Algorithms

  • Analyst methods are modular so that results from other processing tools (e.g., Bloodhound, Cardinal, PMCC) can be used in later analysis steps and vice versa.

  • Algorithms are written to be data source agnostic so analysis can be performed regardless of data source once IO method is understood

    • Waveform analysis utilizes an ObsPy Stream object for data ingestion
    • Detection information is stored in .json or ascii .dat files that can be ingested using functions in infrapy.likelihoods

Station Level Processing

Array Processing

  • Beamforming estimates parameters of coherent signals
  • Capabilities include methods to characterize transients as well as persistent signals
  • Transient signals are identified using standard Bartlett beaming
  • Persistent signals can be investigated using Minimum Variance Distortionless Response (MVDR) or MUltiple Signal Classification (MUSIC) algorithms

Detection

  • Adaptive Fisher statistics determine when to declare a detection

Network Level Processing

Association

  • Events are identified using a pair-based Bayesian algorithm that defines the association between detection pairs from their joint-likelihood and identifies events via hierarchical clustering analysis
  • Current implementation utilizes only spatial and temporal coincidence, but additional detection information can further improve event identification
  • Evaluation using a synthetic data set shows some mixing of spatially similar events poorly resolved by network geometry and occasional inclusion of “noise” detections in event clustering

Localization

  • Event analysis using the Bayesian Infrasonic Source Localization (BISL) methodology to estimate both the spatial location of the event as well as the origin time with quantified uncertainty
  • Preliminary analysis of the back projections can be used to define the spatial region of interest or it can be specified by the analysis
  • Analysis identifies the maximum posteriori solution
  • The marginalized spatial distribution is approximated as 2d-normal to define 95 and 99% confidence ellipse bounds
  • The marginalized temporal distribution is analyzed to identify the exact 95 and 99% confidence bounds
  • Likelihood definitions relating detection parameters to spatial and temporal source characteristics are shared between association and localization analysis for consistency

Source Characterization

  • Detections are analyzed to compute the spectral amplitude (magnitude of the Fourier transform) for the arrival waveform

  • Transmission loss statistics are utilized to estimate the infrasonic spectral amplitude at a location near the source from the various observations

  • The near-source spectral amplitude estimate is combined with a source model (currently the Kinney & Graham scaling laws and Friedlander blastwave model) to map the spectral amplitude to blastwave yield