Multitemporal monitoring of forest indicator species using UAV and machine learning image recognition

In natural restoration, it is important to improve the efficiency of monitoring. Remote sensing using unmanned aerial vehicle (UAV) platforms plays a major role in improving monitoring efficiency. UAV platforms are particularly suited for monitoring long-term, multitemporal changes. The objectives of this study were to develop a standard protocol for monitoring multitemporal changes in forest indicator species using a UAV platform, to evaluate multitemporal changes, and to examine the factors contributing to these changes. The study site was a forest located within Takaragaike Park (Sakyo-ku, Kyoto, Japan), and Rhododendron reticulatum was selected as the study species. The distribution of R. reticulatum flowers was identified from 2019 to 2023 using image recognition with artificial intelligence. A mesh was used as a spatial unit, and standardized values (Z values) were tabulated to evaluate the abundance and changes in the number of R. reticulatum flowers over time. The method used in this study showed very high accuracy in image recognition at multitemporal periods, indicating that it is useful for long-term monitoring. It was also able to detect the effects of forest management and the impact of vegetation succession. The new method is expected to become one of the standard protocols for understanding multitemporal changes in forest indicator species.