Multifractal signatures reveal source-dependent differences between tectonic earthquakes and icequakes

Abstract Discriminating between local tectonic earthquakes and cryo-seismic events (icequakes) poses a significant challenge for automated seismic monitoring due to the similarity of their waveforms in the time and frequency domains. To address this, we present a robust classification framework utilizing Multifractal Detrended Fluctuation Analysis (MFDFA) combined with machine learning. The framework represents each waveform by explicitly defined multifractal parameters, which quantify the nonlinear scaling behavior and structural organization of the waveform. A total of 28,022 station-event waveform records obtained from nine seismic stations in the Columbia Glacier region, Alaska, were analyzed under different Signal-to-Noise Ratio (SNR) and data-availability conditions. The analysis of the seismic-network dataset shows class-dependent differences in the multiscale organization of tectonic earthquake and icequake waveforms. Although the expression of individual descriptors varies from one station to another, the complete multifractal representation supports a robust station-specific discrimination under different recording conditions. For each station, an independent Support Vector Machine (SVM) classifier was trained. The mean accuracy obtained from the nine classifiers was 94.28%, while the highest accuracy at an individual station reached 98.01%. These results demonstrate that the proposed multifractal framework provides a physically grounded and highly accurate tool for purifying seismic catalogs in complex glaciated environments.