Application of Neural Networks for Moon Near-Surface Velocity-Model Extraction

Abstract

Characterizing the shallow lunar subsurface is essential for future exploration activities, including landing-site assessment, infrastructure construction and subsurface resource investigation. However, conventional seismic approaches for near-surface characterization commonly depend on prior velocity information and assumptions regarding the geological structure, which are difficult to constrain for legacy Apollo seismic datasets acquired with unconventional geometries. In this study, we propose a data-driven methodology integrating spectral recomposition and neural networks (NNs) for lunar near-surface velocity-model prediction using active-source seismic data acquired during the Apollo missions. The proposed approach reconstructs spectral information associated with seismic wavelets linked to reflected events and incorporates this information as an additional feature for training a fully convolutional NN. To accommodate the Apollo acquisition geometry, the seismic traces were reorganized into combined common-receiver gathers. Synthetic datasets representative of expected lunar near-surface formations, including unconsolidated regolith and underlying consolidated layers, were generated to train the network. The predicted velocity models successfully reproduced the main kinematic characteristics observed in the lunar seismic data, including continuity of reflected events and travel-time trends. Forward modelling using the reconstructed models generated synthetic seismograms consistent with the observed Apollo seismic data, yielding low normalized root-mean-square error values. The results indicate the presence of a shallow low-velocity regolith layer overlying a more consolidated unit, consistent with previous Apollo 16 and Apollo 17 studies. These findings demonstrate that the proposed methodology can provide physically consistent lunar near-surface velocity models directly from seismic data, without requiring prior velocity analysis, highlighting its potential for future planetary seismic exploration and lunar geotechnical investigations.