Multiscale Data Assimilation for Numerical Weather Prediction (Invited Presentation)

The emergence of new computing resources and technologies allows for a significant increase in the effective resolution of the next generation numerical models. Such computational resources also streamline the path for observations from a myriad of existing and new in-situ and remote sensing platforms, which observe a variety of scales, to be ingested in the numerical weather prediction (NWP) models. A next generation data assimilation system is therefore required to effectively analyze the state and its uncertainty across multiple scales, which hereafter is termed as “multiscale data assimilation (MDA)”. In this talk, challenges associated with the MDA for NWP are first discussed. Methods to implement the simultaneous MDA in ensemble-variational (EnVar) DA system is introduced. Recently the simultaneous MDA approach is extended in the multi-resolution ensemble 4DEnVar context and implemented for FV3GFS. It is found that the multi-resolution ensemble 4DEnVar can improve both the global and hurricane track forecasts compared to an equal cost single high-resolution 4DEnVar system. Additionally, the simultaneous multiscale EnVar is further developed and implemented for convective scale weather prediction in the RRFS like and HAFS systems. Experiments are conducted for the 8 May 2003 OKC (Oklahoma City) tornadic supercell and for CONUS MCS. It is found that the simultaneous multiscale EnVar improves the supercell and MCS analysis and subsequent prediction. Experiments are also conducted for hurricane Laura in HAFS. It was found that both the hurricane intensity and track forecast were improved. Detailed diagnostics to reveal the impact of simultaneous MDA for FV3GFS, RRFS and HAFS will be presented. Finally, a novel multiscale data assimilation (DA) method, MLGETKF (Multiscale Local Gain Form Ensemble Transform Kalman Filter) will be introduced.