Groundwater Level Prediction in Golpayegan Aquifer Using ANFIS and PSO Combination

Groundwater level prediction is an essential priority for planning and managing groundwater resources. This study aimed to compare the accuracy of the Neuro-Fuzzy Adaptive Inference System (ANFIS) model with the ANFIS model combined with particle swarm optimization algorithm (ANFIS+PSO) in predicting the monthly groundwater level of Golpayegan aquifer during 2002-2019. For this purpose, monthly data on rainfall, temperature, pan evaporation in the selected meteorological stations, discharge volume of exploitation wells and groundwater level of observation wells have been used. After spatial and temporal analysis, four observation wells with two input data structures (S1 and S2) were selected to predict the groundwater level. The results of trend and homogeneity tests show a 99% significance of groundwater level changes in the selected observation wells 4, 8, 19 and 20 with a sudden drop of 22, 17, 27 and 2 meters before and after June, September, July and August 2010, respectively. The highest and the lowest accuracy of groundwater level prediction is related to observation wells 20 and 4 with root mean square error values (RMSE) of 2.37 and 0.21 m, respectively, related to ANFIS_S1 and ANFIS + PSO_S2 models. Generally, the results of this study indicate that the selection of appropriate structure of input data is more effective than the combination of two models (ANFIS and PSO) in increasing the accuracy of groundwater level prediction. So, that the optimal structure of input data and the combination of optimized algorithm model have increased the accuracy of groundwater level prediction, 44% and 25%,  respectively.