Predicting the Composition of Modern Sand: Machine Learning Applied to a Global Database of Modern-Pleistocene Samples
Glen Sharman
Associate Professor University of Arkansas
Abstract
Terrigenous sand is primarily composed of three components: quartz, feldspar, and rock fragments. The relative amounts of these ingredients are influenced by where the sand came from, including the types of rock present upstream, whether the region is mountainous or flat-lying, how hot and humid the climate is, and other factors. The composition of clastic sand has long been used to decipher the provenance of ancient sedimentary systems, with modal grain proportions linked to end-member geodynamic settings. Although multiple studies have corroborated the general linkage between sand composition and tectonic setting, the actual system of processes that control the composition of sand is complex, particularly because of interactions and feedbacks between fundamental controlling parameters (e.g., source lithology, climate, relief, slope, etc.) within sediment routing systems.
This presentation explores a new global prediction of sand modal composition (GloPrSM) that is based on an inline series of random forest models that predict the total abundance of quartz (Q), feldspar (F), and lithics (L), along with eight sub-grain types, from known values of precipitation, temperature, elevation, slope, basin area, and source lithology. The GloPrSM model is calibrated using modal point count data from >3,200 modern-Pleistocene sand samples compiled from over 50 published sources and predicts global river sand composition for level 8 watersheds (~1,500 km2 mean area) of the BasinATLAS dataset. The GloPrSM model demonstrates that sand composition can be predicted at the global scale and that topographic slope, temperature, and certain rock types are most important in controlling the makeup of sand. In particular, we found that quartz is enriched in hot, low-lying regions near the equator, likely because quartz is more resistant to weathering than feldspar or lithic grains. Examination of how sand composition is predicted to change along river profiles highlights the complexity of sand’s compositional evolution within some sediment routing systems. The GloPrSM model is likely the first global estimate of sand composition in rivers around the world, yielding insights relevant to study of Earth surface processes and potentially impactful to energy assessment and climate research.
BIO
Dr. Glenn Sharman is a sedimentary geologist who studies how geologic processes (e.g., tectonism, climate, sea level) interact to influence the stratigraphic development and evolution of sedimentary systems. Dr. Sharman’s current research focuses on multi-proxy sedimentary provenance analysis and investigating the deep-time record of climate change. Dr. Sharman received a PhD from Stanford University in 2014 and subsequently joined ConocoPhillips in Houston as an exploration geologist. In 2016, Dr. Sharman joined the Bureau of Economic Geology at the University of Texas at Austin as a postdoctoral researcher before joining the faculty at the Department of Geosciences, University of Arkansas. Dr. Sharman was recently promoted with tenure to associate professor.