The Geophysical & Artificial Intelligence Appraisal-Based Groundwater Potential Mapping of the Fullo Centre, Wajir North, using Magnetotelluric Profiles and Multi-Algorithmic Ensemble Models
The Fullo Centre in Wajir North, Northern Kenya, is situated in a delicate and dynamic interplay between the austere physical landscape of a metamorphic basement complex and a resilient human population facing catastrophic water scarcity. The area has experienced consecutive rainy season failures (a three-season drought), triggering ecological collapse, mass livestock mortality, and acute human health crises. The causes of water scarcity do not relate to the lack of precipitation alone, but rather to myriads of geomorphological, orogenic, and hydrogeological variables that prevent effective groundwater recharge and storage. A combined geophysical and Artificial Intelligence-aided mapping of the study area was deemed necessary as a desktop planning tool for borehole siting in the Fullo Centre. The final drilling site is georeferenced at 2.7215172° N and 39.526929° E, at an elevation of 471.19 meters above sea level. The objective of this study was to map depth profiles, resistivity variations, and deep fracture zones using Electrical Resistivity Tomography (ERT) and Audio-Magnetotellurics (ADMT) techniques, alongside a 15-algorithm Machine Learning (ML) ensemble. The ML ensemble processed engineered geophysical variables to predict the groundwater potential of 13 surveyed stations. Quantitative performance metrics revealed that the linear and regularized regression models were the most accurate predictors: Linear Regression and Ridge Regression both achieved perfect R² scores of 1.0000 on the test set, followed closely by Principal Component Regression (R² = 0.9997) and Lasso (R² = 0.9986). The Decision Tree (R² = 0.9850) and Random Forest (R² = 0.9467) performed well, while the Neural Network failed catastrophically (R² = -0.5953), demonstrating that excessive model complexity is counterproductive on small datasets. The consensus ranking, aggregating the top-1 votes across all 15 algorithms, showed that Station 7 received 12 out of 15 votes, Station 1 received 3 votes, and no other station received any top-1 votes. Station 7 was also the only station unanimously selected in the top 3 across all 15 models. Its superior performance is driven by its low deep-zone mean resistivity (61.6 ohm-m), high low-resistivity fraction (100%), and positive resistivity contrast (27%), all of which indicate highly fractured, water-saturated basement rock. Station 7 was subsequently drilled and confirmed as a productive water source, validating the predictive power of the ensemble. Summarily, this study demonstrates that simple linear models, when coupled with geologically informed feature engineering, provide the most reliable and interpretable results for siting sustainable boreholes in arid metamorphic basement terrains.
Authors
- Agatha Njuguna
- Meshack Owira Amimo
Publication Details
- Journal
- American Journal of Science and Advanced Technology
- Published
- 2026-11-12
- Primary Topic
- Geophysical and Geoelectrical Methods
- Type
- article
- Field-Weighted Citation Impact
- 0.00