Machine Learning for Early Warning of Cyanobacteria Blooms in Lake Champlain
Conference Year
January 2020
Abstract
Cyanobacteria blooms are a major problem in Lake Champlain, producing toxins that can harm swimmers, poison pets, and disrupt aquatic ecosystems. Every summer, bays along the lakeshore see recurrent blooms that close beaches and endanger public health. Blooms are driven by the complex interactions of multiple factors. Phosphorus is thought to be a key driver, but nitrogen, water temperature, and mixing by wind and weather events may all play a role.
Untangling these multiple drivers makes it difficult to understand, let alone predict, the formation and timing of blooms. While severe blooms can be identified visually, measuring cyanobacteria levels in water samples is time and resource intensive, making it impractical to track levels on a daily basis or catch rising levels before a bloom becomes visible. Monitoring cyanobacteria would be simpler if levels could be reliably predicted from other factors that are easier to measure. Predicting blooms a few days in advance would benefit public safety by giving municipalities advanced warning to prevent or better-manage blooms and their public health impacts.
Machine learning algorithms can find patterns among multiple variables that might be missed by traditional statistical methods. They may lend insight into this problem by helping to identify combinations of factors linked to bloom formation, or even allowing us to forecast blooms a few days in advance by using data on current water and weather conditions. We will apply a variety of machine learning methods to a unique, public, long-term data set produced by the VT Department of Environmental Conservation. We will analyze ten years of data on water quality and cyanobacteria levels from around Lake Champlain with the goal of predicting cyanobacteria levels from easily measured water quality indicators.
Primary Faculty Mentor Name
Gillian Galford
Faculty/Staff Collaborators
Tim Laracy, Wilton Burns
Status
Graduate
Student College
Rubenstein School of Environmental and Natural Resources
Program/Major
Natural Resources
Primary Research Category
Vermont Studies
Machine Learning for Early Warning of Cyanobacteria Blooms in Lake Champlain
Cyanobacteria blooms are a major problem in Lake Champlain, producing toxins that can harm swimmers, poison pets, and disrupt aquatic ecosystems. Every summer, bays along the lakeshore see recurrent blooms that close beaches and endanger public health. Blooms are driven by the complex interactions of multiple factors. Phosphorus is thought to be a key driver, but nitrogen, water temperature, and mixing by wind and weather events may all play a role.
Untangling these multiple drivers makes it difficult to understand, let alone predict, the formation and timing of blooms. While severe blooms can be identified visually, measuring cyanobacteria levels in water samples is time and resource intensive, making it impractical to track levels on a daily basis or catch rising levels before a bloom becomes visible. Monitoring cyanobacteria would be simpler if levels could be reliably predicted from other factors that are easier to measure. Predicting blooms a few days in advance would benefit public safety by giving municipalities advanced warning to prevent or better-manage blooms and their public health impacts.
Machine learning algorithms can find patterns among multiple variables that might be missed by traditional statistical methods. They may lend insight into this problem by helping to identify combinations of factors linked to bloom formation, or even allowing us to forecast blooms a few days in advance by using data on current water and weather conditions. We will apply a variety of machine learning methods to a unique, public, long-term data set produced by the VT Department of Environmental Conservation. We will analyze ten years of data on water quality and cyanobacteria levels from around Lake Champlain with the goal of predicting cyanobacteria levels from easily measured water quality indicators.