Potential predictability of the MJO in the ISVHE multimodel framework

Skillful prediction of the Madden Julian Oscillations (MJO), using dynamical models, has been one of the main objectives of the climate community over the past decades. Due to its far reaching influences on the weather and the climate in the tropics as well as in the mid latitudes, accurate prediction of the amplitude and location of the MJO in real time would be helpful in understanding and predicting a wide range of phenomena ranging from, tropical cyclones and monsoon active/break cycles, to rainfall variability over the extra tropics. Some of the present day numerical models are capable of predicting the winter MJO up to 3-4 weeks in advance. Is there further scope for extending this intraseasonal predictability? What would be the theoretical upper limit for MJO predictability? The Intraseasonal Variability Hindcast experiment (ISVHE) focuses on understanding the present day MJO prediction capabilities of dynamical models, both practical and potential. In the multimodel framework of ISVHE, the potential predictability of the MJO is explored for eight coupled models with varying degrees of MJO prediction skills. Analyzing the divergence of atmospheric MJO trajectories starting from close by initial states under perfect model assumption, estimates of MJO predictability are made for the eight models and the dependence of MJO potential predictability on the geographic locations and the strength of MJO are analyzed. MJO predictability is estimated from both the deterministic and ensemble mean hindcasts, giving values of 20-30 days and 35-45 days respectively, for most of the models examined. Exploring the dependence of predictability on the phase of MJO during hindcast initiation reveals a slightly higher predictability for hindcasts initiated from MJO phases 2, 3 6 or 7 in three models with higher prediction skill. The analysis also indicate that the “Maritime Continent prediction barrier” may not be an actual predictability problem. The fact that the model's hindcast skill is less than the estimated predictability limits by about 5-7 days suggests that more skillful MJO forecasts can be afforded through further improvements of dynamical models. In addition, analysis of the forecast “spread error” relationship in the different ensemble prediction systems (EPS) indicates that significant improvements in MJO prediction can also be brought out through focused improvements in the structural and model fidelity aspects of existing EPSs. The idea that the theoretical upper limit to predictability of MJO is as high as five weeks is encouraging for our efforts towards developing an ideal prediction system for intraseasonal variability.