Ibadan Journal of the Social Sciences
Volume 23, No. 2, 2025
DOI: 10.36108/ijss/5202.32.0251
Drought Hazard Mapping and Groundwater Yield Enhancement through AI-driven Predictive Analytics and Geospatial Multi-criteria Evaluation
Josiah Ayomide Oluwaleye1,2, John Faith Oyelakin1, Deborah Ayodele-Olajire2, Olalekan John Taiwo2
1Geographic Information System Unit, International Institute of Tropical Agriculture, Ibadan, Nigeria
2Department of Geography, University of Ibadan, Ibadan, Nigeria
Abstract
The integration of Artificial Intelligence (AI) with Geospatial Multi-Criteria Evaluation is essential in data processing automation and in predicting future trends. This study investigates drought hazard and identifies groundwater yield zones in Nasarawa State, Nigeria. Nine hydrogeological factors were analysed spatially by utilising data derived from remote sensing and ancillary sources. The factors include a decadal 12-month Standardized Precipitation Index (SPI-12), Groundwater Storage Anomaly (GWSA), Rainfall, lithology, lineament density, drainage density, slope, aspect, and land use/land cover (LULC). Image processing and rose diagram generation, which depict lineament orientations, were performed using R and Python programming languages.The consistency ratio derived from the Analytical Hierarchy Process (AHP) was 0.0197, indicating that the assigned weights and the matrix were significant and coherent. The SPI-12 analysis for ten years revealed near-normal to mild drought conditions across the state, with the eastern region experiencing more pronounced dry periods. The local government areas most impacted by drought are Lafia and Awe, located in the eastern region of the study area.The application of an AHP established GWSA as the primary driver of groundwater yield with a weight of 38%. The AHP-generated map provided a generalized overview, which indicates a low yield in the eastern part and a high yield in the northern part. Artificial Intelligence (AI) was integrated to further enhance the groundwater yield AHP results; the result obtained revealed more detailed and nuanced patterns that AHP could not identify. Random Forest and Extreme Gradient Boosting (XGBoost) achieved accuracies of 99% and 98% respectively, which indicates their strong performance in predicting groundwater yield. The Precision-Recall and other graphs indicate the superior performance of Random Forest over XGBoost. The XGBoost performed well but showed a high false positive rate of 61% for moderate groundwater yield zones and missed an instance of high groundwater yield, unlike Random Forest. The models identified Keffi, Kokona, and others as high groundwater yield zones, while Toto, Awe, and others were classified as low groundwater yield zones.
Keywords: Drought, ground water yield, GeoAI, analytical hierarchy process, Geographic Information System, remote sensing