Abstract Archive Select a year below to view:
Artificial neural networks for filling data gaps and improving hydrologic simulations in coastal watersheds
Proceedings of the 2020 Mississippi Water Resources Conference
Year: 2020 Authors: Upadhyay P., Linhoss A.
The accuracy of streamflow measurements and models is very important in water-resources planning and management. Traditionally, streamflow is simulated using physically based hydrologic models such as the Soil and Water Assessment Tool (SWAT). Previous studies have also compared SWAT with Arti?cial Neural Network (ANN) models to understand which give the best results. Traditional hydrologic models have the advantage that they simulate mechanistic processes which enables users to understand and quantify physical components of the system. In some cases, physical processes of a system are either unknown or unimportant. Here ANN models have the advantage that they can predict the relationship between the inputs and outputs of a process without an understanding of the physical characteristics of the system. We propose to combine the SWAT model with an ANN model to improve stream?ow estimation. We employ each of these models according to their strengths. The ANN model was used to fill gaps in time series data and estimate unknown physical processes. The data developed by the ANN model was used as an input into the SWAT model to simulate streamflow. The models were developed for a coastal watershed in Florida, which drains to the Biscayne Bay. Biscayne Bay has been recently designated as one of ten habitat focus areas across the country by the National Oceanic and Atmospheric Administration (NOAA). The Biscayne Bay region is unique because of its karst geology, flat topography, system of regulated surface canals, and the oligotrophic nature of the Bay. In the future, the developed model will also be used to simulate water quality in Biscayne Bay, Florida.