A comparative study of artificial neural networks and support vector machines for predicting groundwater levels in the Ziveh Aquifer–West Azerbaijan, NW Iran

In many parts of the world, especially where surface water resources are rare or not available, groundwater as the largest source of freshwater is used for domestic, agricultural, and industrial water needs. Groundwater use has increased dramatically which has led to groundwater depletion with negative effects. To protect these water resources, their monitoring and management are essential. Long-term groundwater level (GWL) data from individual wells provide the information needed to monitor groundwater depletion locally (for aquifers) and then be compiled into regional and national. One of the most important approaches in groundwater resource management (GRM) is to achieve a suitable model for predicting GWL behaviour or estimating the parameter that affects it. In this study, artificial intelligence approaches include artificial neural networks (ANNs) in two different types of recurrent neural networks (RNNs) and feed-forward neural networks (FNNs), as well as support vector machines (SVMs), used to predict groundwater levels (GWLs) of the Ziveh Aquifer. Parameters including precipitation, temperature, and water levels of seven piezometers with monthly periods have been considered input data, and water levels during the same period have been used as the model’s output for a 14-year statistical period (2005–2018). Root mean square error (RMSE) and correlation coefficient ( R 2 ) have been used to study models and compare their efficiency. Despite the inherent potential of each model in the prediction of water levels, the results suggest the relative superiority of SVM compared to ANN. The RMSE results for two SVM training and testing steps were 0.36 and it was equal to 0.38 and 0.41 for FNN and RNN models respectively. To achieve a higher efficiency rate of artificial intelligence models, the use of more dependent variables and different algorithms for future study is recommended.