Delineation of groundwater resources of a given area is importance for management of groundwater resources. This is often done manually by combining various geo-scientific datasets in Geographic Information System (GIS) environment, which is time consuming and is prone to subjective bias and also suffers from other human induced uncertainties and difficult to cope with increasing volumes of data. The explosive growth of data leading to ‘rich data, but poor knowledge’ dilemma yet we have challenges to be solved. Artificial Intelligence (AI) has been successfully used in fields such as robotics, process automation in engineering, industry, medical and domestic households. Artificial Intelligence tool have the able to bridge this gap by augmenting the human capabilities in understand science far better than before. Incorporating AI into groundwater potential mapping greatly improves computation speed, reduces the subjectivity nature of manual mapping and lessens human induced uncertainties. The software platform includes artificial intelligence algorithms such as artificial neural networks, support vector machines, random forest, index-overlay and fuzzy logic.
The software platform is semi-automatic to allow the user to control some of the processes yet automating the other processes. The possible inputs to the AI for training includes; aquifer types, topographic slope, lineament and drainage density, land-use / land-cover (LULC), distance to lineaments, distance to streams and soil clay content. Yield values of selected boreholes are used as training outputs.
The software was tested using data gathered for the area surrounding Maluti-a-Phong in the Free State Province of South Africa. The area was chosen because of recent drought which has hit the country and local municipalities are searching for groundwater resources for building wellfields to supply local communities with fresh water. The groundwater potential map of the area was validated using borehole yield values of boreholes which were not used for modelling. Good correlation values as high as 0.85 was obtained between model values and borehole yield. The final groundwater potential map was divided into four zones; very good, good, poor and very poor. Based on this study, it is concluded that the high groundwater potential zones can be target areas for further hydrogeological studies.
The usage of the software proved to be efficient in minimising the time, labour and money needed to map large areas. The results of which can be used by local authorities and water policy makers as a preliminary reference to narrowed down zones to which local scale groundwater exploration can be done. AI should be viewed as augmented intelligence as it aid the decision-making process rather than replacing it. Data-driven approaches should also be knowledge-guided for efficient results.