Development of a Smart Model for Groundwater Level Prediction Based on Aquifer Dynamic Conditions

In recent years, drought and demand growth in most parts of the county have caused a dramatic increase in using groundwater for water supply purposes. Besides, unplanned excessive discharges from aquifers have led to aquifer degradation. In most integrated water resources management models, groundwater simulation is used for taking into account discharge constrains, however, the most cases the optimal solution is not achievable.Hence, artificial neural network models may be replaced by groundwater numerical simulation models. In this paper, a methodology based on dynamic artificial neural networks (DANN) is developed for simulating groundwater table. Karaj aquifer is used as the case study and its groundwater numerical model (PMWIN) is calibrated using such measured groundwater characteristics as hydraulic conductivity and specific storage. The results of the numerical model are presented to DANN for training. In the proposed procedure, the total recharge, discharge, and groundwater level in previous time intervals are used as the inputs to the DANN model. The model output is the groundwater level at the end of the time interval. In this study, the development of the model is accomplished in four steps that consist of developing the aquifer simulation model and its calibration, producing the input-output data set for DANN training, training the DANN model for various structures, and selecting the best structure for use in the optimization model. The result shows that the proposed DANN model is more efficient in simulating groundwater level fluctuations than the static artificial neural network (SANN) models.