Zircon transport histories revealed by automated classification of scanning electron microscope images

Wind, waves, and currents leave distinct imprints on the sand that they transport. Microscopic features like pits and scratches on the surfaces of sand grains thus record information about how sand has been transported in the past. For various historic fields of inquiry, ranging from sedimentary geology to forensics, sand textures can provide detailed, otherwise unobtainable insight into past surface environments. Recent work has emphasized the value of sand texture analysis by improving texture quantification, thereby reducing human bias. Most work has focused on quartz sand due to its ubiquity, but quartz sand is susceptible to alteration over geologic timescales that can render it unsuitable for microtexture analysis. Zircon is much less susceptible to alteration, making it a more suitable mineral to analyze very ancient (less than 500 Ma) sand-bearing deposits. Here, we develop and use a machine learning tool, SandAI-Z, to classify electron scanning microscope images of zircon sand grains by depositional environment: here, aeolian, beach, or fluvial. The tool was adapted from SandAI, a machine learning classifier originally developed to accomplish the same task using quartz grains. Given the different crystallographic and mechanical properties of both mineral phases, this new tool was necessary to accurately analyze zircon grains. The tool is freely available to use with no coding required: users can submit images online and obtain results in a matter of seconds. The modified algorithm classifies individual grains of sand from modern and ancient samples of known depositional history with high accuracy (73.5%) on our test dataset, expanding the possible application of quantitative, automated microtexture analysis to Earth’s entire metasedimentary record.