Mapping Paddy Fields Using Satellite Images and Machine Learning to Identify High Temperature-Induced Sterility in Nankoku, Japan

Naoyuki Hashimoto, Haruki Yamada, Shiho Matsuoka

High temperature-induced rice sterility has become a major issue in Japan; thus, the conditions influencing this sterility must be better understood to identify effective countermeasures. In this study, a random forest-based sterility estimation model was developed using the sterility rate measured via a field survey and satellite images. Applying this model to Nankoku, Japan, we attempted to map fields based on their sterility rates and visualize the spatial distribution of sterility. The results showed that the rate of change in reflectance from the heading stage until an effective accumulated temperature of 350 °C was reached was an effective model variable. Applying this model to map fields where rice sterility occurred from 2022 to 2024 revealed that more than 41% of the fields in Nankoku may have been damaged, suggesting that many fields might be at risk of adverse effects from high temperatures. The 3-year average sterility rate revealed areas with a high concentration of paddies with a low sterility rate, suggesting that investigating the environment and cultivation management techniques in these areas could provide insights to reduce the sterility rate. Moreover, the growth process up to the heading stage may contribute to the increase in the sterility rate. In the future, we plan to conduct a longitudinal survey based on the generated map to further investigate the relationships between cropping type, cultivar, and weather conditions to develop countermeasures.