Modeling Dissolved Oxygen (DO) Concentration Using Different Neural Network Techniques

Abstract

The concentration of dissolved oxygen (DO) is important for the healthy functioning of aquatic ecosystems, and a significant indicator of the state of aquatic ecosystems. DO is a parameter frequently used to evaluate the water quality on different reservoirs and watersheds.In this study, two different ANN models, that is, the multilayer perceptron (MLP) and radial basis neural network (RBNN), were developed to estimate DO concentration by using various combinations of daily input variables, pH, discharge (Q), temperature (T), and electrical conductivity (EC) measured by U.S. Geological Survey (USGS). The data of Fountain Creek Stream - Gauging Station (USGS Station No: 07106000) which cover 18 years daily data between 1994-2011 were used. The ANN results were compared with those of the multiple linear regression (MLR). Comparison of the results indicated that the MLP and RBNN performed better than the MLR model. The RBNN model with three inputs which are pH, Q,and T was found to be the best model in estimation of DO concentration according to the root mean square error, mean absolute error and determination coefficient (R2) criteria.