Modeling of land subsidence induced by groundwater withdrawal using Artificial Neural Network (A case study in central Iran)

Land Subsidence due to the groundwater over-exploitation is a significant problem in some areas which experience urbanization and expansion of agriculture and industry. In this study, the land subsidence of the Aliabad plain of Iran has been modeled using the artificial neural network (ANN) method. In this regard, a multi-layer perceptron has been used to model the land subsidence measured from Sentinel-1 images from 2015 to 2016. Groundwater dropdown, the thickness of alluvial sediments, the aquifer sediments' transmissivity, and elasticity modulus have been considered as four ANN model’s inputs variables and land subsidence as a single output. The results show that the ANN model has the ability to predict Aliabad subsidence with good accuracy (R2 = 0.74, R=0.94, RMSE= 0.02 m, MSE = 0.0006). Then a sensitivity analysis was performed in order to determine the impact of input parameters and the results indicate groundwater fluctuations as the most effective one. Model validation was achieved by comparing the ANN results with the calculated land deformation by DInSAR technique. An unused dataset including the four specified input parameters have been used, to assess the generalization of the ANN model. The model produces a proper prediction of land deformation with the new dataset.