In Tunisia, water resources are already limited in space and time. Their sustainable management is a national priority for ensuring water security and development of the country. The Medjerda river basin is the largest watershed in Tunisia, where groundwater resources are used in conjunction with surface water. While local surface water resources are relatively well managed, groundwater resources are more hidden and difficult to conceptualize; additionally, they have the gap of groundwater data. Hydrogeological mapping of groundwater resources is one of the main tools for the controlled development of groundwater resources. Remotely sensed surface indicators of groundwater provide useful data where practical classical alternatives are not available. Integrated Remote Sensing (RS) and Geographic Information System (GIS) are widely used in groundwater mapping. Locating potential groundwater targets is becoming more convenient, cost-effective than invasive methods and efficient with the advent of number of satellite imagery. The nature of Remote Sensing-based groundwater exploration is to delineate all possible features connected with localization of groundwater. The main goal of this study is to investigate the machine learning models for Groundwater Potential Mapping (GPM) using GIS and RS at the Medjerda river basin.
This study includes the analysis of the spatial relationships between Transmissivity and various conditioning factors such as elevation, slope, curvature, river, lineament, geology, soil, rainfall, and land use. Eighteen groundwater-related factors were collected and extracted from topographic data, geological data, satellite imagery, and published maps. About 60 groundwater data of transmissivity were randomly split into a training dataset 70 % for training the model and the remaining 30 % was used for validation purpose.
Subsequently, GPM was produced using weights-of-evidence with logistic regression and functional tree models, classified as very high, high, moderate, low, and very low zones. Finally, the Receiver Operating Characteristic (ROC) curves for all the groundwater potential models were constructed and the areas under the curves (AUC) were computed. These results of GPM can be helpful for future planning in groundwater resource management and land use planning.