Unlocking the potential of single stations to replace seismic arrays

We introduce Virtual Seismic Arrays, which predict full array recordings from a single reference station, eliminating the need for continuous deployment of all stations. This innovation can reduce costs and logistical challenges while maintaining multi-station functionality. We implement a Virtual Seismic Array using a deep learning encoder-decoder approach to predict transfer properties between stations. Training on recordings from the Gräfenberg array in the secondary microseism frequency band allows us to retrieve models capturing transfer characteristics between stations. These models form the Virtual Seismic Array. To evaluate performance, we beamform original and predicted waveforms to detect dominant secondary microseism sources. We assess three scenarios: one aligning with the training dataset, another with two regimes in training but testing on one, and a third where training data does not align with the testing regime. Our results show strong agreement between predicted and original beamforming results, demonstrating the potential of Virtual Seismic Arrays.