Two types of inversely graded bands in spaced stratification identified by microtextural analysis using a convolutional neural network

ABSTRACT Spaced stratification is a sedimentary structure consisting of multiple inversely graded bands, traditionally interpreted as traction-carpet deposits formed under high-concentration sediment gravity flows. Although widely reported from coarse-grained sandstones deposited by deep-sea turbidity currents, pyroclastic density currents, and hyperconcentrated flows, its formation mechanism remains controversial. Competing hypotheses—ranging from freezing of traction carpets to bedform migration and turbulent sweep–fallout cycles—have lacked quantitative validation from natural deposits. To address this, we conducted a microtextural analysis of spaced stratification in ancient turbidites to clarify its internal variability and depositional implications. We analyzed coarse-grained sandstone samples from the Upper Cretaceous Izumi Group (southwest Japan) using a convolutional-neural-network (CNN)-based semantic segmentation model. The model automatically distinguished grains from the matrix in cross-sectional images with 78.7% accuracy at a resolution of 5 µm per pixel, enabling high-throughput quantification of grain size and orientation. Multivariate analyses of the measured grain data reveal two distinct types of inversely graded bands: 1) Type A, characterized by coarser mean grain size, bimodal distributions of imbrication angles, and pronounced internal variability and 2) Type B, composed of finer grains with unimodal, upstream-dipping imbrication and more uniform textures. These results demonstrate that deposits previously grouped under spaced stratification actually encompass at least two microstructurally distinct types that are likely formed through different depositional dynamics. The CNN-based image analysis provides a new methodological framework for objective, quantitative characterization of sedimentary microtextures at high spatial resolution. Further experimental and numerical studies replicating these two types of banded structures under controlled flow conditions will be essential to constrain their formation processes and refine paleoenvironmental interpretations of high-concentration deposits of sediment gravity flows.