Newly Discovered Temperature-Related Long-Period Signals in Lunar Seismic Data by Deep Learning

Abstract

Lunar seismic data are essential for understanding the Moon's internal structure and geological history. After five decades, the Apollo data set remains the only available one and continues to offer significant value for current and future lunar seismic data analyses. Recent advances in artificial intelligence for seismology have identified seismic signals that were previously unrecognized. In our study, we utilized deep learning for unsupervised clustering of lunar seismograms, leading to the discovery of a new type of long-period lunar seismic signal that existed every lunar night from 1969 to 1976. We then conducted a thorough analysis covering the timing, frequency, polarization, and temporal distribution characteristics of this signal to study its properties, occurrence, and probable origins. This signal has a physical cause instead of artificial, such as voltage changes, according to its amplitudes during peaked and flat modes, as well as the digital converter status. Based on its relation to the lunar temperature and documents on Apollo instruments, we conclude that this signal is likely induced by the cyclic heater, with several unresolved questions that might challenge our hypothesis. Excluding interference from this newly identified signal is crucial when analyzing lunar seismic data, particularly in detecting lunar free oscillations. Our research introduced a new method for discovering new types of planetary seismic signals and helped advance our understanding of Apollo seismic data. Furthermore, the discovery of this signal holds valuable implications for the design of future planetary seismometers to avoid encountering similar issues.