The objective of this research is to explore the limits of dynamical predictability of the Madden-Julian Oscillation (MJO). A more complete understanding of the MJO is a requisite step for meeting a number of CLIVAR-GOALS objectives due to the significant role it plays within our climate system. For example, it has been shown that the MJO has considerable influence over Asian-Australian monsoon activity, western North American winter and central South American summer rainfall variability, tropical storm and hurricane development, and possibly the timing and strength of El Nino / La Nina events. As influential as the MJO is on our weather and climate, estimates of its theoretical limit of predictability have yet be determined. While it is known that deterministic prediction of most weather phenomena is expected to provide useful skill at lead times of about 6-10 days, maybe longer in some instances, and that for ENSO at leads times of about 12-18 months, the corresponding metric for the MJO has not been quantified. All previous studies addressing this sort of question have been based on either statistical prediction models or dynamical studies of predictive skill but with models that poorly simulated the MJO. Thus the questions of how predictable, in principle, the MJO might be, and over what lead times, are still unknown.

To help address this question, ensembles of "twin" predictability experiments were carried out with the NASA/GLA AGCM using specified annual cycle SSTs. This model was chosen due to its relatively realistic MJO representation. Initial conditions were taken from a 10-year control simulation during periods of strong MJO activity identified via extended EOF analysis of bandpassed tropical rainfall. From this analysis, 15 cases were chosen when the MJO convective area was located in the Indian Ocean, Maritime continent, western Pacific Ocean, and central Pacific Ocean, respectively, making 60 cases in total. In addition, 15 cases were selected which 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 VP200 and rainfall data. Analysis of this measure indicates that useful predictability for this model’s MJO extends out to at least 20-25 days, somewhat longer (shorter) for VP200 (rainfall). The predictability measure shows some modest dependence on the longitudinal location of the convection. In addition, the predictability during periods of weak MJO activity is significantly diminished. Ongoing work, to be presented at the conference, involves examining the dependence of this result on initial condition perturbation size/structure, season, mid-latitude activity, and ENSO conditions, as well as its implications for mid-latitude weather and short-term climate predictability.