Depth-Informed Bayesian Stratigraphic Correlation via Latent Signal Reconstruction

Following the success of Bayesian age modelling in chronostratigraphy, Bayesian approaches are increasingly being applied to quantitative stratigraphic correlation through uncertainty-aware alignment and latent signal reconstruction. Existing Bayesian methods, including BIGMACS, StratoBayes, and StratMC, have shown that probabilistic frameworks can integrate proxy observations and stratigraphic information within flexible alignment models to infer section correlations more rigorously than deterministic matching approaches. Inspired in particular by StratMC, we develop a Bayesian stratigraphic correlation method that retains the concept of a common proxy signal, but replaces mandatory requirements of explicit age priors with a depth-informed simplex parameterization. Our framework models the shared component of the proxy signal across neighboring sections using a Gaussian process, while the latent age increments between adjacent samples is parameterized on a simplex and assigned a Dirichlet prior informed by observed depth spacing. This parameterization incorporates depth information directly into the prior while preserving stratigraphic superposition, thereby discouraging implausible stretching or compression in the alignment of unevenly sampled intervals. The result is a correlation-oriented Bayesian framework designed for sections in which proxy observations and sample depths are available but robust geochronologic constraints are absent. In contrast to deterministic warping schemes, our method yields a full posterior distribution over the shared component of each proxy and latent mappings between sections, allowing uncertainty to be propagated into downstream quantitative stratigraphic interpretation. The framework offers a bridge between proxy modelling and stratigraphic matching while reducing dependence on external chronological priors. We suggest that our strategy has particular promise for alignment of densely sampled neighboring sections lacking reliable absolute chronology.