plSSN : 0374-8111 | elSSN : 2287-8815
plSSN : 0374-8111 | elSSN : 2287-8815
plSSN : 0374-8111elSSN : 2287-8815
Comparative Performance Analysis of Statistical Downscaling Methods for Reproducibility Assessment of Simulated Winter Low Water Temperatures in the South and West Coastal Seas of Korea: A Case Study for the Winter of 2023-2024
박명희·임병준·김창신·이준수*
국립수산과학원 기후변화연구과
This study evaluated five statistical downscaling methods-Empirical quantile mapping (EQM), detrended quantile mapping (DQM), delta change method (DCM), delta method (DM), and quantile delta mapping (QDM)-to more accurately reproduce observed low-temperature events using a regional ocean modeling system (ROMS)-based model. The analysis focused on 12 coastal stations in the seas south and west of Korea during the winter of 2023-2024. As the training and evaluation periods were identical, this study assessed model reproducibility rather than future predictive performance. The results from the root mean square error (RMSE) and Bland-Altman analyses indicated that the distribution-based methods, namely EQM and DQM, were generally more accurate. However, the optimal method depended critically on local oceanographic characteristics. The simple DM was most effective in stable environments with low variability, whereas EQM excelled in tide-dominated or topographically complex areas, and DQM performed best in regions influenced by freshwater discharge and meteorological changes. In contrast, QDM exhibited poor performance. These findings emphasize that a station-specific correction strategy is essential for advancing coastal low-temperature warning systems and provide a foundation for developing independently validated prediction models.
Statistical downscaling, Low water temperature, Bias correction, Model reproducibility