Determining Depositional Environments Using Quartz Microtextures and Deep Learning

As sand moves across Earth’s landscapes, the shapes of individual grains evolve and microscopic textures accumulate on their surfaces. Because transport processes vary between depositional environments, the shape and suite of microtextures imprinted on sand grains provide insights into their transport histories. Previous efforts to link microtextures to specific transport environments have demonstrated that microtextures can provide important information about the depositional environments of rocks with few other indicators. However, several drawbacks have hindered broad adoption of the technique: analyses rely on 1) subjective human assessment of microtexture occurrence, which can yield biased, error-prone results; 2) a non-standard nomenclature and number of microtextures; and 3) relatively large samples (>150 grains) to obtain statistically significant results, the manual documentation of which is extremely labor intensive.

We addressed these limitations by developing a deep learning model that classifies scanning electron microscope (SEM) images of sand grains by transport environment with 93.3% accuracy. The model was trained, validated, and tested on microscope images of sand grains from modern environments around the globe and Triassic-Cretaceous rocks with known depositional environments. Training data encompass a range of transport environments (fluvial, aeolian, glacial, coastal) and diagenetic alteration states. Grains were imaged using a variety of SEM instruments to ensure broad applicability. We further tested the model on Triassic-Cretaceous rocks of known origin, and two Cryogenian samples of contested origin. Results demonstrate that the model provides a versatile new tool to quickly and automatically constrain the depositional histories recorded in individual grains of quartz sand.