Application of machine learning to groundwater spring potential mapping using averaging, bagging, and boosting techniques

Abstract

Abstract Determining groundwater potential is vital for groundwater resource management. This study aims to present a comparative analysis of three widely used ensemble techniques (averaging, bagging, and boosting) in groundwater spring potential mapping. Firstly, 12 spring-related factors and a total of 79 groundwater spring locations were collected and used as the dataset. Secondly, three typical ensemble models were adopted to predict groundwater spring potential, namely, Bayesian model averaging (BMA), random forest (RF), and the gradient boosting decision tree (GBDT). The area under the receiver operating characteristics curve (AUC) and four statistical indexes (accuracy, sensitivity, specificity, and the root mean square error (RMSE)) were used to estimate the model's accuracy. The results indicate that the three models had a good predictive performance and that the AUC values of the GBDT, RF, and BMA were 0.88, 0.84, and 0.78, respectively. Furthermore, the GBDT had the best performance (accuracy = 0.89, sensitivity = 0.91, specificity = 0.87, and RMSE = 0.33) in terms of the four indexes, followed by RF (accuracy = 0.87, sensitivity = 0.91, specificity = 0.83, and RMSE = 0.36) and BMA (accuracy = 0.76, sensitivity = 0.87, specificity = 0.65, and RMSE = 0.49). This research can provide effective guidance for using ensemble models for mapping groundwater spring potential in the future.