A novel hybrid of support vector regression and metaheuristic algorithms for groundwater spring potential mapping
Depeng Wang,Kailong Huang,Xiwei He,Xu-Xiang Zhang,Yabing Meng
DOI: https://doi.org/10.1016/j.scitotenv.2021.152519
2022-01-01
Abstract:Limited groundwater resources and their overexploitation have become major challenges for sustainable development worldwide. In this study, an innovative hybrid approach was proposed to generate a groundwater spring potential map (GSPM) from the Sarab plain located in Lorestan Province, Iran, which includes the new best-worst method (BWM), stepwise weight assessment ratio analysis (SWARA), support vector machine learning method (SVR), Harris hawk optimization (HHO), and bat algorithms (BA). The first step involved the inventory of a map prepared to contain 610 spring locations. Randomly, 70% of the spring points were selected as training data, and the remaining 30% were selected for validation. Based on the review of the literature and available data, thirteen factors were generated as independent variables. The BWM and SWARA methods were used to identify correlations between the occurrence of springs and factors. Finally, using SVR-BA and SVR-HHO hybrid models, potential maps of groundwater springs were generated and then evaluated with receiver operating characteristic (ROC) and several statistical evaluators such as sensitivity, specificity, accuracy, and kappa index. Validation of the training data set showed that the success rates for the SWARA-SVR-BA, SWARA-SVR-HHO, BWM-SVR-BA, and BWMSVR-HHO models were 92.6%, 93.7%, 95.9%, and 96.4%, respectively. The results revealed that with a small difference, BWM-SVR-HHO performed better in training compared to other models. Evaluation of the prediction rate showed that the values of the area under the ROC curve for SWARA-SVR-BA, SWARA-SVR-HHO, BWM-SVR-HHO, and BWM-SVR-BA were 91.7%, 92.4%, 93.3%, and 94.7%, respectively. According to the results, although all models had excellent performance with more than 90% accuracy, BWM-SVR-BA was more accurate in predicting. The hybrid models presented in this study can be used as an accurate and effective methodology to improve the results of spatial modeling of the probability of groundwater occurrence. (c) 2021 Elsevier B.V. All rights reserved.