Met Office MOGREPS‐G initialisation using an ensemble of hybrid four‐dimensional ensemble variational (En‐4DEnVar) data assimilations

Abstract

AbstractThe Met Office Global and Regional Ensemble Prediction System–Global (MOGREPS‐G) used an ensemble transform Kalman filter (ETKF) to perturb its initial conditions from its operational implementation in September 2008 until December 2019. In 2019, MOGREPS‐G became the first operational atmospheric ensemble to apply hybrid four‐dimensional ensemble variational data assimilation (En‐4DEnVar) to each of the 44 perturbed ensemble members. Other enhancements have also been added, including to the inflation used to improve ensemble spread. The combined impact of these changes on ensemble forecasts is overwhelmingly positive but initially more neutral for deterministic forecasts, which also use the ensemble to represent flow‐dependent forecast errors in their hybrid data assimilation updates. The latter result is not a surprise, because the deterministic forecast's hybrid data assimilation was initially weighted more strongly to the modelled stationary covariance component and not optimised to take full advantage of the upgraded ensemble. A subsequent operational upgrade in December 2020 has introduced shifting in addition to lagging to exploit the ensemble better in the deterministic forecast's hybrid data assimilation by including ensemble members from a previous cycle and also from adjacent forecast lead times to augment the ensemble without having to run additional forecasts. More weight has since been given to the ensemble in the deterministic forecast's hybrid data assimilation in May 2022. A key motive for adopting hybrid 4DEnVar in MOGREPS‐G is to reduce maintenance overheads by virtue of sharing much of the deterministic forecast system's data assimilation code. This also enables the ensemble to assimilate almost all observation types used by the deterministic forecast. The updated system also exploits parallelism better so as to be fast enough for operational use, despite assimilating more observations and being more computationally expensive than the Met Office's ETKF.