Deep Learning
Deep Learning of Suspended Sediment Flux
This project develops a novel deep learning model to predict daily suspended sediment concentration, suspended sediment flux, and river discharge by combining hydrologic data, optical remote sensing, and watershed characteristics. Our approach addresses limitations of existing methods by using satellite and in-situ data without same-day matchups. Focusing on rivers in the conterminous United States, the research demonstrates improved performance through simultaneous modeling of multiple variables and the integration of water color data with hydrology-based predictions. This new methodology aims to enhance our understanding of river processes and provide more accurate, continuous estimates of sediment flux.
Our new hybrid model architecture.