The efficiency of Data-Driven Models for Months ahead Groundwater Level Forecasting Using a Hybrid Gamma Test and Genetic Algorithm Model

In order to implement sustainable groundwater resources management, it is necessary to model the behavior of groundwater level. Groundwater is a nonlinear and complex system which Data-driven models can be modeled this system without approximation and simplification. This study evaluates the performances of data-driven models (Support Vector Regression (SVR) and Artificial Neural Network (ANN)) for forecasting groundwater levels of Hashtgerd plain at 1, 2 and 3 months ahead. One of the most important steps in constructing data-driven models is determining the optimal combination of input variables. In this paper, the optimal composition was determined using the hybrid of gamma test and genetic algorithm, also the optimal length of input composition in teaching period was determined using the M-test method. The results of this study showed that the models for 1 month ahead performed better than for 2 and 3 months ahead forecasts and the accuracy of the models decreased as months ahead increased. The Developed Discrepancy Ratio (DDR) was applied to evaluate the efficiency of models. The results showed that the ANFIS model had the best performance for one month ahead groundwater level forecasting. Also, the results demonstrated that the ANFIS and MLP models were able to provide better performance in groundwater levels forecasting for a shorter-time period (e.g., 1 month ahead). The performance of the SVR model for a longer-time period (e.g., 3 month ahead) was superior to the MLP and ANFIS models.