Testing Hybrid-3DEnVar in the convective scale NWP model AROME-Austria

We test a Hybrid 3-Dimensional Ensemble Variational (Hybrid-3DEnVar) Data Assimilation (DA) method in the limited-area NWP model AROME over Austria at 2.5km horizontal resolution, with a flow-dependent error covariance matrix sampled from a 50-member ensemble.

Rapidly evolving highly non-linear convective-scale processes and the unique orography of the Austrian Alps intensify the complexities of estimating model error correlations. While the climatological error covariance matrix can not well represent the non-linear error growth of convective-scale weather, these errors can be incorporated into the assimilation using the ensemble-based error covariance matrix. We explore 11 weighted combinations of climatological and sampled covariance matrices, ranging from a purely climatological (weight of 0) to a purely ensemble-based (weight of 1) B-matrix, with incremental weight adjustments to the ensemble by 10 percent increments. The pure climatological configuration (3-dimensional variational data assimilation, 3DVar) is the operational DA scheme of GeoSphere Austria and serves as a comparative benchmark for our experiments. Multiple distinct summertime convective weather scenarios with a special focus on local convection were tested, while cold and warm fronts also influenced some of these cases. Aircraft wind and temperature observations are split into assimilated and non-assimilated parts so that the latter serves as validation for the analysis.
The resulting analysis from the Hybrid-3DEnVar configuration outperforms the operational 3DVAR of GeoSphere Austria, indicating a substantial leap forward in forecast accuracy of convective scale weather within Austria’s complex terrain. However, the optimal weight to the ensemble-based covariances for the optimal analysis strongly depends on the weather phenomenon investigated.
Keywords: AROME-Austria, Hybrid-3DEnVar, a 50-member ensemble, convective scale, and non-linear error growth.