Forecasting Groundwater Level in Shiraz Plain Using Artificial Neural Networks

Abstract

Groundwater level forecasting plays an important role in water resources management. Artificial neural network have been used as a robust instrument for this subject. In this paper, four architectures of different neural networks were used for groundwater level prediction in Shiraz Plain and their results were compared by using the statistical measures of mean square error and square of correlation coefficient. Effective parameters on groundwater level such as 5-month precipitation and groundwater level histories, temperature or evaporation, and runoff were utilized as the input data to forecast groundwater level at the next time step as output of the networks. All networks were trained for a ten-year period of data (from 1993 to 2003) and calibrated for an 18-month period (from Apr. 2003 to Sep. 2004). Networks were verified based on groundwater level observations in 29 wells located in the plain for another 18-month period (from Oct. 2004 to Mar. 2006). Results showed that artificial neural networks may be successfully utilized to forecast groundwater levels. Different networks forecasted groundwater level in all wells with acceptable root mean square errors of 0.6–12.17 m. Best overall performance was achieved by feed-forward neural network and the second best by Elman neural network.