Evaluating interannual variability in 1-month CFSv2 forecasts with different lead times at monthly and daily scales in South Korea

Abstract Extended-range forecasts beyond short-term period provide valuable meteorological insights, supporting decision-making across diverse sectors. However, their practical utility is often limited by the inherent uncertainties in initial conditions and model errors, emphasizing the need for thorough evaluation to ensure their effective application. Among various long-range forecast models, the NCEP Climate Forecast System version 2 (CFSv2) stands out with its 6 h initialization intervals and 9-month prediction range, providing a valuable resource for assessing forecast skill across different lead times. In this light, this study assesses the 1-month CFSv2 forecast skill over South Korea, focusing on the impact of varying lead times on predictability of interannual variability of temperature and precipitation. For the summer season from 2011 to 2023, temperature and precipitation on monthly and daily scales from CFSv2 forecasts for each month, with lead times ranging from 6 h to 240 h, were compared with observed values in terms of anomaly correlation coefficients (ACCs). With regard to monthly ACCs, no significant patterns were observed for both temperature and precipitation. On the other hand, daily temperature ACCs revealed that forecasts with shorter lead times exhibited relatively better skill in predicting temperature. Meanwhile, daily precipitation anomalies from the forecasts still displayed very weak correlations with the observed anomalies across all initialization times, highlighting the limitation of forecasts in precipitation. Ultimately, this study aims to enhance our understanding of CFSv2 forecast performance in relation to varying initialization times and serve as a useful basis for more effective utilization of the forecasts in conjunction with advanced approaches such as building a time-lagged ensemble.