Multivariate regression model for predicting oxygen reduction rates in groundwater for the State of Wisconsin

A multivariate regression model was developed to predict zero-order oxygen reduction rates (mg/L/yr) in aquifers across the State of Wisconsin. The model used a combination of dissolved oxygen concentrations and mean groundwater ages estimated with sampled age tracers from wells in the U.S. Geological Survey National Water Information System and previously published project reports from state agencies and universities. The multivariate regression model was solved using the Microsoft Excel solver, with 461 wells used for training and 46 wells held-out for validation. A total of 31 predictor variables were used for model development (56 were tested), including basic well characteristics, soil properties, aquifer properties, hydrologic position on the landscape, recharge and evapotranspiration rates, and land use characteristics. Model results indicate that the mean oxygen reduction rate for the training wells is 0.15 mg/L/yr (ranges from 0.07 to 0.59 mg/L/yr), with a root mean weighted square error of 3.13 mg/L/yr and Coefficient of Correlation (r^2) of 0.49 for the holdout validation data. This data release includes the Microsoft Excel file that represents the final solved regression model, as well as an Excel file that describes all of the predictor variables that were tested with the model.

[
  "010:12"
]

[
  "geospatial"
]