Anthropogenic and Geogenic Factors Affecting Groundwater Arsenic and Potential Drinking Water Exposure in the San Luis Valley, Colorado

Abstract

Groundwater arsenic is a serious health threat in many regions with limited surface water drinking supplies. In the San Luis Valley, Colorado, USA, where most residents rely on groundwater for drinking water, elevated groundwater arsenic concentrations (≥5 μg/L) were observed in 19.8% of private wells. Many recent studies have leveraged data science methods to estimate the spatial variability of arsenic at local, regional, and global scales. These studies typically predict depth-independent arsenic concentrations due to limited well depth information. However, arsenic concentrations may vary greatly with depth due to heterogeneous aquifer geochemistry, including redox potential, pH, and biogeochemical conditions. In this study, we apply random forest modeling to predict elevated groundwater arsenic concentrations (≥5 μg/L) over space and depth continuously across the San Luis Valley using a community-driven data set of 435 groundwater measurements of major ions and trace metals. Our model reveals the influence of several known or hypothesized drivers or proxy drivers of arsenic in groundwater, including surface elevation, fraction of fine-grained aquifer sediments, well depth, soil pH, long-term subsidence, and geothermal influences. Elevation, percentage of fine-grained sediments, and elemental indicators of geothermal fluid influence are some of the most important predictors of arsenic, while long-term subsidence is found to have a weak but positive relationship with arsenic. The model has a specificity of 0.82 and a sensitivity of 0.81 on held-out test data sets. In addition to improving understanding of drivers of arsenic in the region, this model produces pseudo-3D estimates of groundwater arsenic. The model can also be used to assess the time-dependent exposure to arsenic as the depth of domestic wells changes over time.