A digital framework for estuarine stratigraphy: an example of a machine learning approach to paleo-environmental classification and coastal evolution

Abstract

Estuarine successions are globally significant stratigraphic archives, fundamental to understanding coastal evolution, assessing petroleum and groundwater reservoirs, and evaluating carbon storage potential. Yet, their interpretation remains hindered by facies heterogeneity and interpretive subjectivity. This study establishes a new digital framework for estuarine sedimentology by integrating high-resolution core analysis with a machine learning–based sediment classification system (Automated Prediction of Environments using Grain Size: APEGS). Applied to Holocene successions from the River Esk arm of the Ravenglass Estuary (northwest England) and trained on 482 modern reference samples, the workflow objectively discriminates six depositional sub-environments—salt marsh, mud flat, mixed flat, sand flat, tidal bar, and tidal inlet/north foreshore—with reproducibility beyond the reach of lithostratigraphic approaches. The results resolve vertical and lateral facies variability at unprecedented precision, capturing transgressive and highstand infilling phases and revealing tide-dominated early Holocene conditions when the current inner estuary was directly connected to the sea. The method establishes a transferable analytical protocol with international applicability across marginal-marine successions, offering a step-change in the quantitative reconstruction of coastal evolution. By replacing subjectivity in facies classification with a reproducible, scalable, and globally transferable digital toolset, this research sets a new benchmark for the stratigraphic interpretation of estuaries. Its methodological innovation directly informs depositional modelling, resource evaluation, and climate adaptation strategies.