Warming temperatures mean shorter ice seasons for Inuit in Sanikiluaq, Nunavut. Of equal concern is the growing unpredictability of the ice packs used to travel and hunt.
Small polynyas, where ocean currents, wind or other processes prevent ice from forming, can be very dangerous and must be spotted before travelling. While polynyas are found in similar places yearly, climate change makes information about increasingly unpredictable ice conditions more important than ever.
"While conditions have always changed each year, the ice season has now shortened and community members are reporting other effects like thinner ice, new polynyas, more slush and especially more variability in the ice conditions," says Becky Segal, maps manager for SIKU: the Indigenous Knowledge Social Network, a community-driven platform created by the Sanikiluaq-based Arctic Eider Society (AES). "That’s why ice information in SIKU is being used by so many people across the Arctic." SIKU is an app developed by and for Inuit and provides tools and services for ice safety, language preservation and weather.
To work towards sharing accurate and timely data on where polynyas are located and make ice travel safer, AES began working with Waterloo Engineering’s VIP Lab. Using data validated by local Inuit, the lab is developing machine learning models to identify potentially hazardous open-water areas in landfast sea ice using synthetic aperture radar (SAR) imagery. The output of these models will be combined with local knowledge and recent observations of ice conditions on the SIKU mobile app and web platform.
Segal says the AES team was introduced to the VIP Lab by colleagues at the Canadian Ice Service who spoke about the Waterloo Engineering lab’s reputation for computer vision work in radar imagery target detection. The VIP Lab’s Neil Brubacher (BASc ‘21 and MASc ‘24 in systems design engineering), under the supervision of Dr. David Clausi, co-director of the VIP Lab and a systems design engineering professor and Dr. Andrea Scott, associate professor of mechanical and mechatronics engineering, took on the project with support from the Mitacs Accelerate Indigenous Pathways program.
The first order of business was creating a dataset of polynyas that could be used to train machine learning models. Such a dataset was developed collaboratively between the VIP Lab, AES and SmartICE, a community-based organization that supports Inuit-produced, data-driven information products related to local sea ice travel conditions.
Brubacher says trying to automatically identify sea ice and open water areas from SAR imagery is a well-established field of research, with organizations like the Canadian Ice Service considering such algorithms for ice chart production in support of shipping route navigation. What sets this work from the VIP Lab and AES apart is the small size of the polynyas being targeted and the work’s community audience.
"The novel research angle of this work is looking at the very small size and sparsity of these landfast ice polynyas in the SAR imagery," he says. "It's a bit of a needle in the haystack problem."
The models the team have developed can detect 90% of target polynyas under clear imaging conditions, and they are continuing to improve precision by looking at how detections change across a series of images.
Brubacher says the ability to fuse the knowledge of his Waterloo VIP Lab colleagues and the Inuit community he worked closely with in Sanikiluaq has been very beneficial.
“Being at Waterloo in the VIP Lab and being surrounded by people who are on the very cutting edge of these deep learning systems grounds the work in a level of confidence and credibility,” says Brubacher. “Discussing this research with colleagues and mentors who have a breadth of expertise in computer vision and remote sensing improves the quality of the research tremendously.”
After two years of working on the project, Brubacher had the opportunity to travel to Sanikiluaq this January and discuss the research with the community.
“It was an incredible depth of knowledge around sea ice, weather, and travel safety practices that I was very privileged to experience and listen in on,” he says. “It provided an important context to the local knowledge that these new tech systems are interfacing with.”
The in-person experience also included Brubacher seeing polynyas up close for the first time after examining them for two years from his computer screen. He also experienced the changing weather in person, with temperatures sitting around –5 Celsius compared with the typical –20 that has been typical for January in the region.
Neil Brubacher presenting during his trip to Sanikiluaq this January
"There’s more variability," Brubacher says. "I’ve heard Inuit say that ice that used to be predictable isn’t as predictable anymore. But I’ve also seen an optimism towards adapting to these changing conditions, including a desire to develop effective digital technologies that can complement local Indigenous knowledge. Deriving community-prioritized information from remote sensing imagery is one example of this."
Brubacher says the developed polynya detection models will be potentially expanded to other coastal communities across the Arctic, and that they are continuing to investigate ways to combine machine learning and local knowledge to create the most effective community-driven information products.
Segal agrees, saying her team would like to continue working with the VIP Lab, with additional discussions on future projects that include wildlife counting for several species important to Sanikiluaq, like the Eider duck. This ties in with other research in the VIP Lab for whale counting in aerial imagery.
