Water Resources Research Act Program

Details for Project ID 2017VT87B

Application of neural networks to classify erosional and depositional stream reaches in glacially-conditioned Vermont catchments

Institute: Vermont
Year Established: 2017 Start Date: 2017-03-01 End Date: 2018-02-28
Total Federal Funds: $9,998 Total Non-Federal Funds: $19,958

Principal Investigators: Donna Rizzo, Mandar Dewoolkar, Kristen Underwood

Abstract: Water resource managers in Vermont would benefit from a more nuanced representation of sediment transport dynamics in rivers, and how sediment flux varies spatially, from reach to basin scales. A better understanding of sediment transport dynamics will help to identify critical catchment locations and time periods responsible for disproportionate fluxes of sediment (and associated nutrients). Classification of reach-based sediment transport regimes could also help to identify infrastructure at enhanced risk of failure from fluvial erosion during extreme flood events. We aim to demonstrate the utility of advanced, data-driven computational methods to characterize and predict spatial variation in sediment export from glacially-conditioned catchments in response to natural and human disturbances. This graduate research proposal requests resources to support an additional dissertation chapter for PhD candidate, Kristen Underwood. The objective of the research is to further develop an existing set of artificial neural networks (ANNs) to model and predict sediment transport and deposition at the reach scale within study catchments located in southern and northern Vermont. Provisional results of a Self-Organizing Map (SOM) and Counterpropagation Network (CPN) show promise for classifying the sediment transport regime of river reaches in these catchments. Newly-available stream power and infrastructure data sets and geoprocessing models developed by collaborating PhD candidate, Ian Anderson, represent an opportunity to further refine data inputs to these ANNs, with expected improvements in ANN performance. Expected outcomes include a set of computational tools that can be used to estimate sediment transport regimes of Vermont river networks at the reach scale, and which can be linked to GIS for visualization. Results will support river corridor planning efforts of the VT Agency of Natural Resources (VTANR) to classify river reaches by their sensitivity to future vertical and lateral adjustment, with classifications considered in a tactical basin planning context to identify and prioritize channel and watershed restoration and conservation projects. Results will support hazard management activities of the VT Agency of Transportation (VTrans) relating to infrastructure (i.e., bridges, culverts, roads) at risk from fluvial erosion. Opportunities will also be explored with VTANR staff for the possible interface of computational tools with the VTANR Stream Geomorphic Assessment database and web-based Natural Resources Atlas.