Replacing spatial distance with waveform similarity in nearest-neighbour earthquake clustering

Earthquake clustering methods based on nearest-neighbour distances provide a powerful framework for identifying seismic sequences and distinguishing clustered from background seismicity. Classical formulations rely on spatial proximity and therefore require accurate hypocentral locations, which limits their applicability to single-station template-matched catalogues with only a subset of located earthquakes. Here we introduce a new clustering approach that replaces spatial distance by waveform similarity, quantified through normalised cross-correlation, while retaining the established time–magnitude scaling of nearest-neighbour methods. Waveform similarity is treated as a distance in a feature space, and its effective dimension is estimated directly from the data using a maximum-likelihood intrinsic-dimension estimator. This allows the definition of a similarity-based nearest-neighbour distance with a clear statistical interpretation as the expected number of background events in time–similarity–magnitude space. 

The method is specifically designed for single-station template-matched catalogues, where waveform similarity provides constraints on source proximity and fault association. We test the approach on natural seismic sequences in Switzerland using template-matched catalogues and benchmark the results against clusters obtained from double-difference relocated catalogues. This study aims to assess whether waveform similarity can robustly replace spatial distance in nearest-neighbour clustering while preserving the statistical and physical interpretability of the method, and to evaluate its potential for the analysis of dense template-matched catalogues from sparse or single-station seismic deployments.