Evaluation of predictive performance of the Subseasonal Experiment (SubX) models to recent multi-year droughts in South Korea

Abstract

To cope with potential adverse impact of multi-year drought, accurate sub-seasonal forecast of precipitation is essential. This study evaluates the predictive performances of the five Subseasonal Experiment (SubX) models (ECCC-GEPS6, EMC-GEFSv12, ESRL-FIMr1p1, GMAO-GEOS_V2p1, and RSMAS-CCSM4) for the precipitation variations during two recent multi-year drought events (2007−2010 and 2013−2016) in South Korea. Based on various verification metrics for ensemble, deterministic, and categorical forecasts, the prediction skills of these five SubX models at subseasonal forecast times (1−4 weeks) are quantitatively estimated. Results show that the SubX models generally better predict the persistence stage of an emerging drought than the drought recovery stage in all forecast times. For all verification metrics, the skillful predictions are observed in the first two lead weeks (1−2 weeks). Although, those are markedly reduced in the last two-lead weeks (3−4 weeks), it can still suggest reliable prediction skill depending on the drought stage. The multi-model ensemble approach shows the best prediction skill, and EMC-GEFSv12 which has the most ensemble member presents the better predictive performance than other models. Results from the sensitivity of the number of ensemble members show that seven or more ensemble members enhance the prediction skills significantly, particularly in the 3 and 4th lead weeks. The findings of this study suggest that the need for selective use of the SubX subseasonal forecasts for South Korea droughts.