A First Step toward Global Ensemble-Based Data Assimilation at Convection-Allowing Scales Using MPAS and JEDI

Using the Model for Prediction Across Scales (MPAS) interfaced with the Joint Effort for Data Assimilation Integration (JEDI) software, we performed four global three-dimensional ensemble–variational (3DEnVar) data assimilation (DA) experiments that look ahead to future global convection-allowing modeling systems. Specifically, three 3DEnVar experiments were executed on a global variable-resolution mesh with;3-km horizontal cell spacing over much of North America and;15-km horizontal cell spacing elsewhere. The fourth 3DEnVar experiment was executed on a global quasi-uniform 15-km mesh (without the;3-km region). Flow-dependent background error covariances (BECs) were provided by either global quasi-uniform 15- or 30-km MPAS-based 80-member ensemble Kalman filters. All experiments produced continuously cycling analyses every 6 h for 35 days, and 0000 UTC analyses initialized deterministic 8-day forecasts on the variable-resolution mesh. The experiments differed in terms of assimilated radiance observations, BEC resolution, and 3DEnVar mesh (variable-resolution or quasi-uniform). Increasing BEC resolution did not yield better forecasts, while assimilating more radiances unambiguously improved forecasts. Performing DA on the variable-resolution mesh rather than on the quasi-uniform 15-km mesh yielded only small (yet sometimes statistically significant) impacts on global temperature, wind, and moisture forecasts but clearly led to more skillful precipitation forecasts over the central–eastern conterminous United States through;48 h. Our variable-resolution 3DEnVar experiments likely represent the first examples of continuously cycling DA on a global mesh with a large area of;3-km horizontal cell spacing. However, our experiments were simplified relative to operational DA systems and not well tuned, so they are best viewed as proof-of-concept demonstrations that set baselines for future studies.