DEVELOPMENT OF A PREDICTIVE MODEL FOR TASTE AND ODOR EPISODES IN REGIONAL DRINKING WATER RESERVOIRS

Taste and odor episodes are a major issue water utilities face in Alabama, which occur when high concentrations of odorous compounds, such as geosmin synthases, are detected in reservoirs, giving water a musty odor. Local utilities, including Auburn Water Works, Opelika Utilities, and Columbus Water Works, have each experienced taste and odor issues and identify them as high-priority problems due to humans being extremely sensitive to the compounds, being able to detect them by taste at less than 10 ppt. These episodes lead to public distrust, though the compounds are not harmful to humans. Predictive models have previously been made, though most tend to focus on algal blooms, although only a small subset of algae actually produce geosmin, rather than the actual compound synthases. The objective of this research is to develop a model for predicting taste and odor episodes, focusing on geosmin synthase, and which specific bacteria produce this synthase, rather than predicting the algal blooms in general.

The water utilities regularly collect water samples for characterization and regulation and agreed to collect additional samples to deliver to our research lab for further analysis. The samples were filtered for DNA extraction, so that qPCR and ion chromatography could be performed to examine the amount of geosmin synthase present, along with key anions in the water samples. This data was then entered into R Studio to develop a classification and regression tree (CART) model for predictive purposes. CART modeling gives an output that is a nonlinear decision tree, which was chosen due to the ability to show the nonlinear behavior of most aquatic systems.

Though the dataset is not complete, preliminary data has been entered into the software to run CART models for each utility with the water variables found. Our lab has confidence that these predictive models will have the power to allow utilities time to prepare responses to the compounds in the taste and odor episodes to prevent distrust from the public. In the coming months, a comprehensive dataset will be entered into the software and statistical measures will be evaluated to determine the fit of the model for each utility. Once these models are complete, manuscripts will be put together and can be used by the utilities along with the academic community in the future.