Researchers’ tool to be part of European Space Agency’s Climate Change Initiative

New machine learning model predicts lake ice conditions with 94 per cent accuracy To advance climate change monitoring and public safety, researchers at the University of Waterloo have developed new tools that bring an unprecedented level of accuracy to identifying lake ice conditions.

The researchers are the first to use machine learning models for processing satellite radar altimetry data that can identify between open water, thin ice, growing ice, or melting ice with 94 per cent accuracy.

The algorithm is a major step forward for climate change monitoring and will be implemented as part of the European Space Agency’s (ESA) Climate Change Initiative to assess lake ice trends over time. It is also significant for public safety and recreation because the presence or absence of lake ice in Northern communities affects transportation on ice roads, food security, recreational activities, and tourism.

"There is a long history of low-resolution satellites providing inaccurate ice condition estimates," said Jaya Sree Mugunthan, master’s student in Waterloo’s Department of Geography and Environmental Management. "As satellite technology has evolved, we now have higher resolution data to inform our classification work and combined with state-of-the-art machine learning algorithms, it allows for these research and societal advances."

In a recent study, the researchers analyzed the largest lakes in the northern hemisphere located in Canada, U.S., Sweden, and Russia over a three-year-period. The lakes differed in size, elevation, geographical location, and seasonal ice changes, allowing them to test the algorithm’s rigour.

"We are at a stage where the algorithms we created are achieving high accuracies," said Claude Duguay, professor in the Department of Geography and Environmental Management. "As soon as the data becomes available, we can run it through the algorithm and report on the ice conditions, which is available to the public."

Future work will test new machine-learning approaches using data from satellites that were launched by the ESA last year. The researchers will also use the model on smaller lakes, like in Ontario’s cottage country.

This research is in keeping with Waterloo’s commitment to being a leader in sustainability research and education as the institution has been a catalyst for environmental innovation, solutions and talent for 50 years.

The study, Machine learning based classification of lake ice and open water from Sentinel-3 SAR altimetry waveforms , appears in the journal Remote Sensing of Environment.