Comparison of Geostatistics, Artifitial Neural Networks and Adaptive Neuro-Fuzzy Inference System Approaches in Groundwater Level Interpolation (Case study: Ghazvin aquifer)

A precise Groundwater level interpolation is one of the attractive subjects in groundwater studies. In this study the performance of different approaches such as geostatistics (Kriging), neural networks (MLP and RBF) and adaptive neuro-fuzzy inference system (ANFIS) in a groundwater level estimation is examined in order to identify the best approach for interpolation. Ghazvin aquifer was chosen as the case study for this study. The coordinate of monitoring wells are used as input and groundwater level is used as output parameter in neural networks and adaptive neuro-fuzzy inference system. In all of approaches, 20 percent of monitoring wells are used as test data. The results showed that accurate predictions can be achieved with an adaptive neuro- fuzzy with R2=0.98. Duo to smaller amount of MSE, the MLP neural networks with R2=0.93 can give more accurate results than RBF (R2=0.9) and kriging methods (R2=0.95). The results also showed that geostatistic approach can give more accurate results than RBF neural networks.