Comparison of Artificial Neural Network and Kriging models in predicting groundwater quality parameters (SAR, TDS and EC) of Dezful Andimeshk plain

 Due to the complexities in the nature of ground water systems, it sounds like a demanding job to model either the time or the location of ground water. However, artificial neural networks have a high capability to model both complicated and non-linear models. Besides, Geostatistic Methods are, to a good extent, accurate in modelling ground water.

 The aim of this study is to simulate groundwater quality parameters (SAR, TDS and EC) of Dezful Andimeshk plain using ANN-PSO and geostatistical models. For this purpose, information from 61 observation wells in Dezful-Andimeshk plain has been used. Neural network model inputs including qualitative parameters SO42- ، pH ، HCO32-، Na+، Mg2+، Ca2+، TDS، SAR and EC were considered.

 The results of simulation with intelligent model showed that the highest accuracy of ANN-PSO model in simulation is related to EC, SAR and TDS parameters, respectively. The results of interpolation by geostatistical method showed that the highest accuracy of Kriging model in simulation is related to EC, TDS and SAR parameters, respectively. The general results obtained from the simulation of groundwater quality parameters showed that the ANN-PSO model is more accurate in simulating the groundwater quality parameters of the plain in Andimeshk than the Kriging model. So that the value of R2 for simulating SAR, TDS and EC parameters using ANN-PSO model in the test phase is 0.92, 0.918 and 0.955 respectively and using kriging model is 0.902. 0.915 and 0.931 were estimated.

 The results of this study also showed that the combination of intelligent models with optimization algorithms is used as a useful tool to simulate groundwater quality parameters.