A Comparison of AI Methods for Groundwater Level Prediction in Burkina Faso

---

Résumé

Groundwater serves as a valuable resource to supplement surface water, and its extensive utilization underscores the importance of precise groundwater level predictions. Burkina Faso confronts a critical challenge in the domain of sustainable groundwater resource management, underscoring the need for accurate forecasts of groundwater levels to enable efficient resource allocation and ensure long-term sustainability. This study introduces a robust framework that uses state-of-the-art Artificial Intelligence methodologies to predict groundwater levels across six strategically located piezometers in Burkina Faso’s Central Plateau region. The dataset combines piezometric Measurements, Rainfall, and vegetation indices that serve as a multifaceted feature space for model training. We systematically evaluated the performance of three specific machine learning models-NeuralProphet, XGBoost, and Long Short-Term Memory to determine which machine learning model offers the most robust predictions, enabling more effective and sustainable groundwater management. We observe that the XGBoost model outperforms its counterparts in terms of predictive accuracy. The findings of this study offer critical insights into the temporal variations in groundwater levels, thereby contributing to the formulation of more efficient water resource management strategies and facilitating data-driven decision-making processes in the target region.

Mots-clés

Groundwater, Environmental science, Water resource management, Geography, Geology, Geotechnical engineering