Abstract
<jats:p>Full moment tensor inversion of shallow mining-induced seismicity is affected by sparse station coverage, near-surface effects, and structured waveform mismatch that is not adequately represented by random-noise assumptions. We present a synthetic diagnostic study using the Jacobina mining district, Brazil, as a realistic reference case for local network geometry, shallow source depths, and empirical waveform contamination. Tests were performed under three contamination regimes: band-limited Gaussian noise, phase-randomised empirical contamination, and structured empirical contamination constructed from small recorded events, thereby preserving source--path--site waveform structure. The phase-randomised case preserves the empirical amplitude spectrum while randomising temporal phase relationships, allowing the effects associated with the empirical amplitude spectrum to be distinguished from those associated with preserved temporal waveform structure. We compare conventional full-waveform inversion with a P--S segmented strategy and examine the effects of prescribed source case, depth, station geometry, and moment tensor constraints. Gaussian contamination produces smooth, variance-dominated degradation. Structured empirical contamination produces stronger mechanism errors and, for the CLVD-like, mixed, and double-couple cases, can inflate spurious non-double-couple components, whereas phase randomisation largely restores Gaussian-like behaviour. The ISO-dominant case remains comparatively stable in orientation but shows persistent decomposition bias. Denser station coverage reduces the amplitude and depth dependence of empirical-contamination errors, although residual degradation can persist at greater depths. P--S segmentation generally reduces sensitivity to structured contamination, especially where P- and S-wave packets are differentially perturbed. Constraint tests show that restrictive source parametrisations can produce solutions that are insensitive to increasing contamination but systematically biased when incompatible with the true mechanism. These results indicate that Gaussian-based thresholds should be treated as comparative stability references and that station geometry and phase-specific waveform treatment should be evaluated jointly when assessing inversion robustness under structured contamination.</jats:p>