Spatial Prediction of the Groundwater Potential Using Remote Sensing Data and Bivariate Statistical-Based Artificial Intelligence Models

• Published in: Water resources management 2022-11, Vol.36 (14), p.5461-5494
• Main Authors: Ye, Yong, Chen, Wei, Wang, Guirong, Xue, Weifeng
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Atmospheric Sciences; Bayesian analysis; Bivariate analysis; Civil Engineering; Conditioning; Data analysis; Data mining; Decision trees; Earth and Environmental Science; Earth Sciences; Environment; Forests; Geotechnical Engineering & Applied Earth Sciences; Groundwater; Groundwater management; Groundwater potential; Hydrogeology; Hydrology/Water Resources; Mathematical models; Methods; Modelling; Numerical models; Parameters; Radial basis function; Rank tests; Regression models; Remote sensing; Resource management; Statistical analysis; Statistical prediction; Water resources
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-022-03307-w

Evaluation of the groundwater potential in a given region must be performed in groundwater resource management. The main purpose of this research was to simulate the groundwater potential in Wuqi County, China, based on five artificial intelligence models. The five data mining methods included the decision forest by penalizing attributes (forest PA) model, rudiment of the radial basis function network (RBFN) model, certainty factor (CF) model, logistic regression (LR) model and naïve Bayes (NB) model. According to local geographic environment characteristics, a total of sixteen conditioning factors were selected, the two-variable CF model was used to calculate the weight of factor subclasses, and multicollinearity analysis and the SVM method with different filter types were used to assess the independence and importance, respectively, of the conditioning factors. In addition, to obtain better model prediction results, the relevant modeling parameters of the forest PA and RBFN models were optimized via the tenfold cross-validation method in the modeling process. Finally, corresponding groundwater potential maps (GWPMs) of each model were generated, groundwater areas with a high potential were identified, and the consistency between the groundwater potential maps was compared by calculating the kappa statistical index. The area under the receiver operating characteristic curve (AUROC) was used to verify the accuracy and success rate of each model. The Wilcoxon signed-rank test method was also used to evaluate the significance of several benchmark numerical models. The results indicated that all five models achieved a satisfactory performance, and the models with optimized parameters (forest PA and RBFN models) realized greater predictive capabilities than those of the other unoptimized models. The results of this study could provide a certain guiding significance for rational management of groundwater systems and could contribute to the formulation of measures to ensure the best future use of groundwater energy.