High-Resolution real-Time Diagnostic Platform (realTPA) for Forest Fire and Landslide Early Warning

Climate change has intensified forest fires and landslides across mid-latitude regions. Conventional early warning systems rely mainly on meteorological data and lack spatial precision. This study presents the real-Time diagnostic Platform and Action system (realTPA), developed by the OJEong Resilience Institute at Korea University. realTPA delivers 100 m resolution risk mapping, processes multi-source data within one hour, and updates warnings every three hours. By integrating atmospheric variables with land surface indicators such as vegetation indices, topography, and fuel conditions, the system improves spatial specificity for disaster-prone areas. The forest fire model (CatBoost) achieved an AUC of 0.94, while the landslide model (Random Forest) reached an accuracy of 0.93 and AUC of 0.98. Operational validation over the Korean Peninsula (2025–2026) showed that 76% of 510 observed fire events occurred within 'Dangerous' or higher risk zones, and 89% of landslides fell in high or very high susceptibility classes. Case studies of the March 2025 Uiseong-gun fire and July 2024 Buyeo-gun landslide confirmed strong diagnostic reliability. Although technical performance is high, field utilization remains constrained by institutional barriers. realTPA offers a scalable framework for near real-time climate disaster early warning.