UVM Theses and Dissertations
Format:
Print
Author:
Hamshaw, Scott D.
Title:
Dept./Program:
Civil and Environmental Engineering
Year:
2014
Degree:
MS
Abstract:
Over 1,000 river miles in Vermont are either impaired or stressed by excessive sedimentation. Higher streamflows and incised river channels have resulted in increased bed and bank erosion. As the climate in the Northeast is, expected to feature greater and more frequent precipitation events and winter rainfall, the potential for increased sediment loading from erosion processes in the watershed and along the channel are high and a major concern for water resource managers and other stake holders. Typical sediment monitoring comprises periodic sampling during storm events and is often limited to gauged streams with flow data.
Continuous turbidity monitoring enhances our understanding of river dynamics by offering high-resolution, temporal measurements to better quantify the total sediment loading occurring during and between storm events. Artificial neural networks (ANNs), that mimic learning patterns of the human brain, have been effective at predicting flow in small, ungauged rivers using local climate data. This study advances this technology by using an ANN algorithm known as a counter-propagation neural network (CPN) to predict discharge and suspended sediment in small, ungauged streams.
The first distributed network of continuous turbidity sensors was deployed in Vermont in the Mad River Watershed, located in Central Vermont. The Mad River and five tributaries were selected as a test bed because seven years of periodic turbidity sampling data are available; it represents a range of watershed characteristics, and because the watershed is also being used for hydrologic model development using the Distributed-Hydrology-SoiIs-Vegetation Model (DHSVM). Comparison with the DHSVM simulations will allow estimation of the most-likely sources of sediment from the entire watershed and individual subwatersheds. Inaddition, recent field studies have commenced in quantifying erosion from unpaved roads and streambanks in the same watershed.
Periodic water quality sampling during stonn events enabled turbidity versus total suspended solids relationships to be established. In sub-watersheds with monitored turbidity and stage, I5-minute precipitation, soil moisture and air and water temperature data are also being collected. Stage sensors and theoretical rating curves are used to validate the flow predictions from the CPN. The real-time turbidity data are used to train and test the suspended sediment predictions from the CPN network at each site. The turbidity data are also used for training the CPN on a subset of tributaries and testing on the remaining subwatersheds. Reasonable estimates of suspended sediment discharged from the tributaries and the main stem of the Mad River are calculated and compared enabling a more reliable foundation for building a sediment budget. Results of this study will assist managers in prioritizing mitigation projects to reduce impacts of sediment loading.
Continuous turbidity monitoring enhances our understanding of river dynamics by offering high-resolution, temporal measurements to better quantify the total sediment loading occurring during and between storm events. Artificial neural networks (ANNs), that mimic learning patterns of the human brain, have been effective at predicting flow in small, ungauged rivers using local climate data. This study advances this technology by using an ANN algorithm known as a counter-propagation neural network (CPN) to predict discharge and suspended sediment in small, ungauged streams.
The first distributed network of continuous turbidity sensors was deployed in Vermont in the Mad River Watershed, located in Central Vermont. The Mad River and five tributaries were selected as a test bed because seven years of periodic turbidity sampling data are available; it represents a range of watershed characteristics, and because the watershed is also being used for hydrologic model development using the Distributed-Hydrology-SoiIs-Vegetation Model (DHSVM). Comparison with the DHSVM simulations will allow estimation of the most-likely sources of sediment from the entire watershed and individual subwatersheds. Inaddition, recent field studies have commenced in quantifying erosion from unpaved roads and streambanks in the same watershed.
Periodic water quality sampling during stonn events enabled turbidity versus total suspended solids relationships to be established. In sub-watersheds with monitored turbidity and stage, I5-minute precipitation, soil moisture and air and water temperature data are also being collected. Stage sensors and theoretical rating curves are used to validate the flow predictions from the CPN. The real-time turbidity data are used to train and test the suspended sediment predictions from the CPN network at each site. The turbidity data are also used for training the CPN on a subset of tributaries and testing on the remaining subwatersheds. Reasonable estimates of suspended sediment discharged from the tributaries and the main stem of the Mad River are calculated and compared enabling a more reliable foundation for building a sediment budget. Results of this study will assist managers in prioritizing mitigation projects to reduce impacts of sediment loading.