Haque, I, Törnman, W, Martinsson, J, Vanegas-Palacio, C, Viljoen, N, Primadiansyah, A & Svanberg, E 2026, 'The benefits of unconstrained, data-driven velocity models for 
seismic processing in caving operations', in A van As, D Cumming-Potvin & J Wesseloo (eds), Caving 2026: Proceedings of the Sixth International  Conference on Block and Sublevel Caving, Australian Centre for Geomechanics, Perth, pp. 1-12, https://doi.org/10.36487/ACG_repo/2645_47
(https://papers.acg.uwa.edu.au/p/2645_47_Martinsson/)
Abstract: Microseismic monitoring is essential for panel caving operations, but a complex and changing environment makes it challenging to obtain accurate event location. This study evaluates an unconstrained, data-driven rock mass model (including velocity) for high-precision seismic processing at the Deep Mill Level Zone (DMLZ) panel cave in the Grasberg district. The approach uses the Bayesian estimation of mining-induced seismicity (BEMIS) framework, an automated, real-time system designed for the highly irregular and heterogeneous ‘Swiss cheese’ appearance of the underground rock mass. By combining Bayesian inference with adaptive calibration, event location accuracy and precision is improved, revealing clusters, structures and features that improve the understanding of cave propagation. The study also incorporates high-resolution P- and S-wave tomography by BEMIS, producing detailed velocity models with quantified uncertainties that enable dataquality filtering for reliable interpretation. The tomograms show low-velocity zones that closely match the evolving cave shape and high-velocity areas exhibiting stress accumulation, and highlight areas with poor ground conditions. In combination, the precise event locations and tomography provide new and valuable insights into cave evolution and extraction performance at the DMLZ. Overall, the study demonstrates that data-driven, Bayesian seismic processing offers a robust solution for precise velocity model estimation and cave monitoring in complex mining environments.

Keywords: caving, seismic processing, data-driven velocity model, mining-induced seismicity, seismic tomography, event location accuracy