Prediction of ground water level of Bostan Abad using combining artificial intelligence models

Bostan Abad aquifer located in the East Azerbayjan Province is the main supplier of the region's water needs. Using a numerical model has some limitations such as high complexity, costly, time consuming and a lot of data demanding. For this reason, in the present study the artificial intelligence models including feed forward neural networks (FNN), recurrent neural networks (RNN) and gene expression programming (GEP) were used for prediction of groundwater level changes. Classification of parameters was carried out before modeling, due to their heterogeneity of the aquifer. Precipitation, evaporation, discharge of Ojan River and groundwater level at a time before (t0-1), were used as input parameters in the models. Despite the acceptable results of all three models, based on the average RMSE of each cluster in the training and testing steps, combining the artificial intelligence models using a non-linear neural network as a combiner was adopted to achieve better results than three individual models. The results show decreasing of the average error with a value of 17% in the RMSE for each category in the Supervised Intelligent Committee Machine (SICM) compared to each individual model The SICM was adopted to evaluate the effect of reducing 30 and 50% of the extraction well discharges on groundwater level.The results indicated that the increasing water level in most of piezometers are remarkable. This reflects the high impact of pumping in the amount of groundwater fluctuation is relatively higher than climate change in the study area.