Comparison of SVM and ANN models in simulation the Groundwater level of observation wells in Nahavand Plain - Hamedan Province

In this research, support vector machine (SVM) and artificial neural network (ANN) were used to simulate the monthly Groundwater level of Nahavand plain located in Hamedan province. Twenty years data (1997-2017) were used by engaging Matlab software. From these data, 14 years were used for training, 3 years for calibration and finally 3 years used for model validation. The statistical comparison of the results was also attempted with the aid of correlation coefficient (r) and standard error (SE). Four observation wells were used along with the variables of Groundwater level, precipitation, evaporation and temperature to simulate the Groundwater level. The highest accuracy among these two models is SVM model, which has SE = 0.11 in the training mode and 0.03 in the test mode. Also, the correlation coefficient in the test mode is 98%. Considering the appropriate accuracy of SVM method in simulating the groundwater level, the results indicate that, the RBF kernel function, with a variance of 6523 and a gamma of 527.23 for the optimal mode. The results of this study showed by using the SVM approach is more realistic in simulation of groundwater level and evaluation of the input parameters. In addition this could help in reducing the number of input parameters as well as can show appropriate accuracy for the simulation.