Document Type : Original Article
Authors
1
Department of Civil Engineering, Taft Branch, Islamic Azad University, Taft, Iran.
2
Department of Water Engineering, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran.
Abstract
Objective: The aim of this study is to simulate the fluctuations of the groundwater level of the Arak plain using MODFLOW model and GMDH neural network.
Method: In this paper, MODFLOW transient model is calibrated for an eight-year period (April 2006 to March 2014) with a monthly time step. Then, this model is validated for a two-year period (April 2014 to March 2016). The values of R2, NSE, and RMSE indices for the transient model are 0.9081, 0.7390, and 0.9226, respectively, while the values of these indices for the validation stage are 0.6783, 0.8948, and 0.9721, respectively. In the next step, the GMDH model is used to simulate the fluctuations of the groundwater level. In this case, 80% of the data are used for training the GMDH model and 20% of the remaining for testing the GMDH model. The values of R2, NSE and RMSE indices are calculated as 0.9319, 0.9192 and 0.2285 for the network training stage and 0.9817, 0.9865 and 0.2542 for the testing stage.
Results: According to the results of this study, even though both models have good efficiency for simulating groundwater level fluctuations, groundwater level fluctuations can be simulated more accurately using GMDH than MODFLOW model. But using the MODFLOW model, hydrogeological analyzes can be done more easily.
Conclusions: The results of this study show that when the purpose of modeling is only to simulate the groundwater level, the GMDH model is more suitable, but when the main purpose of the simulation is to investigate the hydrogeological conditions, the MODFLOW model is more suitable.
Keywords
Main Subjects
)