Enhancing spatial prediction of groundwater-prone areas through optimization of a boosting algorithm with bio-inspired metaheuristic algorithms

Abstract

Abstract Groundwater resources are essential for ensuring a consistent water supply in many regions. Groundwater potential maps (GPMs) can be utilized in many ways to estimate the quantity, quality, and distribution of subsurface water, supporting the decision-making processes of numerous stakeholders. This study contributes to improving the accuracy of GPMs, focusing on implementing Geospatial Artificial Intelligence (GeoAI) models. For this purpose, the accuracy performance of the Extreme Gradient Boosting (XGBoost) algorithm is improved in this study. To do this, two such popular metaheuristic algorithms, i.e., invasive weed optimization (IWO) and biogeography-based optimization (BBO), are integrated into the XGBoost algorithm for modeling and spatial prediction of the areas prone to groundwater. Three models—XGBoost, XGBoost-IWO, and XGBoost-BBO—are implemented within the Python programming environments to execute spatial modeling and generate predictive maps. The evaluation of results unfolds in two stages: model validation and GPM validation. For the training data, the root mean square error (RMSE) and mean absolute error (MAE) indices were 0.165 and 0.121 for XGBoost, 0.13 and 0.087 for XGBoost-IWO, and 0.114 and 0.082 for XGBoost-BBO, respectively. The test data showed similar trends, with XGBoost yielding RMSE and MAE values of 0.424 and 0.295, XGBoost-IWO at 0.416 and 0.287, and XGBoost-BBO at 0.39 and 0.28. XGBoost-BBO, XGBoost-IWO, and XGBoost had a prediction accuracy higher than other models. The respective area under the curve (AUC) of GMPs using receiver operating characteristic (ROC) curves for XGBoost, XGBoost-IWO, and XGBoost-BBO were 81.8 %, 83.1 %, and 83.7 %. Using bio-inspired metaheuristic algorithms, the GPM accuracy rate has improved further. The study of groundwater resources demonstrated how geological feature extraction by GeoAI may help employ advanced techniques.

Keywords

• Groundwater resource
• Spatial prediction
• Geospatial artificial intelligence (GeoAI)
• Bio-inspired algorithms
• Water scarcity