Predicting the Direction of Groundwater Flow Using Geospatial Data Analysis

Abstract

Groundwater, water flowing beneath the Earth's surface, provides the largest and most accessed source of freshwater. When groundwater is contaminated, the pollutant will disperse and travel in the same direction as the flow of groundwater, which directly threatens the integrity of drinking water and irrigation. All instances of groundwater contamination incur environmental, health, and monetary costs, but when not mitigated promptly, these costs can increase drastically. Currently, the method for determining the direction a contaminant plume will travel requires physically visiting the site and surveying the groundwater. This project addresses this issue by leveraging geospatial data and statistical learning methods. The aims of this project were two-fold. First, we aggregated known features, relevant to the direction of groundwater flow, at sites across the United States into a database. Having a centralized source of data regarding these properties is an improvement on the current system of sparse, disjoint, and at times inaccessible data sets. Second, we utilized that data in conjunction with machine learning techniques to develop a model that receives latitude and longitude as inputs and generates a prediction of the direction of groundwater flow at any location within the United States. Having accurate predictions directly improves efficiency by reducing response times and overall mitigation costs. We validated our model predictions against the known direction of groundwater flow using the smallest angle differences between the two.