Clustering Observation Wells Network and Forecasting Groundwater Level by Artificial Neural Networks (Case Study: Marageh Plain)

The purpose of this study was to cluster the observation well networks in Marageh Plain in East Azarbaijan and to predict the groundwater level by artificial neural networks. Primarily, by Hierarchical WARD clustering method, 20 observation wells of Maragheh Plain with over 15 years data period were clustered. Then, a cluster with 3 homogenous subclusters was selected and the representative of each subcluster was determined. Artificial neural networks with a multilayer perceptron structure utilizing back-propagation algorithm was used to simulate the representative groundwater level of each subcluster. The results indicated considering monthly temperature data as input for the artificial neural networks caused disorder of the network while considering lag time for the input data increased the accuracy of the estimated groundwater levels. Based on the results, the minimum and maximum RMSE between the observed and calculated values were 0.26 m and 0.63 m, respectively. Also the Maximum and minimum quantities of R2 were 0.86 and 0.82, respectively.