Differential arrival times for source location with DAS arrays: tests on data selection and automatic weighting procedure

Distributed Acoustic Sensing (DAS) technology is currently used to monitor seismic activity, offering a unique spatially-dense representation of the along-the-cable strain wavefield. Traditional seismic networks typically rely on the timing of specific seismic phases to estimate source locations. In this context, DAS arrays may fail to provide accurate traveltimes because of spatially-heterogeneous waveforms. The motivations are (but not limited to) the directional sensitivity, the heterogeneous cable ground-coupling and the enhanced sensitivity to lateral variations in the medium elastic properties. The resulting fluctuations in signal-to-noise ratios of the dense DAS channels pose significant challenges in the automatic picking of body phases, e.g., P-wave Absolute Arrival Times (P-ARTs). Consequently, the complex distribution of the estimated traveltimes impacts the accuracy of event locations, especially if incorrect assumptions on error statistics (e.g., Normal distribution) are considered. In this study, we address this issue by exploiting the intrinsic DAS measurements' spatial density and testing selected P-wave Differential Arrival Times (P-DATs) for source location. We estimate P-DATs for all the possible DAS channel pairs by identifying the time delay corresponding to the peak of each cross-correlation function. Subsequently, we select P-DATs based on two criteria: interchannel distance and cross-correlation index value. This procedure is often employed to reduce the risk of mixing delay times from coherent and incoherent waveforms. As a first test, using a probabilistic inversion (Hamiltonian Monte Carlo method), we demonstrate how the selected P-DATs provide a better constraint on the event's azimuthal direction compared to P-ARTs. Then, as a second experiment, we move from a subjective selection of P-DATs. To do so, we test a fully-automated and data-driven covariance matrix weighting procedure, in a probabilistic inversion scheme. Specifically, we compute posterior probability distributions for both the physical parameters (event location) and hyperparameters related to data features (interchannel distance and cross-correlation index thresholds). In this scheme, the hyperparameters define each weight along the diagonal of the covariance matrix. These tests offer useful insights into the utilization of P-DATs for event location with DAS. Moreover, we provide an automatic approach to avoid subjective biases based on pre‐conceptions in the a-priori data selection.