Predictions of Tornado-like Vortex Embedded in the 8 May 2003 Oklahoma City Tornadic Supercell Initialized from the Sub-Kilometer Analysis Produced By the GSI-Based EnVar Data Assimilation System

Previous studies suggested that tornado-like vortices (TLVs) can be a good predictor of tornados. Given the intrinsically small size, an increasing number of studies have attempted to simulate the TLVs using a sub-kilometer horizontal grid spacing. These studies however often initialize the high-resolution prediction with an analysis at a coarser resolution, mostly larger than or equal to 1 km. Such a coarse resolution initial condition (IC) may not capture the fine scale characteristics and circulation associated with the TLVs.

For an ensemble-based data assimilation, running all ensemble members at the sub-kilometer can be computationally expensive. The GSI based dual resolution EnVar is therefore extended where the control analysis is at the sub-kilometer resolution whereas the ensemble is at the coarser kilometer resolution. The impact of the sub-kilometer analysis on the prediction of TLVs is therefore investigated. Specifically, in the dual-resolution (DR) approach (Exp-DR), the control analysis is at the 500-m sub-kilometer resolution and the ensemble covariance is at the 2km resolution. In a coarse single resolution experiment (Exp-SR), all control and ensemble members are run at the 2km resolution. In both experiments, the free control forecast is run with a 500-m horizontal resolution. The 8 May 2003, Oklahoma City, tornadic supercell storm is selected for the study.

The prediction covers the entire period of two observed tornadoe outbreaks. A TLV in Exp-SR reaches the high end of category 0 on the enhanced Fujita scale (F0), while two TLV episodes respectively reaching F1 and F0 are produced in Exp-DR. The two strong surface vorticity swaths derived from Exp-DR fits with the observed tornado damage track well in both locations and timing. In comparison, Exp-SR only produces one strong surface vorticity swath and fails to capture the second development of observed tornadic vortices. Such significantly different simulated TLVs suggest that the importance of initializing the simulation with a sub-kilometer analysis. Comparison between the two experiments for each DA cycle reveal that stronger cold pool, higher hydrometeor mixing ratios, and more intense kinetic fields with finer scale features are produced in Exp-DR than Exp-SR due to the accumulated impacts of integrating and analyzing high-resolution fields.

To isolate the impacts of each analyzed field on the subsequent TLV predictions, five additional field-replaced experiments are conducted by replacing Exp-a) hydrometeor mixing ratios (QRAIN, QSNOW, QGRAUP, and QICE), Exp-b) thermodynamic field (T), Exp-c) kinetic fields (U, V, and W), Exp-d) kinetic and thermodynamic fields (U, V, W, and T) in Exp-SR with those fields in Exp-DR, and Exp-e) halving vertical velocity (W) in Exp-DR. A similar TLV with Exp-SR is produced in Exp-a) and early stage of Exp-b) and a second TLV is re-intensified during 2220-2230 UTC in Exp-b); while all experiments of Exp-c) – Exp-e) produce 2 TLVs with the first one similar with Exp-DR. Relative to Exp-DR, Exp-d) almost resemble its development of the second TLVs; a weaker and shorter-lived re-intensified TLV is generated in Exp-b); Exp-e) produces a weaker TLV with a slower rate of re-intensification. Detailed diagnostics reveal that updated hydrometeor mixing ratios slightly influence the subsequent forecasts; the thermodynamic field from Exp-DR can enhance the strength of subsequent rear flank cold pool and prevent the overlarge rear flank downdraft (RFD) through increasing the low-level pressure perturbation; the kinetic fields contribute to the timing of TLVs development, e.g., the timing of the first TLV weakens and re-intensified into the second TLV. More details will be presented on the conference.