Feature-based Magnitude Estimates for Small Earthquakes in the Yellowstone Region

In this study we use features derived from short-duration waveforms segments and event location information in support vector machines to produce local magnitude estimates that are consistent with the existing University of Utah Seismograph Stations (UUSS) earthquake catalog for the Yellowstone Volcanic Region.

This code implements the methods and produces the figures described in Armstrong et al., which has been accepted at BSSA in May 2025.

Magnitudes are produced using station and distance corrections and code from Holt et al., (2022)

Armstrong et al., 2025

Data

• data/
  • catalogs/yellowstone.events.2024.csv - UUSS events in the Yellowstone region through 2023
  • catalogs/yellowstone.arrivals.gains.2024.csv - Arrivals associated with the events, includes station gain information
  • catalogs/yellowstone.events.ypml-v5.2024.csv - Info for events with a YP21 M_L computed
  • catalogs/yellowstone.amps.ypml-v5.2024.csv - Arrival info the YP21 events
  • features/p_features.2024.csv - waveform features for UUSS P arrivals
  • features/s_features.2024.csv - waveform features for UUSS S arrivals
• Format UUSS catalogs for use in magnitute recalibration using reformatAmpCatalogs.ipynb
• Features created using mlmodels.features
  • P features generated with pMagnitude.py
  • S features generated with sMagnitude.py
• Train and test splits made using make_feature_splits.py

Methods

• Recursive Feature Elimination
  • Main implementation in feature_selection.CustomRFECV
  • Run for P arrival data in rfecvAllPStations.ipynb
  • Run of S arrival data in rfecvAllSStations.ipynb
• Final SVR models
  • P models produced in trainPModelsWithSelectedFeatures.ipynb
  • S models produced in trainSModelsWithSelectedFeatures.ipynb
• Single Gradient Boosted Tree models
  • using all P arrivals in allPStationsSimpleBoostedTree.ipynb
  • using all S arrivals in allSStationsSimpleBoostedTree.ipynb
• Examine event location feature importance in examineDistanceCorrections.ipynb
• Analysis of WY.YDC P model in notebooks/results/YDC

Figures

• Figure 1 produced in featureStationMaps.ipynb
• Figure 2 produced in ModelResultsP.ipynb
• Figure 3 produced with Adobe Illustrator in psuedoCode_v2.ai
• Figure 4 produced in RFE_example_fig.ipynb
• Figure 5 produced in combinePandSFeatureSelectionHeatmaps.ipynb
• Figure 6 produced in summaryFigsR2Boxplots.ipynb
• Figure 7 produced in summaryFigsResiduals.ipynb
• Figure 8 produced in combinePandSPredictions.ipynb
• FIgure 9 produced in summaryFigsResiduals.ipynb