Appli cation and Comparison of Integrated Time Series and Artificial Neural Network Model for Prediction of the Variations of Groundwater Level (Case study: Plain Marvast)

Prediction of groundwater level fluctuations is an essential step for management, appropriate planning and efficient use in arid and semi-arid regions. According to the existing information, the general trend of groundwater hydrograph represents declining trend and continuous falling of groundwater level during the last years in the Marvast plain. In this study, integrated time series and neural network models was used to predict the fluctuations of groundwater levels in the Marvast plain. For this purpose, the groundwater level data of 1987-2009 time period were provided and different integrated time Series models and artificial neural network were fitted to the data. The efficiency and accuracy of ARIMA model for predicting future values was assessed using the Root Mean Square Errors (RMSE) and Akaike information criterion. The results of different ARIMA states showed the ARIMA (1, 1, 0) is the best-fit time series model. Three train functions of Levenberg-Marquardt, Resilient Back Propagation and Scaled Conjugate Gradient were used for the feed-forward back propagation neural network. The results showed Levenberg-Marquardt function is the best train function to predict groundwater level. Comparing the results of ARIMA (1,1,0) and neural network feed-forward with back propagation algorithm using RMSE, MAE and CE statistics showed neural network model is relatively superior to the Integrated Time Series model.