A Four-Dimensional Ensemble Variational (4DEnVar) data assimilation system for the Model for Prediction Across Scales (MPAS)

A global Four-Dimensional Ensemble Variational (4DEnVar) data assimilation system for MPAS is constructed with the Joint Effort for Data assimilation Integration (JEDI). Global cycling data assimilation experiments with analyses ranging from 120–30 km show that 4DEnVar updates have lower mean errors than comparable 3DEnVar experiments. Cycling experiments with the ensemble run at a coarser resolution are also conducted and shown to perform well. Furthermore, it is shown that the inclusion of temporal localization in the form of a time decaying Gaussian has a substantial positive influence on 4DEnVar analyses. Lastly, extended forecasts initialized from the 4DEnVar analyses are compared with forecasts initialized from 3DEnVar analyses. A particular focus is on the prediction of clouds in forecasts initialized from 4DEnVar analyses versus 3DEnVar analyses.