your poster, size it first, and then click it to add a picture to the poster.
RESEARCH POSTER PRESENTATION DESIGN © 2011 www.PosterPresentations.com
QUICK TIPS (--THIS SECTION DOES NOT PRINT--)
This PowerPoint template requires basic PowerPoint (version 2007 or newer) skills. Below is a list of commonly asked questions specific to this template. If you are using an older version of PowerPoint some template
features may not work properly.
Using the template
Verifying the quality of your graphics Go to the VIEW menu and click on ZOOM to set your preferred magnification. This template is at 50% the size of the final poster. All text and graphics will be printed at 200% their size. To see what your poster will look like when printed, set the zoom to 200% and evaluate the quality of all your graphics before you submit your poster for printing. Using the placeholders To add text to this template click inside a placeholder and type in or paste your text. To move a placeholder, click on it once (to select it), place your cursor on its frame and your cursor will change to this symbol: Then, click once and drag it to its new location where you can resize it as needed. Additional placeholders can be found on the left side of this template. Modifying the layout This template has four different column layouts. Right-click your Mouse on the background and click on “Layout” to see the layout options. The columns in the provided layouts are fixed and cannot be moved but advanced users can modify any layout by going to VIEW and then SLIDE MASTER. Importing text and graphics from external sources TEXT: Paste or type your text into a pre-existing placeholder or drag in a new placeholder from the left side of the template. Move it anywhere as needed. PHOTOS: Drag in a picture placeholder, size it first, click in it and insert a photo from the menu. TABLES: You can copy and paste a table from an external document onto this poster template. To make the text fit better in the cells of an imported table, right-click on the table, click FORMAT SHAPE then click on TEXT BOX and change the INTERNAL MARGIN values to 0.25 Modifying the color scheme To change the color scheme of this template go to the “Design” menu and click on “Colors”. You can choose from the provide color combinations or you can create your own.
© 2011 PosterPresenta.ons.com 2117 Fourth Street , Unit C Berkeley CA 94710 [email protected]
Student discounts are available on our Facebook page. Go to PosterPresentations.com and click on the FB icon
Progress in Signal-Based Bayesian Monitoring
Introduc4on
David A. Moore1, Kevin Mayeda1, Stephen C. Myers2, Min Joon Seo1, Stuart J. Russell1 University of California, Berkeley1 and Lawrence Livermore Na.onal Laboratory2
Signal Envelope Model • Global seismic monitoring for the Comprehensive Nuclear-Test-Ban Treaty (CTBT) aims to recover the time, location, depth, and magnitude for all seismic events in the magnitude range of interest. • Data from the International Monitoring System (IMS) are processed in real time at the International Data Centre (IDC) in Vienna. Our goal is to improve the sensitivity and accuracy of automated processing at IDC.
Signal-‐Based vs. Detec4on-‐Based Monitoring
Signal-‐based Bayesian monitoring CONCLUSIONS • Bayesian methods are a powerful and promising approach to monitoring problems. In particular, signal-based monitoring improves on traditional methods by making full use of the available data, with potential for significant performance improvements. • The SIG-VISA prototype system includes an empirically-calibrated, physics-based model of observed waveform envelopes. • Gaussian Process regression enables learning of flexible, data-driven probabilistic models of spatial parameters. • Waveform matching / correlation effects can be recovered from Bayesian inference in a spatially conditioned model, enabling localization from a single detection in some cases.
REFERENCES Arora, Nimar S., Stuart J. Russell, Paul Kidwell, and Erik Sudderth, “Global seismic monitoring as probabilistic inference,” In NPIS 23, MIT Press, 2011. Harris, D.B, “A wave-form correlation method for identifying quarry explosions,” Bull. Seism. Soc. Am., 81, 2395–2418, 1991. Harris, D. B. and D. A. Dodge, “An autonomous system for grouping events in a developing aftershock sequence,” Bull. Seismol. Soc. Amer ., vol. 101, no. 2, pp. 763–774, 2011. Mayeda, K., Hofstetter, A., O'Boyle, J.L., and Walter, W.R. (2003). Stable and transportable regional magnitudes based on coda-derived moment-rate spectra. Bulletin of the Seismological Society of America, 93(1), 224-229. Myers, S.C., G. Johannesson, and W. Hanley (2007). A Bayesian hierarchical method for multiple-event seismic location. Geophys. J. Int., 171, 1049-1063. Schaff, D. P., Bokelmann, G. H. R., Ellsworth, W. L., Zanzerkia, E., Waldhauser, F., & Beroza, G. C. (2004). Optimizing Correlation Techniques for Improved Earthquake Location. Bull. Seism. Soc. Amer, 94(2), 705-721. Slinkard, Megan E., Dorthe B. Carr, Stephen L. Heck, and Christopher J. Young (2011). Towards an automated waveform correlation detector system. In Monitoring Research Review, Tucson, Arizona. Thorbjarnardottir, B. S. and J. C. Pechmann (1987). Constraints on relative earthquake locations from cross-correlation of waveforms. Bull. Seismol. Soc. Am., 77, 1626–1634.
ACKNOWLEDGEMENTS We gratefully acknowledge the support of DTRA for this work; also the support of CTBTO for the work on NET-VISA and the technical cooperation of CTBTO personnel including Ronan Le Bras and Jeff Given. We thank Sheila Vaidya of LLNL and Ola Dahlman, former Chair of CTBTO Working Group B, for their moral support and encouragement in this project. We also thank Megan Slinkard and Regina Eckert of Sandia National Labs for sharing with us their data on correlated arrival waveforms.
