Potential Benefits of Convection-Allowing Initial Conditions in a Variable-Resolution Global Data Assimilation System

Using the Joint Effort for Data Assimilation Integration (JEDI) software interfaced with NSF NCAR’s Model for Prediction Across Scales (MPAS), we performed two parallel experiments with global three-dimensional ensemble–variational (3EnVar) data assimilation (DA) systems. One experiment performed 3DEnVar DA on a global variable-resolution mesh that had ~3-km horizontal cell spacing over much of North America and ~15-km cell spacing over the rest of the globe. The second 3DEnVar experiment was exactly as the first, except the DA system was executed on a global quasi-uniform 15-km mesh (without a ~3-km mesh-refinement region). Both experiments produced continuously cycling 3DEnVar analyses every 6 hours for 35 days, assimilated conventional observations and clear-sky radiances, and ingested background error covariances provided by a global, 15-km, MPAS-based ensemble Kalman filter. Deterministic 192-h forecasts on the global variable-resolution mesh were initialized from both sets of 3DEnVar analyses; for the quasi-uniform 15-km DA system, 15-km 3DEnVar analyses were interpolated onto the variable-resolution mesh to initialize the 192-h forecasts.

Performing DA on the variable-resolution mesh rather than on the quasi-uniform 15-km mesh yielded small, but sometimes statistically significant, improvements to global forecasts of dynamic and thermodynamic variables. Additionally, performing DA on the variable-resolution mesh led to better precipitation forecasts through ~48 h over the central–eastern conterminous United States. While this result echoes limited-area modeling studies finding that short-term high-resolution precipitation forecasts are best when initialized from corresponding high-resolution analyses—rather than from coarser downscaled analyses—this study may be the first to document that benefits of high-resolution analyses for short-term high-resolution precipitation forecasts can be realized in global modeling systems. Furthermore, we have some evidence to suggest that these benefits may extend to longer forecast lead-times in global modeling systems than in limited-area modeling systems.

These experiments are part of longer-term efforts at NSF NCAR to develop global convection-allowing ensemble-based DA systems, which will also be briefly discussed.