Extended-range Multi-scale Prediction in the Tropics for the DYNAMO time period

The performance of 30-day simulations of the Navy Global Environmental Model (NAVGEM) and the Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS®) forecast system, run over a large domain in the tropics and subtropics, are evaluated under several metrics. The time period of interest is the DYNAMO (Dynamics of Madden Julian Oscillation) field experiment period, starting late October 2011. The NAVGEM experiments are run at an effective 37-km resolution, and the COAMPS experiments are run at 45-km and 27-km resolution. The COAMPS domain covers the tropics and subtropics from the eastern Atlantic to the eastern Pacific, with lateral boundary conditions provided by global analyses. For both systems, the SST, provided by the NCODA (Navy Coupled Ocean Data Assimilation) system, is either held fixed to the initial value (fixed SST) or updated with analyses daily (observed SST). The extended integrations have been verified using observed OLR, TRMM precipitation estimates, and global analyses.

For both NAVGEM and COAMPS, comparison of the fixed and observed SST integrations indicate that the fixed SST has a small impact on the precipitation and wind biases for the first two weeks, but has a substantial impact after that. For NAVGEM, SST also has a large impact on land surface temperature, demonstrating the “propagation” of the SST influence from ocean areas to land areas. NAVGEM forecasts started at different times during the month of November indicate some skill in predicting the MJO evolution if the forecast is started during the MJO event. For COAMPS, comparison of the 27-km and 45-km resolution experiments indicates that while the 27-km simulation outperforms the 45-km simulation for OLR, the advantages of higher resolution for other metrics are unclear. The 45-km simulation has smaller errors for the wind and precipitation fields. Examination of time-longitude diagrams of equatorial precipitation shows that the 45-km COAMPS simulation with observed SST and analyzed lateral boundary conditions has a remarkably good representation of the late-November Madden Julian Oscillation (MJO). The 45-km simulation with fixed SST, and the 27-km simulations, had weaker representations of the MJO. Overall the results demonstrate the importance of accurate SSTs for predictions beyond “weather” time scales of 1-2 weeks for both the global and mesoscale forecast systems.