2nd UF Water Institute Symposium Abstract

   
Submitter's Name Renee Murch
Session Name Poster Session: Hydrologic, Biogeochemical and Ecological Processes 1
Category Hydrologic, biogeochemical and ecological processes
Poster Number 231
 
Author(s) Renee Murch,  INTERA, Inc.
  Patrick Tara,  INTERA, Inc.
  Doug Munch, St. Johns River Water Management District
  Xinjian Chen, St. Johns River Water Management District
   
  Utilization of Artificial Neural Networks (ANNs) for Hydrologic Modeling Applications
   
  In recent years, the application of artificial neural networks (ANNs) to model hydrologic processes has become an increasingly attractive alternative to other types of statistical models. ANNs can be an efficient way of modeling hydrologic processes in situations where explicit knowledge is not available or the system is just too complicated to represent numerically. Two applications of ANNs for hydrologic modeling are presented: a salinity model and a spring flow model. The networks were trained, validated, and utilized as predictive tools to develop time series for periods of interest where observed data was not available.

ANNs were effectively utilized in order to develop a statistical model for the prediction of top and bottom salinity at the mouth of the Manatee River. A total of six ANNs were developed based on flow regime and flow location (top or bottom). Total ungauged flow, recorded stage, and nearby lake discharges were utilized as input data for each of the ANNs. The ANNs consistently out-performed the multiple linear regression models and were therefore utilized in predictive mode to develop top and bottom salinity time series from 1982 through 1984. The resulting predictive time series was utilized by the Southwest Florida Water Management District to support MFL development for the Manatee River.

ANNs were also successfully applied to estimate the discharge from White Springs. White Springs, located in Hamilton County, is of particular of interest because it is a first magnitude spring with historical significance. In recent years, flow reversal in White Springs has occurred, making it particularly difficult to quantify the spring flow. Using the best and most complete data available, two statistical models were developed to develop a flow time series for White Springs: a multiple linear regression and an artificial neural network (ANN). Using nearby stage and well data, two ANNs were developed to estimate White Springs discharge. The networks were trained using all available White Springs flow data. Validation of the networks demonstrated that the networks were able to successfully estimate the direction of flow. Utilization of both networks to determine average spring flow for 1993 through 1994 resulted in comparable results, with the ANNs out-performing the multiple linear regression models during the training period.

These studies suggest that feed-forward back propagation ANN based modeling can be effectively applied as an alternative approach to other statistical models for the estimation of hydrologic variables. And, as with any modeling technique, careful selection of explanatory variables is essential to achieve optimal performance.