457 Pathways to Better Prediction of the MJO Initiation over the Indian Ocean

Monday, 11 January 2016
Ajda Savarin, RSMAS, Miami, FL; and S. S. Chen

Handout (11.4 MB)

The convective initiation and eastward propagation of the Madden-Julian Oscillation (MJO) over the tropical Indian Ocean are not well understood and are poorly predicted by the numerical weather prediction (NWP) and climate models. Recent advancements in high-resolution, coupled atmosphere-ocean model development and observations from the Dynamics of the MJO (DYNAMO) field campaign have provided new pathways to improve model prediction of the MJO convective initiation. This study systematically examines the effects of model physics and resolution on model prediction by conducting numerical experiments using a coupled atmosphere-ocean model. The atmosphere and ocean components of the coupled model are the Weather Research and Forecasting (WRF) model version 3.6.1 and the Hybrid-Coordinate Ocean Model (HYCOM, v2.2.98), respectively. The model simulations are compared to satellite, ship, and aircraft observations from DYNAMO during a strong MJO initiation even in November 2011. These observations have led to significant model improvements.

In this study, we first conduct a control simulation using the coupled model with nested grids of 36-, 12-, and 4-km horizontal resolutions and 36 vertical levels in WRF, and a uniform resolution of 0.08° and 32 vertical levels in HYCOM. Three model experiments are performed to investigate the impacts of 1) cloud-permitting vs. cumulus parameterization, 2) air-sea coupling, and 3) parameterization of air-sea fluxes on prediction of the MJO initiation.

In the first experiment, we remove the 4-km cloud-permitting domain. The remaining 12- and 36-km nests use the Kain-Fritsch cumulus parameterization. Poor representation of mesoscale convective systems (MCSs), and a possible decoupling of the boundary layer from the free troposphere above it are seen in the lower resolution experiment, degrading model prediction in terms of both convection and near surface zonal westerly winds associated with the MJO (compared with the control simulation).

The second experiment uses an uncoupled WRF with the same configuration as in the control simulation, except with no coupling to the ocean. The SST is fixed in time from the model initial time. The uncoupled WRF exhibits a high bias in the amount of precipitation that can be linked to a high bias in surface latent heat fluxes. In comparison, the surface latent heat flux bias in the coupled WRF-HYCOM used in the control simulation is on average reduced by a factor of 1.5, with larger reduction at high wind speeds. The effect of sea-surface cooling following the leading edge of the MJO event in the control simulation also reduces the amount of precipitation produced, especially after a few days into the simulation.

Although coupling to the ocean helped correct part of the atmospheric model bias in excessive precipitation, the high biases in air-sea latent and sensible heat fluxes, as well as precipitation, are still present compared to observations. To further investigate model error, we conduct a third experiment in which we improve the surface layer parameterization in WRF by reducing a parameterized “convective velocity” to 50% of its value. The original parameterization was based on land surface data, and considering ship observations collected during DYNAMO, we find that its reduction is better suited when simulating ocean areas. This experiment shows a significant improvement in the surface latent heat flux at low wind speeds compared with the control simulation. This is reflected in a reduction of precipitation amount by about 18% on average, bringing it closer to observations and also improving the representation of the leading edge of convection.

These results show that the cloud-permitting resolution, air-sea coupling, and model physics, microphysics and surface air-sea fluxes are crucial for improving model prediction of the MJO convective initiation over the Indian Ocean. Further investigation of the coupled model representation, the PBL parameterization, and their impact on the MJO prediction are underway.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner