A method to improve the stability and accuracy of ANN- and SVM-based time series models for long-term groundwater level predictions. (May 2016)

• Record Type: Journal Article
• Authors: Yoon, Heesung; Hyun, Yunjung; Ha, Kyoochul; Lee, Kang-Kun; Kim, Gyoo-Bum
• Is Part Of: Computers & geosciences. Volume 90(2016)Part A
• Page Start: 144
• Page End: 155
• Publication Date: 2016-05
• Subjects: Groundwater level -- Artificial neural network -- Support vector machine -- Recursive prediction -- Weighted error function; Environmental policy -- Periodicals; 550.5
• DOI: 10.1016/j.cageo.2016.03.002
• Languages: English
• ISSNs: 0098-3004

Abstract

Abstract: The prediction of long-term groundwater level fluctuations is necessary to effectively manage groundwater resources and to assess the effects of changes in rainfall patterns on groundwater resources. In the present study, a weighted error function approach was utilised to improve the performance of artificial neural network (ANN)- and support vector machine (SVM)-based recursive prediction models for the long-term prediction of groundwater levels in response to rainfall. The developed time series models were applied to groundwater level data from 5 groundwater-monitoring stations in South Korea. The results demonstrated that the weighted error function approach can improve the stability and accuracy of recursive prediction models, especially for ANN models. The comparison of the model performance showed that the recursive prediction performance of the SVM was superior to the performance of the ANN in this case study. Highlights: ANN- and SVM-based long-term groundwater level prediction models were developed. We proposed a weighting factor approach for recursive prediction models. The method can enhance the performance of the long-term groundwater level prediction model.