Groundwater Vulnerability Model

Oregon Department of Environmental Quality

The Oregon Department of Environmental Quality (DEQ) was seeking empirical, data-driven approaches to focus its limited resources on efficiently identifying potential sites that are more likely to have groundwater contamination. DEQ, with assistance from Apex Companies and Exa Data & Mapping, developed a statistical model to predict the probability of groundwater contamination. First, scientists at DEQ and Apex developed a statewide dataset consisting of groundwater well contamination status from chlorinated volatile organic compounds (cVOCs) from the State’s Ambient Water Quality Monitoring System (AWQMS), along with spatial attributes that may be associated with an increased probability of contamination.  Exa joined the team in 2022 to assist with the development, testing, and refinement of the statistical model used to predict well contamination.  The dataset contained approximately 1,700 wells. Preliminary analysis used logistic regression models (LRM), but the method found to be most promising was a random forest model (RFM) – an ensemble method that improves prediction accuracy by aggregating across thousands of classification tree model results.   The RFM resulted in high sensitivity and specificity in predicting the cVOC contamination status of the observations in the testing data. 
 
With the RFM, covariates obtained for any location (e.g., distance from contaminated sites, land use, and soil characteristics available from public databases) can be used to predict contaminant potential for any point within the state of Oregon. A map of contamination potential across the state can then be used to focus groundwater sampling for cVOCs to prioritize locations of high groundwater vulnerability for site assessment. Application and testing of this model are currently in process. This step involves scripting in ArcPython to generate the data layers from the public databases, aligning these layers to allow covariates to be estimated on a grid across the state of Oregon.  Exa is tasked with troubleshooting and streamlining the ArcPython and R modelling scripts to allow ODEQ to update and implement the model inhouse in the future.