Periodic and autoregressive models of groundwater level dynamics

Abstract

In this paper, authors examine the groundwater level oscillation at groundwater station Bogatić near Šabac (Serbia), for the period 2002-2014. Analysis is done by transforming the observation data into Fourier series within fundamental period of 5 years, and by deriving a model based on the autocorrelation properties of the recorded data. One should note that this is the first time that such analyzes are preformed for the groundwater level oscillation observed in Serbia. The derived model based on groundwater level periodicity is represented by a combination of sine and cosine waves. Analysis shows rather satisfying statistical accuracy of the model, with R 2=0.707 and mean absolute error of 0.34m, which fails to capture the highest values of the groundwater level for the analyzed period. Second model is expressed in a form of a simple correlation between the two successive observations, with a time delay τ=1 and 4. Both models are derived for the period 2003-2009, and their prediction accuracy is checked for the period 2010-2014. Both of these models provide more accurate predictions in comparison to the model based on the groundwater level periodicity.