Characterizing non-E. coli coliforms as indicators of groundwater susceptibility via “big data”, geostatistical analysis, and machine learning

Abstract

Individuals reliant on private wells for drinking water face a significantly higher risk of acute gastrointestinal illness (AGI) compared to those serviced by public water systems, with approximately 1 in 10 Ontarians (~1.6 million people) belonging to this subgroup. Historically, groundwater microbial contamination studies have focused on E. coli presence/absence, with the roles of non-E. coli coliforms (NEC) and microbial magnitudes (CFU/100 mL) in assessing groundwater susceptibility consequently remaining understudied. This PhD thesis seeks to address these gaps via analyses of the Ontario Microbial Water Quality Dataset (OMWQD), which contains >1 million private well samples collected from ~300,000 wells between 2010 and 2021. Three distinct thesis objectives were designed and fulfilled, each analyzing a suite of groundwater quality parameters (i.e., NEC concentration, E. coli concentration, and the NEC:E. coli concentration ratio): 1) develop and evaluate provincial groundwater Contamination Indices which reflect 12 years of groundwater quality data; 2) characterize south Ontario contamination cycles and identify immediate and/or sustained effects of three extreme weather events on microbial contamination; and 3) develop a series of parameter-specific models to explain long-term groundwater contamination across Ontario. Contamination Indices were spatially compared to enteric infection rates and well density, with findings confirming that indices are not biased by rural population density and, based on statistically significant associations with infection rates, represent appropriate spatiotemporal reflections of long-term groundwater quality. Contamination cycle analyses identified E. coli concentration and the NEC:E. coli ratio as complementary metrics, with concurrent interpretation of their seasonal signals indicating that contamination via bypass mechanisms dominates winter months. Developed models suggest NEC may serve as appropriate indicators of the potential for generalized contamination, as the NEC model exhibited high goodness-of-fit on testing datapoints (91.9%). Overall, this work provides foundational evidence for the continued use of NEC as a groundwater quality indicator, improves our knowledge of spatiotemporal variations in contamination mechanisms across Ontario, and provides transferrable results due to the high number of private well samples, multi-year study period, and the study region’s diverse climate, geology, and topography.