SIG-VISA is a probabilistic generative model of seismic event origins, propagation, and observed waveform envelopes, including event signals along with station background noise:
• NET-VISA is a detection-based Bayesian monitoring system whose performance is limited by the classical, bottom-up, threshold-based detections algorithms used in station processing. It misses about 2-3 times fewer events than SEL3. • SIG-VISA, a signal-based system, uses generative models that span the range from events to waveform traces. This approach has several qualitative advantages over NET-VISA, with the potential for significantly improved sensitivity and localization performance.
Blue dots and triangles are primary seismic stations.
events
detections
waveform signals
SEL3
sta.on processing
NET-‐VISA
SIG-‐VISA
model inference model inferenc
e
Bayesian monitoring with a generative approach Pθ(world) describes prior probability for what is (events) Pφ(signal | world) describes forward model (propagation, measurement, etc.) Detection-based Bayesian monitoring: P(world | f (signal)) ∝ Pφ(f (signal) | world) Pθ(world) where f (signal) = set of all detections Signal-based Bayesian monitoring: P(world | signal) ∝ Pφ(signal | world) Pθ(world)
• The signal model encodes a distribution over waveform envelopes at each station given parameters for all hypothesized events.
• Each observed envelope is modeled with a shaped template combined with a stochastic modulation signal for each arriving phase, and a background noise process at each station:
• SIG-VISA uses a paired-exponential envelope template whose arrival time, amplitude, and decay rate depend on event magnitude and the event->station path. Template fits to historical signals are used to train predictive models for new observations.
LeJ: An observed envelope showing the P and S arrivals and subsequent coda decays. The red line indicates the template fit.
• Waveform shape is known to be highly repeatable across events with the same location and source mechanism (Thorbjarnardottir and Pechmann, 1987; Harris, 1991).
• SIG-VISA captures this effect by replacing the independently sampled modulation signal with a signal conditioned on the event location, causing nearby events to generate correlated waveforms:
observedEnvelope = template x exp(modula.on) + sta.onNoise
• This causes a statistical “waveform matching” effect to emerge from inference in the probabilistic model.
• Signals are represented parametrically as a sum of basis functions (e.g. Fourier basis), with coefficients modeled by a spatial Gaussian process.
• The current automated system (SEL3) detects 69% of real events and creates twice as many spurious events. • 16 human analysts find more events, correct existing ones, throw out spurious events, and generate LEB (“ground truth”). • Unreliable below magnitude 4 (about 1 kiloton).
Seismic Event
Travel 4mes
Sta4on 1 noise
Path-‐specific signal characteris4cs
Seismic Event
Travel 4mes
Sta4on 2 signal envelope
Sta4on 2 noise
Path-‐specific signal characteris4cs
Sta4on 1 signal envelope
0
0.5
1
1.5
AFI CTA DAV FITZ GNI
Med
ian Ab
solute Residual
Amplitude Predic4on (ver4cal component, 2-‐3Hz)
Baseline Linear GP
0
0.01
0.02
AFI CTA DAV FITZ LPAZ GNI
Med
ian Ab
solute Residual
Coda Decay Predic4on (ver4cal component, 2-‐3Hz)
Baseline Linear GP
Waveform Matching Path-‐Specific Characteris4cs
Observed envelope amplitudes and decay rates vary with the event source location, with effects from distance and depth as well as local variation in seismic structure.
We model these effects empirically using a nonparametric spatial Gaussian Process regression (kriging) model, which learns for each seismic station a probability distribution over template parameters for events at every point on the Earth:
Coda decay ra
te
The nonparametric model captures local variation, yielding improved predictions relative to a distance-dependent linear model:
Spatial model of coda decay rates observed by Charter Towers station. (red = sharper decay).
The spatial model enables inference from signal characteristics to possible event locations:
Prior location density
Synthe.c reference envelope (red) and a sampled envelope (blue) from a Gaussian-‐process genera.ve model for a nearby event loca.on.
The same reference envelope (red) and a sampled envelope (blue) from a distant event loca.on.
Fourier features for two nearby and correlated events (red, blue) vs. a distant event (green).
Posterior loca.on density of IMS event 4689462 from a single sta.on observa.on (MKAR) using a waveform matching model trained on data from 99 other events, shown in red. The posterior peak is 8.4km from the true event loca.on, marked by a green star.
• Station noise is modeled by an autoregressive process, retrained hourly at each station to capture regional and temporal variation in noise properties.
• Observed signals are modeled independently across a range of narrow frequency bands, automatically learning spectral characteristics and frequency-specific decay rates.
• Amplitude for each phase within each frequency band is predicted from event magnitude using a physics-based source model (Brune, 1971; Mueller and Murphy, 1971) and a path-specific learned transfer function (next column).
FUTURE WORK • Computationally efficient algorithms for training and probabilistic inference at global scale. • Model refinement based on geophysical expertise. • Dependent modeling of array station elements. • Putting it all together: a practical, effective global monitoring system.
Correlated event waveforms (IMS events 4686108, 4689462, correla.on .60).
Par.al posterior densi.es condi.oned on one (leb) and five (right) Fourier coefficients.
Posterior location density given coda decay rate
Coda decay rate likelihood
Right: A synthe.c envelope generated from the template with an autoregressive modula.on process.