Evaluation of the Climate Forecast System (CFS) Model in Predicting the Madden Julian Oscillation (MJO) during Boreal Winter

Abstract

The Madden Julian Oscillation (MJO) is crucial for bridging weather and seasonal forecasts due to its influence on sub-seasonal to seasonal (S2S) conditions. This study evaluates the Climate Forecast System version 2 (CFSv2) model’s performance in predicting the MJO index, utilizing the ‘mjoindices’ tools developed by Hoffman. We analyze MJO index predictions from reanalysis and CFSv2 model data for the boreal winter months (November through January) spanning 2011 to 2021. Performance evaluation employs deterministic metrics such as Root Mean Square Error (RMSE) and Correlation Coefficient (CC), as well as probabilistic metrics including Continuous Ranked Probability Score (CRPS) and Brier score. The criteria for an active MJO event are defined as an MJO index exceeding 1 for five consecutive days. Results show that while RMSE and CC metrics indicate limited correlation between model predictions and observations, probabilistic metrics offer deeper insights. CRPS reveals that the CFSv2 model performs best in short-term predictions with a value of 0.368 at lead 1, although accuracy decreases with extended lead times. The Brier score for weekly predictions at lead 1 is 0.352, reflecting a lower accuracy for predicting active MJO events, but stability in prediction quality across subsequent leads is observed.