Application of artificial neural networks to estimating DO and salinity in San Joaquin River basin

Abstract

In the current research, which is based on recorded and collected data from one of the San Joaquin River data recording stations, an artificial neural network (ANN) model was developed to simulate water quality parameters. Then, the results were compared with traditional salinity formula to optimize parameters. We also chose the best parameters for estimation of dissolved oxygen (DO) through an ANN model. In both models, we used feed-forward perceptron training algorithm along with Levenberg–Marquardt as the learning algorithm and tansign(x) as transfer function algorithm. To simulate the salinity, we used more than 5,000 water quality data-sets. Also, we used two groups of data-sets, with 16,000 related data for simulating DO. We developed a highly precise model and verified the results with the most recommended formula. Mean squared error is 12.5 for the presented model and 9061 for the traditional formula (Salinity = 0.64*EC). We also recommend a formula whose result is very close to the pilot-recorded data. It showed a large disagreement between traditional formula and the proposed model. We used MATLAB software to optimize the design parameters of the model.

Keywords:

• Water quality
• ANN model
• Simulation
• San Joaquin River
• Electrical conductivity

Acknowledgment

The authors are grateful to Dr H. Moradi and Dr Sohrab Soori for their editorial and revision assistance. They are also thankful to San Joaquin River Monitoring Stations Control Board for providing data for the current analyses.