Application and performance evaluation of time series, neural networks and HARTT models in predicting groundwater level changes, Najafabad Plain, Iran

Abstract

In this study, the application and performance of three models, including artificial neural networks, HARTT model and time series models for predicting the groundwater level in Najafabad plain were evaluated. For this purpose, SPSS, MATLAB and HARTT software packages were used to construct and evaluate each of these methods. Different parameters were used as inputs to estimate the groundwater level in each of these models. The results showed that all three models are capable of estimating the groundwater level with relatively good accuracy. However, the accuracy of seasonal ARIMA models and neural networks is higher than that of the HARTT model, though the performances of the two models are relatively similar. Based on the goodness of fit results, the seasonal ARIMA models are more appropriate for groundwater level prediction, because of the lower error, but the seasonal ARIMA models require some assumptions on the errors that make the modeling difficult, while the artificial neural network models have no such problems. But artificial neural networks have some limitations, such as the lack of a comprehensive theory and the need for a large number of observations due to the nonlinear structure of the networks. In addition, neural network inputs, like coefficients of regression models, are not interpretable and act as a black-box model. Therefore, they are only suitable for prediction and are not applicable for analyzing and interpreting the effects of each parameter on the groundwater level. Also one of the benefits of ARIMA models is that it can be used for short-term groundwater level prediction without the need for input data. This is especially important for areas without observation data or areas where data are limited. Furthermore, the results show that in addition to the ability of groundwater level prediction, the HARTT model can predict the effects of precipitation changes on groundwater level and the time delay between precipitation and its effect on the groundwater level, which this delay for different wells, based on the different effective parameters, is different.