Groundwater level Forecasting in Tassuj Plain-East Azarbaijan by artificial neural networks

The main water resource in arid and semi-arid regions is groundwater. Therefore, its study is important. The purpose of this study was groundwater level forecasting in Tassuj-Azarbaijan Plain in the future by artificial neural networks. Used data was from October 2002 to September 2013 (11 years). After determining boundary of Tassuj plain, water and observation wells where located out of boundary, was deleted. Then by Tissen method, polygons of observation wells were determined and input and output water into each polygon through rainfall, evaporatin and water wells were calculated. Finally, by designing 4 different architectures, the best network to forecast groundwater level was determined. Multilayer perceptron with back propagation error algorithm was used to simulate and forecate groundwater level. The results showed, considering monthly air temperature as input data in networks, was confused them. Network with 1 to 5 time lags in input data was the best. On the best network, values between calculated and observed data in 10 observation wells were greater than 90% and RMSE values were less than 80 cm. Finally, groundwater level was predicted for 24 next months (from October 2013 to September 2015). On the forecasting data, falling the groundwater level will be continued in the 24 next months as much as 1.3m.