Forecasting Spatiotemporal Water Levels by Neural Kriging Method in Tabriz City Underground Area

Groundwater level variations can essentially affect the execution of many engineering projects. Accordingly, due to the projects underway in Tabriz district and especially Tabriz Underground Project (METRO), spatiotemporal prediction of the groundwater level is crucial. Due to the aquifer complexity in the Tabriz area, there are problems in using classical mathematical models. In this research a combination of the artificial neural networks and Geostatistic models were applied as a new method for spatiotemporal prediction of groundwater levels using selected pizometers. For this purpose, the different neural networks were examined for groundwater level forecasting in central piezometer and an optimal ANN architecture was identified. This ANN structure was then used for modeling the selected piezometers. The results of these models were used as the inputs of the geostatistics model for forecasting spatial groundwater level in the study area. Two year monthly groundwater level prediction data in selected piesometers resulted by ANN modeling were among these input data. In order to obtain a high efficiency model, different methods of the geostatistic model were used. Finally the obtained model was tested by water level data in piesometers other than those used for model calibration. The results of this hybrid model were acceptable.