Multi-scale Characterization of Seismic Noise and Signals in an Underground Coal Mine

Seismic monitoring in underground coal mining environments is influenced by various anthropogenic and natural noise sources. The background noise, predominantly of mechanical origin, shows strong spatial and temporal variability. Some highly impulsive sources share common characteristics with genuine seismic events. Routine blasting activities within the mine and from surrounding regions also contribute significantly to the recorded data. Mining-triggered sources such as microseismicity, subsidence, roof falls, and occasional sensing of tectonic earthquakes originating from distant locations further contribute to the recorded data. The combined influence of these sources strongly affects the performance of conventional processing workflows, frequently resulting in false detections and event misclassifications.

In this study, continuous seismic data recorded in an underground coal mine using eight short-period seismometers over a six-month duration are analysed to characterise signal and noise properties across temporal, spectral, and spatial domains. Spectral persistence, correlation metrics, and multichannel signal-processing techniques are used to identify dominant noise sources and assess their influence on the recorded waveforms. Persistent mechanical activity is shown to dominate the spectrum, with numerous harmonics and broadband noise, motivating the use of multiscale decomposition methods.

We evaluate the performance of Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD) for multiscale analysis. Our results show that EMD can introduce spurious low-frequency modes that are absent from the original signals and can therefore be misinterpreted. In contrast, VMD’s constrained-bandwidth formulation yields more physically meaningful scale separation. The multivariate extension of VMD (MVMD) has been used for better mode alignment and correlation across channels.

Overall, these results demonstrate the advantages of constrained, multivariate multiscale methods for the characterization of signal and noise with implications for improving seismic monitoring and event classification in complex environments.