Groundwater modeling using artificial intelligence methods (Case study: Dezful-Andimeshk plain)

Groundwater is one of the most important water resources on earth, and groundwater level and groundwater salinity studies are very important to protect and plan the water resources, especially in the arid and semiarid areas, such as Iran. Groundwater quantitative and qualitative testing is time-consuming and costly. Therefore, using the models to simulate the quantity and quality of groundwater has become common.

In recent decades, the artificial intelligence models were tested for the simulation of aquifers in terms of the complex and nonlinear properties of groundwater systems. The present study stimulated the groundwater level and groundwater salinity parameters of Dezful-Andimeshk plain using ANN and ANN + GA models, and finally compared their results with measured data. The data collected for input to two models include meteorological data and groundwater quality parameters gathered from 2011 to 2018.

The results showed that the optimal model is to simulate ANN + GA (Artificial Neural Network + Genetic Algorithm) groundwater level with sigmoid tangent stimulus function, and the optimal model is to simulate ANN + GA groundwater salinity with sigmoid logarithm stimulus function. MAE and RMSE statistics have the minimum and  has maximum value for the model (In test phase, for the groundwater level RMSE=7.47, MAE=9.5 and R2=0.979 and for the groundwater salinity RMSE=6.81, MAE=7.74, and R2=0.99).

Therefore, optimizing the artificial neural network model using a genetic algorithm is very useful, effective and reduces errors and saves time and money.