Spatio-temporal Analysis of Groundwater Level Using Clustering Method Combined with Artificial Neural Network

Long-term planning and proper management of groundwater resources utilization are essential to ensure a reliable supply of water to countries, especially in arid and semi-arid regions. Therefore, it is necessary to employ appropriate models to predict the spatial and temporal fluctuations of aquifers and their future behavior. This study aimed to apply zoning strategies to Miandoab aquifer and predict its spatial and temporal groundwater level using an artificial neural network. First, the six parameters of transmissivity coefficient, groundwater level, ground elevation, withdrawal, rainfall, and discharge were spatially clustered to identify their effect on the simulation model. Three clustering approaches of single-parameter, three-parameter and integrated-parameter were evaluated using some statistical indices. The number of suitable clusters was determined using silhouette width. Groundwater level data (2002-2012) from 77 observational wells were used for model training and validation. Results showed that the correlation clustering approach performs better than the other methods. Precipitation, aquifer recharge, aquifer discharge, and groundwater level of the previous month were inputs to the back-propagation artificial neural network (ANN) for predicting a two-year period of groundwater level. The results showed that the correlation coefficients of variation in 6 clusters were 0.71- 0.97, and the RMSE variations were 0.19 - 0.58, indicating appropriate accuracy of this approach for predicting groundwater level.