Potential Predictability of the Madden–Julian Oscillation

The objective of this study is to estimate the limit of dynamical predictability of the Madden–Julian oscillation (MJO). Ensembles of “twin” predictability experiments were carried out with the NASA Goddard Laboratory for the Atmospheres (GLA) atmospheric general circulation model (AGCM) using specified annual cycle SSTs. Initial conditions were taken from a 10-yr control simulation during periods of strong MJO activity identified via extended empirical orthogonal function (EOF) analysis of 30–90-day bandpassed tropical rainfall. From this analysis, 15 cases were chosen when the MJO convective center was located over the Indian Ocean, Maritime Continent, western Pacific Ocean, and central Pacific Ocean, respectively, making 60 MJO cases in total. In addition, 15 cases were selected that exhibited very little to no MJO activity. Two different sets of small random perturbations were added to these 75 initial states. Simulations were then performed for 90 days from each of these 150 perturbed initial conditions. A measure of potential predictability was constructed based on a ratio of the signal associated with the MJO, in terms of rainfall or 200-hPa velocity potential (VP200), and the mean-square error between sets of twin forecasts. This ratio indicates that useful predictability for this model's MJO extends out to about 25–30 days for VP200 and to about 10–15 days for rainfall. This is in contrast to the timescales of useful predictability associated with persistence forecasts or forecasts associated with daily “weather” variations, which in either case extend out only to about 10–15 days for VP200 and 8–10 days for rainfall. The predictability measure shows modest dependence on the phase of the MJO, with greater predictability for the convective phase at short (< ~5 days) lead times and for the suppressed phase at longer (> ~15 days) lead times. In addition, the predictability of intraseasonal variability during periods of weak MJO activity is significantly diminished compared to periods of strong MJO activity. The implications of these results as well as their associated model and analysis caveats are discussed.