Coherent Source Subsampling for Ambient Noise Correlation Analysis: A Himalayan Case Study

Ambient noise interferometry enables the retrieval of inter-station surface wave responses through cross-correlation and linear averaging of continuous seismic response, under the assumption that the seismic wavefield is equipartitioned, with energy uniformly distributed over all propagation directions. In practice, however, ambient noise sources are highly non-uniform in both space and time, leading to biased estimates of the inter-station response between station pairs. If sources located within the stationary-phase zone can be identified and only the corresponding cross-correlation windows are selected for averaging, the inter-station response can be more accurately approximated, resulting in improved causal–acausal symmetry. Deep learning-based coherent source subsampling has been shown to effectively identify stationary-phase noise sources, thereby enhancing the recovery of physically meaningful inter-station surface wave responses.

In the Himalayan region, linear averaging of ambient-noise cross-correlations often does not yield causal–acausal symmetry and fails to recover inter-station surface wave response. In this study, we use data driven coherent source subsampling approach to systematically identify ambient-noise cross-correlations associated with stationary-zone sources prior to averaging. In this study, data from 19 stations deployed along a linear profile in the Kumaon–Garhwal Himalaya, spanning the periods 2005–2008 and 2011–2012, are analyzed. The continuous seismograms during the mentioned period were divided into 30-minute windows, and inter-station cross-correlations were computed for 167 station pairs. Using a symmetric variational autoencoder with discrete latent variables, we subsampled cross-correlation windows into distinct source states and select those corresponding to the stationary-phase zone, characterized by pronounced causal-acausal symmetry and maximum time lag. Averaging cross-correlations associated with the stationary-phase source state enhances the inter-station surface-wave dispersions, with causal and acausal branches yielding similar dispersions. These results show that coherent source subsampling provides an effective framework for improving ambient-noise interferometry in complex Himalayan geological settings.