Constructing Mammal Occurrence Status and Deep Learning Training Data from Camera Trap Surveys in Wind Farm

In South Korea, there has been an increase in the development of onshore wind farms as part of national strategies to achieve carbon neutrality. However, concerns have been raised regarding the potential ecological impacts on terrestrial mammals posed by the placement of these facilities in forested mountain areas. In order to establish a baseline for the purpose of impact assessment, camera trap surveys were conducted at two onshore wind farms - Gasan (GS) and Moochang (MC) - and differences in community composition and spatial activity patterns were examined. A total of 40 camera traps were deployed, yielding 1,875 usable images between May and December 2023 (excluding legally protected species). Eleven species from eight families were documented. GS exhibited a relatively even assemblage, dominated by Cervidae (31.6%), Mustelidae (28.0%), and Sciuridae (14.5%), whereas MC showed lower evenness and strong dominance by Mustelidae (29.0%), Cervidae (28.6%), and Suidae (21.2%). Spatial distribution mapping indicated that GS detections were concentrated along boundary ridgelines, while MC detections clustered near access roads and lower slopes. This reflects different habitat-use patterns under wind farm development. During the data processing stage, a significant amount of effort was required for the manual annotation and formatting of the data. In order to streamline this process, a user-assisted labeling tool was developed that automates metadata extraction and produces both monitoring-ready datasets and deep learning-compatible annotations. While not directly related to the ecological objective, this workflow exemplifies a pragmatic approach to reducing processing demands and paving the way for future AI-driven analyses. This study provides one of the first standardized datasets of mammal occurrence for Korean wind farm sites, and highlights the necessity for longterm, repeatable monitoring supported by scalable data-management frameworks.