SourceSpec – Earthquake source parameters from S-wave displacement spectra

This is a pretty big release coming after several months of work on a large dataset of more than 5000 events in Mayotte.

Please read through the changelog to discover all the improvements and new features.

You will need to update your old config files via source_spec -U CONFIG_FILE_NAME

Note that v1.5 is no more compatible with Python 2!

• Input/output:
  • Write output files into a subdirectory of OUTDIR, whose name is the event id
  • Support for HYPOINVERSE-2000 output files
  • Removed autodetection of hypo71 file paths (specific to CRL case)
  • SQLite output: added radiated energy, weighted averages, errors on parameters, number of observations, hypocentral location and origin time
  • Removed -C argument to apply station correction to spectra. Now spectra are automatically corrected if residuals_filepath is specified in the configuration file
  • Save an additional event parameter to output files: average quality factor
  • Save additional station parameters to output files: source radius, Brune stress drop, source radius, quality factor
  • Mark outliers in .out file and in html report
  • Colored console output for log messages! (Not supported on Windows)
• Processing:
  • New parameter for setting the width of the spectral smoothing window in terms of frequency decades: spectral_smooth_width_decades
  • Compute spectral weights after spectral correction (when a station residuals file is specified via residuals_filepath)
  • Removed configuration parameter trace_format
  • New configuration parameter sensitivity to provide a constant sensor sensitivity (flat response curve), which ovverides any response curve provided in metadata
  • New parameter for manually specifying trace units: trace_units (defaults to auto)
  • New approach for trace clipping detection (requires just one configuration parameter, named clip_max_percent)
    • Check for trace clipping only in the processing window
    • Use histogram of samples to detect clipping
  • Fix for wrong component used for 'SV' spectra (#3)
• Inversion:
  • New config option: Mw_0_variability. Allowed variability around Mw_0 during the main inversion. Previously hardcoded to 0.1
  • New inversion methods for grid sampling:
    • grid search (very slow!)
    • importance sampling of the misfit grid using k-d tree (faster, but less accurate)
  • Fix for Basin-hopping algorithm not running
• Post-Inversion:
  • New set of post-inversion parameters to reject certain inversion results, per-station: pi_fc_min_max, pi_t_star_min_max, pi_bsd_min_max, pi_misfit_max
  • Reject inversion results when inverted fc is within 10% of fc_min or fc_max
  • Fix: use logarithmic error width for weighted logarithmic averages
    • previous way of computing weighted logarithmic averages was not correct!
  • Option to reject outliers using the IQR (interquartile range) method: parameter nIQR
  • Support for non symmetric error on station spectral parameters
  • Compute additional, per-station parameters: source radius, Brune stress drop and quality factor
  • Compute errors for all station parameters
  • Compute weighted averages for all event parameters (except radiated energy)
  • Compute spectral residuals using weighted average spectral parameters
• Plotting:
  • Source parameter box plots to evaluate parameter dispersion across stations and visually detect outliers
  • Misfit plot (2D and 1D) when using grid sampling
  • cartopy removed as installation dependency, since it is not easily installable via pip
  • Use GSHHS database to draw coastlines.
    • New config option: plot_coastline_resolution
  • Correctly show circles on maps with diagonal smaller than 1 km
  • Fix plotting map colorbar on Matplotlib 3.5
  • Make average and errorbar lines more visible on map colorbar
  • Fix for error on plotting fc map, when only one station is available
  • Fix trace plot scaling for traces with larger signal outside the plot window
  • Do not plot 'H' spectrum if there is only one instrument component (since it will coincide with the only component)
  • Plot uncorrected spectrum when station correction is used