Report on a Preliminary Test of Roadkill Observation System in Amami-Oshima Island, Japan

Abstract

 Recording and storing the accurate location and the correct species names of roadkill is necessary when researches want to use roadkill data for analysis and countermeasures. Application program (apps) and websites can collect and store consistent, long-term and extensive data on target areas. Our object was to clarify issues by testing a trial app (roadkill observation system) which can create a database with photos and can search for species names of photos using AI for collecting roadkill data in Japan. Roadkill data were collected from September 2019 to February 2020 on Amami-Oshima Island. We evaluated the accuracy in two perspectives; the distance from roadkill location to the nearest road because compared to their positioning error, the accuracy of species names uploaded. Users input the correct species name for 66 uploaded images; however, six images were too unclear to identify by AI. The median distance from roadkill locations based on smartphone GPS to the nearest road was 3.0 m. Most GPS locations were within 20 m. Our trial of the roadkill observation system app would be effective for collecting roadkill data.