Using machine learning to predict aquifers from groundwater chemistry data

Abstract

Systems and methods include a computer-implemented method for predicting locations of aquifers. Missing measurements in groundwater chemistry data from aquifers are replaced with calculated equivalents. The groundwater chemistry data is split into a model building/optimizing dataset and a model validation dataset. The groundwater chemistry data is normalized. A dimensionality reduction is performed on the groundwater chemistry data, including identifying principal components. An aquifer prediction model is generated and trained using the model building/optimizing dataset and machine learning. Performance of the aquifer prediction model is validated using the model validation dataset. Locations of aquifers are predicted during production of a new well using the aquifer prediction model and groundwater chemistry data for the new well. Settings, targets, and parameters of equipment used for drilling are changed based on the predicted locations of aquafers and the groundwater chemistry of the aquifers.