Groundwater Level Modeling and Forecasting using the Time Series Models (Case Study: the Plains of Hamadan Province)

Regarding the reliance of the agricultural and industrial sections and the drinking water on the groundwater resources in Hamadan province, the modeling and forecasting groundwater level fluctuations to utilize the resources is a basic necessity. One of the usual method in this way is the utilization of the time series models that give simply and clearly good short-term forecasts if the models are used in the correct way. Therefore, the raw data of piezometers in the plains of Hamadan province are taken and after the preprocessing job and using the Thiessen polygon, the time series of each plain is formed. The Mann-Kendall test showed deterministic trend in all the time series of the plains which consequently it is needed to detrend by excluding the trend term from the time series. Subsequently, the unit root test is carried out for whether the time series are stationary, and then using the Box-Jenkins method, seasonal ARIMA models are applied to the sample data and the bests are selected. Afterwards, the ARIMA models are used in the 12 months forecasting that gives the good out-of-sample forecasts, which in all the plains the lowest Pearson's correlation coefficient and the highest root mean square error are calculated 0.93 and 0.73 m, respectively, for the Hamadan-Bahar plain. Moreover, the best 12-months forecast is obtained in the Kaboudarahang plain with a Pearson's correlation coefficient of 0.99 and a root mean square error of 0.20 m.