PREDICTIVE MODELING FOR HAB MONITORING IN PHOENIX CANALS

The Phoenix canals transport water from the local reservoirs and rivers to treatment plants before being distributed to the community. They also provide a source of recreation to residents in the valley. The canals are stocked with two types of fish, white amur and carp. Blooms of filamentous cyanobacteria have been reported in the canals along with the associated byproducts, such taste and odor compounds. A time series of monthly sampling at multiple canal sites has been conducted since 1998 and parameters such as 2-Methlyisoborneol (MIB) and geosmin, nutrients, organic carbon, and temperature are publicly available through the Regional Drinking Water Monitoring Program. The mean concentrations of MIB and geosmin were 5.5 ng/L and 2.9 ng/L, respectively, in the canal waterways from 1998 to 2022 while filamentous cyanobacteria reached up to 1900 particles/ml during summer of 2002. This study will utilize this public data set to model byproduct hotspots in the Phoenix canals with the use of machine learning and spatial modeling. Principal component analysis selected parameters that are correlated with the occurrence of taste and odor compounds. As a monitoring tool for local resource managers, we created classification trees to determine the location of odorous water samples from environmental measurements and maps predicting compound hotspots. Trees were trained with 80% of data and tested against the remaining 20% to measure model accuracy. We achieved a model accuracy of 87% for MIB and 74% for geosmin.

Primary Presenter: Ashley Foster, Arizona State University (ashleynicolefoster16@gmail.com)

Authors:

Ashley Foster, Arizona State University  (anfoste4@asu.edu)

Susanne Neuer, Arizona State University  (Susanne.Neuer@asu.edu)

Morteza Abbaszadegan, Arizona State University  (Morteza.Abbaszadegan@asu.edu)

Peter Fox, Arizona State University  (Peter.Fox@asu.edu)