Assessing risk of groundwater pollution exposure from sea level rise in California

Sea level rise (SLR) will cause a groundwater table rise in coastal aquifers, and this can trigger exposure to toxic chemicals via direct contact with contaminated water or through vapor intrusion. This study presents deep learning- and explainable artificial intelligence (XAI) models to assess risk of polluted groundwater exposure due to SLR in California. Our method quantifies risk based on (1) SLR-related groundwater rise projections and (2) a prioritization of 2296 hazardous sites according to a geochemical indexing algorithm applied to groundwater data from the California State Water Resources Control Board. Our results indicate that nearly half a million people in California may be potentially exposed to marginal or poor groundwater by the end of the century. We show that feature collinearity renders numerous indicators on socioeconomics, demography, health and environment as dependent- and thus insignificant risk predictors. Our XAI models furthermore indicate that features on total population and features related to demographics (households of color and low income) and pollution source (proximity to wastewater dischargers) exert strongest controls on risk. However performance differences between randomly- and spatially cross-validated models highlight generalization challenges due to spatial autocorrelation. While the former model yields better results for California-specific analysis, only the latter model should be applied for unbiased risk assessments in other coastal regions.