Predicting Water Quality Using Machine Learning

Bridget Armstrong (ESR), Theo Sarris (ESR), Louise Weaver (ESR), Helen Morris (University of Canterbury), Judith Webber (ESR), David Wood (ESR)

Groundwater is utilised for vital services, including drinking water, irrigation, and source water. It is imperative that groundwater resources are protected from increasing anthropogenic activities and that close monitoring of groundwater aquifers is undertaken. Currently, the status of groundwater is monitored by testing specific water quality parameters, including major ions, nitrate-nitrogen, ammonia-nitrogen, silica, iron, and manganese. Faecal indicator bacteria are also monitored as a proxy for pathogenic bacteria from faecal contamination.

Currently, these tests are all reactive and indicate a past problem if 'issues' are detected. There is a need for a fast, proactive method to assess the status of groundwater.

While groundwater systems can contain pathogenic bacteria, not all bacteria in this environment are harmful. Nonpathogenic bacteria communities in groundwater aquifers play an essential role in nutrient processing. The composition of these bacterial communities adapts to even subtle changes in the chemical composition of a groundwater system. Changes in bacterial communities could potentially be used as an early warning of a change in water quality, such as nitrate contamination.

Chemical parameters and environmental DNA (eDNA) were compiled from groundwater wells across New Zealand. The machine learning modellings

Classification and Regression Training (CART) and Random Forest were used to train various regression algorithms. Several bacterial Phyla were indicated as significant predictors of nitrate levels, i.e., Proteobacteria and Thaumarchaeota.