Abstract Archive Select a year below to view:
Using Machine-Learning Models to Predict Concentrations of Nuisance Constituents in Groundwater of the Mississippi Embayment
Proceedings of the 2019 Mississippi Water Resources Conference
Year: 2019 Authors: Knierim K., Kingsbury J., Haugh C.
Machine-learning methods were used to map groundwater quality, including specific conductance (SC), total dissolved solids (TDS), and chloride, in the Mississippi River Valley Alluvial aquifer (MRVA) and the Claiborne aquifers of the Mississippi Embayment regional aquifer system. The MRVA aquifer is used widely for irrigation and locally for public supply, and high concentrations of chloride and iron can limit groundwater use. The middle Claiborne and lower Claiborne aquifers of the Mississippi Embayment are largely confined with few water-quality concerns, but higher salinity zones occur with depth as groundwater becomes more mineralized. Explanatory variables, including surficial spatial datasets (such as soil properties and land use), groundwater-flow model output (such as groundwater flux and age), and well characteristics (such as depth to screened interval) were used in boosted-regression tree models to predict SC and chloride concentrations throughout the aquifers at depth zones used for drinking water supply. TDS concentrations (which has a secondary maximum contaminant level of 250 g/L) were modeled using the correlation between SC and TDS. Surficial explanatory variables were attributed to individual wells using a 500-meter buffer, which for confined aquifers likely does not reflect the recharge zone for groundwater to the well. However, predicted values of hold-out data (not used to train the model) were in good agreement with observed values for SC models (r2 = 0.62). Important predictors included surficial variables such as land-surface elevation, landscape geomorphology, well position within the aquifer system, and land use. In particular, land-surface elevation may be a good indicator of whether the well screen was located in a confined or unconfined area of the aquifer system. Therefore, mapping groundwater quality across a confined aquifer system using surficial datasets is possible, and predicted concentrations were improved when groundwater-model variables were included as explanatory variables in machine-learning models.