SEISMIC SITE QUALITY ASSESSMENT IN NORTH SUMATRA USING SPECTRAL DENSITY ANALYSIS AND MACHINE LEARNING-BASED CLUSTERING

Abstract

Seismic noise strongly influences the accuracy and reliability of earthquake monitoring, particularly in tectonically active regions such as North Sumatra. This study investigates the quality of seismic stations by analyzing noise characteristics using Power Spectral Density (PSD), Probability Density Functions (PDFs), and machine learning clustering. PSD was computed through the Fast Fourier Transform (FFT) and compared against the New High Noise Model (NHNM) and New Low Noise Model (NLNM) benchmarks. Noise variability was further quantified using PDFs, while fuzzy c-means (FCM) clustering was applied to classify temporal noise patterns. Results from the MUTSI seismic station demonstrate strong diurnal and weekly variability, with horizontal components (SHE and SHN) exhibiting significantly higher noise levels and fluctuations than the vertical component (SHZ). Noise amplitudes peaked during morning hours (06:00–09:00 UTC), correlating with anthropogenic activity, and decreased substantially at night, indicating that optimal recording conditions occur during late evening to early morning. FCM clustering identified five dominant noise regimes, separating stable low-noise baselines from sporadic high-noise anomalies likely associated with human activity or instrumental disturbances. These findings highlight the importance of integrating spectral analysis with clustering techniques to evaluate seismic station performance, improve real-time monitoring, and guide optimal site selection and operational scheduling for earthquake detection.