The Madden-Julian Oscillation moist static energy budget in CMIP5 models

Contributions by terms of the moist static energy (MSE) budget to the maintenance and propagation of convection in the Madden-Julian Oscillation are studied using 16 coupled climate models (14 from CMIP5 plus two additional coupled simulations) and reanalysis data from the European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ERAI). Models are ranked according to how well they simulate the eastward propagation of MJO convection. Top ranked models propagate MJO convection beyond 150E while lower-ranked models struggle to propagate MJO convection beyond the Maritime Continent.

In order to understand why certain models perform better than others, we assess the contributions of vertically integrated MSE budget terms to the vertically integrated MSE anomaly and its tendency, which reveals processes that maintain MJO convection and drive its eastward propagation, respectively. For the maintenance of MJO convection, two findings are consistent with previous studies: 1) longwave heating is the dominant process for maintaining MJO convection in all models, but 2) it is not significantly correlated with MJO skill.

For MSE tendency and MJO propagation, horizontal moisture advection is the dominant term. The column water vapor (CWV) eastward gradient (15S-15N averaged CWV at 160E minus 15S-15N averaged CWV at 100E) is positively correlated with MJO skill at the 95% level. Compared to observations, almost all models overestimate the CWV zonal gradient, with the largest biases found in the highest skill models. When MJO convection is located in the eastern Indian Ocean, average 850 hPa easterly wind anomalies to the east of MJO convection are positively correlated with MJO skill at the 99% level, but all models underestimate the magnitude of the easterly anomalies. The pattern correlations between observed and simulated 850 hPa meridional wind anomalies are also positively correlated with MJO skill at the 99% level, but meridional moisture gradients have no significant correlation with MJO skill.

Surface latent heat fluxes (LH) are negatively (but not significantly) correlated with column MSE. For top ranked models, LH slows MJO propagation, but erroneously promotes propagation in some of the lower-ranked models. A Reynolds decomposition of LH shows that the variability of the wind-driven and thermodynamic perturbations of LH are positively correlated with MJO skill at the 99% and 95% significance levels, respectively, although it is not clear if LH variability is a cause, or simply a consequence, of MJO skill.

Further study is needed to understand 1) the processes that determine the mean CWV field, 2) a model’s ability to replicate MJO wind anomalies, and 3) whether the erroneously large LH contributions to MJO propagation in some models compensates for negative biases in other processes.