arXiv preprint proposes Wasserstein recurrence networks for multiscale paleoclimate pattern detection

The framework converts pairwise 1-Wasserstein distances into a recurrence criterion that remains scale-invariant for one-dimensional Brownian motion, allowing a single probability threshold to flag statistically surprising pattern repetitions regardless of the local sampling interval. This directly addresses the irregular temporal resolution and abrupt regime shifts typical of paleoclimate archives that defeat Euclidean or correlation-based recurrence plots. In the ice-rafted debris example the method isolates events lasting from centuries to tens of millennia within the same record, something conventional recurrence quantification analysis cannot do without arbitrary scale-specific windows. The ice-core application similarly recovers known Dansgaard-Oeschger-like oscillations while surfacing additional shorter recurrences missed by visual or wavelet inspection. Because the threshold is derived from a null model rather than data-driven clustering, the approach supplies an objective statistical control that earlier recurrence studies in paleoclimate lacked. Extension to other unevenly sampled geophysical series appears straightforward provided the underlying process can be approximated as locally Brownian on the scales of interest.