Assessment and classification of Jirisan national park using machine learning, environmental spatial information, and user participation data

For sustainable management of national parks – critical national ecological resources – it is crucial to understand the area’s spatial characteristics and identify users’ perceptions. Moreover, with the increasing emphasis on the need for objective and scientific management of conservation areas, more scientific and quantitative evaluation is required to evaluate conservation areas. This study classified landscape units according to the spatial characteristics of Jirisan National Park in South Korea using a self-organizing map (SOM), an unsupervised learning algorithm. Nine evaluation indicators were selected considering the geographical, natural, ecological, landscape, and human-social environments of Jirisan. Moreover, to assume the perceptions of actual national park users for future management directions, this study used Latent Dirichlet Allocation and image classification using Google Vision API by collecting social media image data created based on user experience. Thirteen evaluation indicators were established by combining the nine established environmental indicators and four perceived major landscape types. The SOM showed eight landscape units in Jirisan National Park with different spatial characteristics. In addition, landscape units with active user activities and high naturalness were distinguished, and spatial distribution patterns of each landscape unit were identified. Identifying the landscape experienced and perceived by park users and understanding how their usage patterns are formed can be critical to managing national parks. The SOM method presented in this study can resolve the complexity of various indicators evaluated from multiple perspectives and assess scientifically national park areas.