Performance Comparison of Artificial Neural Network, Time Series and ANN-ARIMA For Groundwater Resources Index (GRI) Forecasting (Case Study: South of Qazvin Province)

Groundwater drought is one of the drought types that caused by lack of sufficient groundwater recharge. Groundwater Resources Index (GRI) is a method to express the state of this type of drought using ground water level data. Various methods and models have been presented in order to forecast and model, but selecting a reliable model is a difficult task. So, it would be better to use a combination of acceptable models instead of using just one model. In this study, the GRI values over 1984-2011 period were calculated in south of Qazvin province and its relationship with meteorological parameters such as precipitation, discharge, evapotranspiration, temperature (Mean, Max, Min) and large scale climate signals (MEI, SOI, AMM, AMO, PDO) was modeled by artificial neural network based on the Gamma test and in three structures. The results show that SOI and temperature have higher significant correlation with GRI values and also using the meteorological parameters as input parameters lead to improving the artificial neural network performance. Moreover, the ARIMA (1, 1, 3) (2, 0, 1) was selected for forecasting of GRI based on evaluation measures such as AIC and SBC. Finally, ANN-ARIMA modeling revealed better performance compared with the ANN and ARIMA(R2=0.94, RMSE= 0.05).