Insight to groundwater level prediction with feature effectiveness: comparison of machine learning and numerical methods

View largeDownload slide

View largeDownload slide

Close modal

This research aims to compare the accuracy of numerical model (NM) and machine learning (ML) methods for groundwater level prediction (GWLP). Additionally, principal component analysis (PCA) is applied to investigate the impact of each ML input feature on GWLP. GWLP was performed monthly from 2012 to 2019 in the Yazd-Ardakan Plain. The study area's aquifer data and features were analyzed and prepared to develop the conceptual model of MODFLOW and train ML algorithms for GWLP. Considering observation wells (OBWs), operation wells (OPWs), and their latitude and longitude as input features in convolutional neural networks (CNNs), support vector machine (SVM), and decision tree (DT) algorithms, GWLP was performed. The results demonstrate that the most accurate GWLP was achieved by SVM, with root mean square error (RMSE), correlation coefficient (R²), and area under the receiver operating characteristics (ROC) curve (AUC) values of 0.12, 0.90, and 0.94, respectively. For MODFLOW as an NM, RMSE and R² have been estimated with values of 0.84 and 0.68, respectively. Furthermore, PCA presented that the GWL had 71% effectiveness as the most significant feature in GWLP for ML algorithms. An analysis of the modeling results reveals that the ML algorithms provide more accuracy than MODFLOW.

• MODFLOW, a numerical model, is compared with machine learning (ML) methods for groundwater level prediction.

• The results demonstrate the better performance of the MODFLOW model in terms of generalization ability (GeA) and the superiority of ML methods in terms of accuracy in predicting groundwater levels.