Reconstructing Seismic Integrity from Urban Noise: A Deep Learning Approach for Improving Event Detection Martarizal 1, Eko Priyatno 12, Ahmad Kadarisman 12, Santoso Soekirno 1
1. Department of Physics, Universitas Indonesia, Depok, Indonesia
2. Directorate of Instrumentation and Calibration, BMKG, Jakarta, Indonesia
Abstract
Urban-induced noise remains a critical barrier to reliable seismic event detection, particularly in stations located in densely populated environments. This study presents a deep learning-based framework employing a seismic denoising autoencoder (SeisDAE) to reconstruct high-fidelity earthquake signals from noise-contaminated recordings. The model is trained on paired, time-synchronized waveforms from a low-noise reference station and a high-noise urban station, with data segmented into non-seismic intervals and confirmed earthquake events. This structure allows the autoencoder to jointly learn background suppression and event feature preservation. Preliminary evaluations demonstrate that the proposed approach effectively restores the visibility of seismic events in noisy station recordings, aligning them closely with their clean station counterparts. The results suggest that SeisDAE offers a promising direction for enhancing the operational performance of urban seismic networks, particularly in early warning and microseismic monitoring contexts.
Keywords: Seismic denoising, urban noise suppression, deep autoencoder, SeisDAE, earthquake signal reconstruction, time-synchronized seismic data, seismic event detection
