South-Central Section - 52nd Annual Meeting - 2018

Paper No. 10-2
Presentation Time: 1:50 PM

USING MACHINE LEARNING TO MAP PH AND REDOX CONDITIONS IN THE MISSISSIPPI EMBAYMENT REGIONAL AQUIFER SYSTEM


KNIERIM, Katherine J.1, KINGSBURY, James A.2, CLARK, Brian R.3 and HAUGH, Connor2, (1)U.S. Geological Survey, Lower Mississippi-Gulf Water Science Center, 401 Hardin Road, Little Rock, AR 72211, (2)U.S. Geological Survey, Lower Mississippi-Gulf Water Science Center, 640 Grassmere Park, Nashville, TN 37211, (3)U.S. Geological Survey, Lower Mississippi-Gulf Water Science Center, 700 W. Research Center Blvd., Fayetteville, AR 72701

Machine-learning methods are well suited to hydrologic studies because they allow input of continuous and categorical explanatory variables, can accommodate interactions among explanatory variables, and have performed better than linear regression methods. Machine-learning methods were used to map groundwater redox conditions and pH in depth zones of the Mississippi Embayment regional aquifer system that supply groundwater for domestic and public use to approximately 3 million people. Redox processes in this system exert important control on trace-metal solubility and groundwater pH varies based on aquifer materials and time of water-mineral interaction; thus both redox and pH provide important measures of groundwater quality. Surficial spatial datasets, groundwater-flow model output, and well characteristics were used as explanatory variables to model pH and dissolved oxygen in areas of the system that lacked groundwater-quality data. Surficial explanatory variables included soil data (texture, drainage class, permeability, and geochemistry), climate (temperature and precipitation), and land use. Groundwater-flow model variables included groundwater flux, groundwater age, flowpath length, groundwater altitude, water use, and hydrogeologic unit texture. Well characteristic variables included hydrologic position relative to major surface-water drainages and groundwater divides and well-construction data (length of and depth to top of the screened interval). There are limitations to using surficial variables alone to explain groundwater quality, particularly in confined aquifers. However, machine-learning methods may provide a mechanism to find analogs between surficial and groundwater-flow model variables so that groundwater quality can be more accurately predicted in aquifer systems that lack numerical groundwater-flow models or current and reliable water-quality data.