Machine Learning for Water Quality Predictions

The transmission of Giardia spp. cysts to freshwater reservoirs is mediated by rainfall and snowmelt, but the impact on transmission remains unknown—especially for future climate regimes. Climate change has led to a 55-percent increase in heavy rainfall events in the northeast United States, and climate models indicate an increase in the occurrence of rain on snow events.

We used machine learning to provide for a statistical characterization of Giardia spp. transmission as a function of the variability of rain, rain on snow, and soil moisture. This approach demonstrated the relative impact and unraveled the mechanistic relationships of transmission within the noisy patterns of precipitation, soil moisture, and snowfall. We applied this approach to help explain the unusual increase in Giardia cysts which began in the Rondout reservoir in the fall of 2018, persisting through spring 2019. We anticipate this approach may be used to quantify the risk of similar unusual transmission events based on forecasted rainfall and pre-event measurements of soil moisture, temperature, and snowpack.

The ability to predict increased transmission of Giardia spp. in the watershed would be a valuable tool for water supply managers since operational changes can be made to prioritize reservoirs least affected by pathogen transport mechanisms.