Machine-Learning Predictions of Redox Conditions in Groundwater in the Mississippi River Valley Alluvial and Claiborne Aquifers, South-Central United States

Scientific Investigations Map 3468
National Water Quality Program
By: , and 



Machine-learning models developed by the U.S. Geological Survey were used to predict iron concentrations and the probability of dissolved oxygen (DO) concentrations exceeding a threshold of 1 milligram per liter (mg/L) in groundwater in aquifers of the Mississippi embayment physiographic region. DO and iron concentrations are driven by and reflect the oxidation-reduction (redox) conditions in groundwater. Predictions from boosted regression trees, a type of machine-learning model, of iron concentration and DO threshold probability were used to categorize redox zones in the Mississippi River Valley alluvial aquifer (MRVA), middle Claiborne aquifer (MCAQ), and lower Claiborne aquifer (LCAQ). Model predictions indicated that DO concentrations greater than 1 mg/L are uncommon across the MRVA. DO events (where the predicted probability was greater than 0.5) tended to occur on the margins of the MRVA and in upland areas where MCAQ and LCAQ units crop out at the surface or are at shallow depth. Predicted iron concentrations were higher in the MRVA than in the MCAQ and LCAQ. Uncer­tainty in predicted iron concentrations tended to be high in areas where measured concentrations were also high, result­ing in small areas (encompassing less than 1.5 percent of the areal extent of the MRVA) of predicted iron concentrations that exceeded 100,000 micrograms per liter. Despite the large magnitude of overpredicted iron concentrations, the general proportion and spatial distribution of predicted iron concen­trations reflected observed concentrations in groundwater wells. Where the probability of exceeding a DO concentration of 1 mg/L was 0.8 or more and the iron concentration was less than 1,000 micrograms per liter, aquifers were catego­rized as oxic. Oxic conditions were mostly in the uplands where MCAQ and LCAQ units crop out at the margins of the modeled area. The MRVA was mostly anoxic, which was controlled by DO threshold probabilities less than 0.1. The predictions and redox zones support conceptual models of redox conditions in the Mississippi embayment. The MRVA is predominantly anoxic with high iron concentrations. In the Claiborne aquifers (including the MCAQ and LCAQ), groundwater flows along regional flow paths toward the axis of the Mississippi embayment (the approximate location of the Mississippi River), the residence time in the aquifer increases, DO is consumed, and iron concentrations generally increase. Elevated concentrations of trace elements, such as manganese and arsenic, are often associated with reducing conditions in anoxic and mixed anoxic zones, but other factors such as sediment mineralogy affect the occurrence and distribution of these constituents.

Suggested Citation

Knierim, K.J., Kingsbury, J.A., and Haugh, C.J., 2021, 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 3468, 16 p., 3 sheets,

ISSN: 2329-132X (online)

Study Area

Table of Contents

  • Abstract
  • Introduction
  • Hydrogeologic Setting
  • Methods
  • Water-Quality Data Results
  • BRT Model Results
  • Predictions of Dissolved Oxygen Threshold Probabilities
  • Predictions of Iron Concentration
  • Redox Zone Categorization
  • Summary
  • References Cited
Publication type Report
Publication Subtype USGS Numbered Series
Title Machine-learning predictions of redox conditions in groundwater in the Mississippi River Valley alluvial and Claiborne aquifers, south-central United States
Series title Scientific Investigations Map
Series number 3468
DOI 10.3133/sim3468
Year Published 2021
Language English
Publisher U.S. Geological Survey
Publisher location Reston, VA
Contributing office(s) Lower Mississippi-Gulf Water Science Center
Description Pamphlet: v, 16 p.; 3 Sheets: 34.3 inches x 24.7 inches or smaller; Data Release
Country United States
State Alabama, Arkansas, Kentucky, Louisiana, Mississippi, Missouri, Tennessee
Other Geospatial Claiborne Aquifer, Mississippi Rier Valley Alluvial Aquifer
Online Only (Y/N) Y
Google Analytic Metrics Metrics page
Additional publication details