Groundwater potential mapping using machine learning models for Northeastern Karbi Anglong district, Assam, India

Groundwater is an important source of freshwater for mankind that has the unique capability of recharging itself naturally. Unfortunately, decades of over-exploitation has overwhelmed its ability to replenish, so much so that it has recorded unprecedented levels of decline bordering on depletion that is slowly becoming irreversible. In order to reverse the caused damage and prevent further worsening of the situation, it is important to understand the current status and strategize accordingly. The study was undertaken for identifying best model among a variety of machine learning models for groundwater potential mapping in the Karbi Anglong district, Assam, India. Seven groundwater influencing factors of altitude, slope, aspect, land use, normalized difference vegetation index, rainfall and proximity to rivers were chosen for the study. k-nearest neighbours, linear regression, logistic regression, naïve bayes, linear discriminant analysis, quadratic discriminant analysis, decision trees, random forest, support vector machine, gradient boosted trees, and artificial neural networks are the models that were employed in the study and were compared for finding the most accurate model for delineating groundwater potential in the region. The study found that random forest model achieved the highest accuracy of 91.67% and thus concluded it to be the model best suited for such a study and recommends it for similar other geospatial studies.