Experiments with a Variable-Resolution MPAS-JEDI-based Global Data Assimilation System with ~3-km Horizontal Cell Spacing over North America

Using the Joint Effort for Data Assimilation Integration (JEDI) software interfaced with NCAR’s Model for Prediction Across Scales (MPAS), we performed four experiments with ensemble–variational (EnVar) data assimilation (DA) systems using a global variable-resolution model configuration. Specifically, the model’s grid mesh possessed ~6.5 million grid columns, which had ~3-km horizontal cell spacing over most of North America and smoothly transitioned to ~15-km cell spacing over the rest of the globe (Fig. 1). Continuously cycling EnVar analyses were produced every 6 hours for 35 days in Spring 2019. All experiments assimilated conventional observations and clear-sky radiances from AMSU-A sensors, and 0000 UTC analyses initialized 8-day forecasts over the global variable-resolution mesh (Fig. 1). The four experiments differed either in terms of the physics suites used within the MPAS model, whether all-sky AMSU-A and GOES-16 Advanced Baseline Imager (ABI) radiances were additionally assimilated, or the horizontal resolution of the flow-dependent, ensemble-based background error covariances (BECs) used within EnVar DA (provided by either 15- or 30-km MPAS-based ensemble Kalman filters).

The EnVar systems appeared to perform appropriately, as DA statistics indicated the model climate was stable over the cycling period and biases were generally small. All experiments initialized forecasts with diurnal cycles of precipitation over the conterminous United States (CONUS) that generally agreed with those observed throughout the 8-day forecasts. Precipitation forecast skill over the CONUS was very sensitive to the physical parameterization suite but less sensitive to whether 15- or 30-km ensembles provided BECs for EnVar DA. These findings reinforce the importance of a well-performing physics suite for effective DA and suggest that coarser, cheaper ensembles can potentially be leveraged within high-resolution global EnVar DA systems. Additionally, assimilating all-sky AMSU-A and GOES-16 ABI radiances clearly improved 3–8-day precipitation forecasts over the CONUS.

Our EnVar systems are likely the first demonstrations of continuously cycling global DA on a mesh with an area of convection-allowing cell spacing and look toward future operational models. In addition to describing our experiments, longer-term plans at NCAR to develop and demonstrate a global convection-allowing ensemble-based DA system will be discussed.