Automated Weighting Schemes for DAS Data in Geophysical Inversion: A Case Study on Event Location

Distributed Acoustic Sensing (DAS) technology offers meter-scale spatial sampling of seismic wavefields, which improves our seismic monitoring capabilities. However, the resulting data volumes often complicate expert-driven analysis, such as weighting measurements in geophysical inverse problems to mitigate the influence of outliers. To address such difficulties in an automated manner, we focus on a representative problem in seismology, source location, and we test a Bayesian weighting procedure applied to P-wave arrival/differential times estimated from DAS waveforms. This approach estimates the Posterior Probability Densities of physical (epicenter) and non-physical (hyperparameters) parameters. The hyperparameters are designed as a set of thresholds and weights that enable the automatic identification of portions of data associated with higher reliability in the inversion. Specifically, the thresholds are values having the same dimension as specific waveform attributes and/or geometrical descriptors. Weights are then the scaling factors for the variances of data that do not meet these thresholds. Consequently, several possible weighting schemes (thresholds and weights) based on signal-to-noise ratios, cross-correlation indices, and interchannel distances are explored in a Bayesian framework. We present synthetic tests and real-data applications that demonstrate the potential of this method as an alternative to a similar approach without data weighting.