TOKYO – A research team at Sophia University (Tokyo) has published online an “bear encounter AI prediction map” that estimates the likelihood of human-bear contact in 19 areas, with the aim of reducing the frequency of attacks.
The map uses artificial intelligence to predict bear-related risks in areas such as Sapporo in Japan’s northernmost island of Hokkaido and Akita Prefecture on Honshu, based on local environmental data and reported sightings.
The team hopes this tool will help residents, tourists and local authorities prepare for bear activity in their areas, as concerns about bear attacks are also growing in cities and towns.
Data science expert Associate Professor Yusuke Fukazawa and his team have developed an AI-based tool to predict bear-human contact risk by analyzing one to three years of sighting data, forest conditions, road layouts, and population density.
The analysis showed that the risk was higher in the foothills of mountainous regions, along rivers where bears can roam unnoticed, and on roads in remote valleys, while areas with aging populations tended to be more likely to encounter bears.
Based on this system, the researchers published a predictive map on Oct. 20 that color-codes a square kilometer area into five levels, from low to very high, and shows the estimated probability of a bear’s presence in that area ranging from 0-20 percent to 80-100 percent.
In areas where bears have actually been sighted, the model found that accuracy levels varied depending on local conditions across the surveyed locations.
Fukasawa expressed enthusiasm for improving the model’s accuracy by seeking input from bear experts and incorporating additional information such as weather data and bear distribution and movement patterns to improve future iterations.
Japan has brown bears in Hokkaido, and Asiatic black bears in Honshu and Shikoku, one of Japan’s four main islands.
