Structured empirical waveform contamination in moment tensor inversion of shallow mining seismicity: A synthetic diagnostic framework

Sandro Giovani Gomes , Aderson Farias do Nascimento, José Augusto Fonsêca, Gilberto da Silva Leite Neto, Eduardo Alexandre Menezes, Reinhardt Fuck

Moment tensor inversion of shallow seismicity can be strongly affected by sparse station coverage, uncertain source depth, and waveform mismatch caused by near-surface structural complexity. Conventional synthetic tests usually represent these uncertainties using random noise, but such tests may not reproduce the temporally organised waveform distortions present in local-network data. We develop a diagnostic framework to evaluate this problem using the Jacobina mining district, Brazil, as a realistic test case for shallow mining-induced seismicity. Synthetic waveforms were contaminated using three complementary strategies: band-limited Gaussian noise as a stochastic reference; recorded local-event waveforms as structured source–path–site contamination; and phase-randomised versions of the same empirical waveforms, preserving their Fourier-amplitude spectra while destroying the original temporal phase organisation. We tested two observed six-station configurations and a 12-station combined configuration over source depths of 0.1–5 km. Conventional full-waveform inversion (FW-I) was compared with a P–S segmented inversion (PS-I), in which P- and S-wave windows are treated separately. Gaussian and phase-randomised contamination produce similar ensemble-level behaviour, with generally smooth increases in moment-tensor rotation as contamination increases. Structured empirical contamination produces a more heterogeneous response: many source–depth–
orientation configurations remain weakly affected, whereas a smaller subset develops much larger errors. This behaviour is clearest under FW-I. For the compact six-station network at the highest contamination level, structured contamination produces rotation errors more than 5◦ larger than the phase-randomised case in 32.1 per cent of realisation-level comparisons, whereas differences below −5◦ occur in only 7.3 per cent. Differences exceeding +10◦ occur in 16.4 per cent of comparisons, compared with 2.6 per cent below −10◦ . Structured temporal organisation therefore preferentially increases the likelihood of large inversion errors rather than uniformly degrading all configurations. Tests with two additional empirical contaminants reproduce the tendency for larger FW-I errors under structured than phase-randomised contamination, but the depth of maximum susceptibility changes among waveforms. The detailed depth response therefore reflects interactions among source configuration, contaminating waveform, and station geometry. PS-I substantially reduces both the magnitude and frequency of large structured-contamination errors, while the 12-station configuration further suppresses their expression through increased observational redundancy. Source-type discrepancies based on moment-tensor eigenvalues show a weaker and more selective response, indicating that a substantial part of the full-tensor error is associated with changes in recovered orientation. Constraint tests additionally show that solutions can appear stable while remaining systematically biased when an inappropriate source constraint is imposed.
These experiments define a transferable stress-testing framework for shallow moment tensor inversion: establish a stochastic reference, introduce representative structured waveform mismatch, remove its original phase organisation while retaining spectral amplitude, evaluate both typical and large-error behaviour, and test whether acquisition or inversion modifications improve robustness. The numerical thresholds are site-specific, but the diagnostic workflow can be adapted to other shallow and structurally complex seismic environments.