Python package for 3D joint hypocenter-velocity inversion on tetrahedral meshes: Parallel implementation and practical considerations

We present a new Python package to locate microseismic events. The principal algorithm is a modified version of the joint hypocenter-velocity inversion technique that was adapted to operate on tetrahedral meshes. Our motivation behind this project stems from the ability of such meshes to better fit irregular geological contacts and steep topography domains. The adaptation of this method to unstructured meshes required modifications to the way the Jacobian matrix and the smoothness constraint matrices were computed. The package is documented, tested and offers many options for downloading data, seismic phase selection and inversion variables (slowness, velocity and/or Vp/Vs ratio). The algorithm was parallelized using task and process-based approaches. Parallelism occurs during the construction of the Jacobian matrix and during the hypocenter location stage. The performance of this package was benchmarked using synthetic data and different velocity models. Results demonstrate that the algorithms, as implemented in this package, are able to locate seismic hypocenters with an uncertainty that ranges between 8% and 11% of the average distance sources-receivers when data are contaminated with 5% noise and minimal constraints on velocity are used. Computational cost estimates show that the implementation of theses algorithms can operate quickly thanks to the implemented parallelism. Depending on number of available cores, it can reduce the time required to compute the Jacobian matrix by 80% and shaves 90% off the time required to locate the hypocenters. A pre-processing step is also proposed to take full advantages of calibration shots available. Doing so allows us to extract additional a priori information to further constrain the initial velocity and Vp/Vs ratio models and improves convergence towards more realistic solutions.