Machine-learning model predictions and rasters of dissolved oxygen probability, iron concentration, and redox conditions in groundwater in the Mississippi River Valley alluvial and Claiborne aquifers

Groundwater is a vital resource in the Mississippi embayment physiographic region (Mississippi embayment) of the central United States and can be limited in some areas by high concentrations of trace elements. The concentration of trace elements in groundwater is largely driven by oxidation-reduction (redox) processes. Redox processes are a group of biotically driven reactions in which energy is derived from the exchange of electrons. In groundwater, this commonly occurs through decomposition of organic matter (carbon) by microbes, which consumes dissolved oxygen (DO). Under low DO conditions, iron (Fe), manganese, and arsenic can dissolve from coatings on aquifer sediments and be released into groundwater. Therefore, predictions of redox conditions (using DO and Fe) are important in the Mississippi embayment for a better understanding of the potential zones of high trace elements in drinking-water aquifers. The Mississippi embayment includes two principal regional aquifer systems: the Quaternary Mississippi River Valley Alluvial aquifer (MRVA) and the Mississippi embayment aquifer system, which includes deeper Tertiary aquifers and confining units. Based on the distribution of groundwater use for drinking water, the modeling focused on the MRVA, the middle Claiborne aquifer (MCAQ), and the lower Claiborne aquifer (LCAQ). Machine learning was used to predict redox conditions?including the probability of exceeding a DO concentration of 1 milligram per liter (mg/L) and Fe concentrations?across the MRVA, MCAQ, and LCAQ. Boosted regression tree (BRT) models (Elith and others, 2008; Kuhn and Johnson, 2013) were developed to predict DO probability and Fe concentration to 1-kilometer (km) raster grid cells of the National Hydrologic Grid (Clark and others, 2018) for 7 aquifer layers (1 MRVA, 4 MCAQ, 2 LCAQ) following the hydrogeologic framework of Hart and others (2008). Explanatory variables for the BRT models included attributes associated with well location and construction, surficial variables (such as soils and land use), and variables extracted from a MODFLOW groundwater flow model for the Mississippi embayment (Haugh and others, 2020a; Haugh and others, 2020b). Output from DO and Fe models were used to classify redox zones, including anoxic, mixed anoxic, mixed oxic, and oxic conditions. Oxic conditions included areas where the probability of exceeding a DO concentration of 1 mg/L was greater than 80 percent and iron was less than 1,000 ?g/L. Anoxic conditions included areas where the probability of exceeding a DO concentration of 1 mg/L was less than 10 percent. Mixed conditions include anywhere that the predicted DO probability was greater than or equal to 10 percent and less than or equal to 80 percent, and either less than 500 ?g/L iron (mixed oxic) or greater than or equal to 500 ?g/L iron (mixed anoxic). Prediction intervals were calculated for DO and Fe by bootstrapping raster-cell predictions following methods from Ransom and others (2017). For a full description of modeling workflow and final model selection see Knierim and others (2020). References for abstract: Clark, B.R., Barlow, P.M., Peterson, S.M., Hughes, J.D., Reeves, H.W., and Viger, R., 2018, National-scale grid to support regional groundwater availability studies and a national hydrogeologic framework: U.S. Geological Survey data release, https://doi.org/10.5066/F7P84B24. Elith, J., Leathwick, J.R., and Hastie, T., 2008, A working guide to boosted regression trees: Journal of Animal Ecology, v. 77, no. 4, p. 802?813. Hart, R.M., Clark, B.R., and Bolyard, S.E., 2008, Digital surfaces and thicknesses of selected hydrogeologic units within the Mississippi embayment regional aquifer study (MERAS): U.S Geological Survey Scientific Investigations Report 2008?5098, https://doi.org/10.3133/sir20085098. Haugh, C.J., Killian, C.D., and Barlow, J.R.B., 2020a, MODFLOW-2005 model used to evaluate water-management scenarios for the Mississippi Delta: U.S. Geological Survey data release, https://doi.org/10.5066/P9906VM5. Haugh, C.J., Killian, C.D., and Barlow, J.R.B., 2020b, Simulation of water-management scenarios for the Mississippi Delta: U.S. Geological Survey Scientific Investigations Report 2019?5116, http://pubs.er.usgs.gov/publication/sir20195116. Knierim, K.J., Kingsbury, J.A., and Haugh, C.J., 2020, Machine-learning predictions of redox conditions in groundwater in the Mississippi River Valley alluvial and Claiborne aquifers, South-Central United States, U.S. Geological Survey Scientific Investigations Map XXXX, https://doi.org/10xxxxx/xxxx. Kuhn, M., and Johnson, K., 2013, Applied Predictive Modeling: Springer, New York, New York, 595 p. Ransom, K.M., Nolan, B.T., Traum, A.J., Faunt, C.C., Bell, A.M., Gronberg, J.A.M., Wheeler, D.C., Rosecrans, C.Z., Jurgens, B., Schwarz, G.E., Belitz, K., Eberts, S.M., Kourakos, G., and Harter, T., 2017, A hybrid machine learning model to predict and visualize nitrate concentration throughout the Central Valley aquifer, California, USA: Science of The Total Environment, v. 601?602, p. 1160?1172, https://doi.org/10.1016/j.scitotenv.2017.05.192.