Noise Reduction in Earthquake Signals Using Singular Spectrum Analysis

Abstract

Seismic waveform analysis is crucial for understanding earthquakes, but noise from instruments, the environment, or human activities frequently obscures essential signals. This study evaluated the ability of Singular Spectrum Analysis (SSA) to separate noise from seismic signals in both synthetic and real datasets from the Flores Sea region, while also assessing its performance across different distances and noise conditions. The SSA process involves embedding the waveform into a trajectory matrix, performing singular value decomposition, and reconstructing the signal using selected components. Synthetic data were modelled using Ricker wavelets with added Gaussian noise, while observed data were taken from three deep-focus earthquakes (Mw 5.2–7.0) recorded at regional and teleseismic stations. Results show that SSA improves the signal-to-noise ratio (SNR) and preserves the original phase better than conventional lowpass or bandpass filters. In clean signals, SSA has minimal impact on picking P-arrivals, but in moderate or noisy events, it significantly enhances arrival detection (up to 3 seconds difference). Component selection is based on spectral content and waveform similarity, and reconstruction uses matrix summation of selected components. Overall, this study demonstrates that SSA is a practical and flexible tool for enhancing waveform signals.