Flash droughts develop quickly due to extreme weather conditions. Their sudden onset and complex drivers make them difficult to detect with a single drought index, especially in regions like South Korea where drought patterns vary by season and area. To overcome this challenge, we created an ensemble drought model that combines multiple drought indices using the Random Forest algorithm. We proposed two types of models: the Non-Seasonal Integrated model (NS) and the Seasonal Hierarchical model (SH). Results showed that the SH model performed better than the NS model, with higher accuracy and better agreement with observed drought-related damage. The SH model was more effective at capturing the spatial and temporal features of flash droughts across different seasons. This study introduces a new method for detecting and analyzing flash droughts, offering a robust and practical tool to improve drought monitoring and response in South Korea.