Machine Learning: Predicting the Water Quality Impacts of Major Forest Fires

Jacobs created the machine learning tool using water quality and operational data collected before and after the Holiday Farm Fire, which began in September 2020 and burned almost 200,000 acres in the Willamette National Forest in Oregon. The fire affected 20% of the McKenzie River watershed, which supplies water to EWEB's 80MGD Hayden Bridge Water Treatment Plant (WTP) that serves 185,000 people. While EWEB implemented timely actions that minimized water quality impacts, such as erosion control measures and revegetation, increased concentrations of metals, nutrients, solids, bacteria and organic carbon were nevertheless recorded during storm events.

To create the machine learning tool, the team used advanced analytics to identify connections between measured parameters and treatment performance before the fire. These results informed the machine learning model to predict when and how WTP performance would be affected.

Machine learning has proven to be a useful tool in recent years to predict the performance of water treatment plants and optimize operations. Today, most water treatment plants are equipped with sensors that continuously collect data. The process of analyzing collected raw data is difficult and time consuming. However, machines can analyze vast amounts of data and use algorithms to model the complex relations between the various parameters and their effect on the treatment process. Machine learning can be used to create more effective water treatment processes and provide early detection of problems.