부산대학교 도서관

The use of Remote Sensing (RS) and Geographic Information Systems (GIS) in modeling and predicting groundwater potential has become a vital first step in groundwater exploration. Demand for water in the semi-arid central Sokoto Basin is fast rising due to limited supply occasioned by poor borehole yields to outright failure. Groundwater data is not readily available in this area and the few available lack wide geographical spread. The aim of this study, therefore, is to integrate the outputs of heuristic and probabilistic models using a Union operation to improve groundwater potential prediction accuracy in a sub-basin with a limited geographical spread of aquifer data. Ten (10) groundwater conditioning parameters were extracted from reliable sources to produce thematic layers necessary for modeling. Thereafter, three (3) initial models; a Knowledge-based Analytical Hierarchical Process (AHP), a novel Information gain-based AHP, and Random Forest were produced, and five (5) groundwater potential zones were delineated. The model results generally revealed the high potential zones to be restricted to the western part of the basin and the low potential zones to the eastern part. Calculated accuracies of the models using the ROC curves vary from 62.7%, 62.8%, and 65.9% respectively. Following this, the initial model outputs were integrated using a Union operation that maximized the extensions of each potential zone and increased model accuracy to 68.2%. Overall, the models exposed the inefficiency of the current borehole drilling regime as most boreholes are sited in low potential zones.