Proceedings of the 2023 Mississippi Water Resources Conference

Year: 2023 Authors: Temple J.M., Paul V.

Groundwater-derived phosphorus (P) has often been dismissed as a significant contributor towards surface water eutrophication, however, this dismissal is unwarranted, making the quantification of P concentrations in groundwater systems immensely important. Machine learning models have been employed to quantify the concentrations of various contaminants in groundwater, but to our best knowledge have never been used for the quantification of groundwater P. The goal of this research was to use a boosted regression tree framework to produce one of the first machine learning model of P variability in groundwater, with the High Plains aquifer serving as the study area. Boosted regression tree models that could explain and predict the statistical variance of P throughout the aquifer (under standard conditions) were developed, but with low predictive capacity. Observed P values from testing dataset compared to predicted values had a R2 of 0.7265, though this value could be skewed by better correlation between the actual and predicted at lower concentrations of P (< 0.05 mg/L). Lack of sufficient data and the raster file extraction method employed were identified as possible reasons for the low predictive capacity, and could be improved by more data. The research provides important variable correlation data that can potentially be incorporated into future studies that aim to further understand P dynamics in groundwater.