AI-BioDynamics: Artificial Intelligence for Biodiversity Mapping and Conservation Decision-Making

Globally, accelerated human activities and climate change have driven a continuous decline in natural ecosystems, resulting in habitat fragmentation and overall biodiversity loss. In response to this crisis, the international community adopted the Global Biodiversity Framework (GBF) at the 15th Conference of the Parties to the Convention on Biological Diversity (CBD COP15) in 2022, highlighting spatial planning–based biodiversity management as a central strategy. Consequently, there is a growing demand for spatial datasets with global consistency and high accuracy to enable quantitative assessment of ecosystem-related indicators. The Global Ecosystem Typology (GET) proposed by the International Union for Conservation of Nature (IUCN) provides a standardized framework and spatial information for consistent ecosystem classification at the global scale. However, the existing GET spatial datasets are produced at a global resolution, which limits their applicability at the national level due to spatial resolution mismatches and reduced classification accuracy for ecosystem types. In this study, we developed an IUCN GET ecosystem map for the Republic of Korea using time-series Landsat satellite imagery. Ecosystem classification was conducted for the period (2020–2024) using machine learning and deep learning approaches, resulting in ecosystem maps comprising 36 classes specific to Korea. The modeling results achieved an overall classification accuracy of approximately 85%, with several ecosystem classes exceeding 90% accuracy. The results of this study enable rapid and efficient detection of long-term and large-scale ecosystem changes. Furthermore, the enhanced precision and accuracy of ecosystem type classification support detailed ecosystem area analysis and provide a foundation for biodiversity conservation–oriented spatial planning.