Abstract:
Groundwater is an important source of pure water and its significance is quite high due to lack of surface water. The amount of surface water alone is not enough to meet demands of ever rising population and increased need of water due to technological advances and increased standard of livelihood. Also, Surface water is not in its purest form and contains a lot of impurities making the groundwater even more important. Hence the need of hour is to increase groundwater sources and manage them effectively for their sustainable growth. The work carried out in this project is to map groundwater potential zones. This mapping can prove to be beneficial in effective management, protection and exploration rich of groundwater prospects. GIS has been widely used in groundwater potential mapping but we ned to move forward and look for even better solutions. Hence, Machine Learning Techniques are incorporated to predict groundwater potential in accordance with GIS. Nine factors were used as effective factors. The factors that were used in our research are Slope degree, Slope aspect, Altitude, Plan curvature, Profile curvature, Topographic wetness index, Slope-length factor, Drainage density and Land use / Land cover. The models adopted here are Boosted Regression tree, Classification and Regression tree and Random Forest. The results of GIS and Machine learning integration for groundwater potential came out to be great increasing significance of machine learning in Groundwater potential mapping. The Area under curve (AUC) of three models namely BRT, CART and Random Forest came out to be 0.8409, 0.8791 and 0.8989 respectively. This concludes that best technique for prediction is Random Forest followed by Classification & Regression tree and Boosted Regression Tree. Since all the values are above 0.8, hence all models are useful for prediction. This research can prove to be helpful in usage of machine learning in various water resources applications.
