Previous high-resolution (3 km) numerical simulations of the 2000 Fort Worth tornado case were rather successful in predicting about 2 hours ahead of time the timing and location of a rotating supercell storm (denoted tornadic storm hereafter) over Fort Worth. The model-predicted low-level rotation center was within kilometers from downtown Fort Worth where and when the tornado touchdown occurred (Xue et al 2002; Wang et al 2001). Intermittent diabatic data assimilation (including radar reflectivity data) over a period of 1 hour proceeding the model initial time was employed to achieve the prediction results.

The evolution of the supercell in the model deviated from the observation in the hour following the time of tornado touchdown, however. Natural growth of errors certainly played a role, but careful examination of the model storms (obtained at either 3 or 1 km resolutions) at high temporal frequency showed that the interference of another storm (denoted splitting storm hereafter) to its southeast had a larger adverse effect. That storm went through a splitting process after which the left-mover intensified and moved north-northeastward and eventually merged with the tornadic storm, hence altering its subsequent evolution. In reality, the left-mover stayed weak while the right mover dominated.

To better understand the behavior of the splitting storm, thermodynamic soundings and vertical wind profiles were extracted from areas surrounding this storm in a 30-min period around the time of splitting. Idealized cloud simulations were performed initialized with these soundings plus a thermal bubble. Identical spatial resolutions and similar physics as the full-physics real-data simulation were used. It was found that the model storm was not very sensitive to the slightly different temperature and moisture profiles but very sensitive to the wind profiles. A small anti-cyclonic curvature found in a hodograph at the mid-lower troposphere was all it took for the left-mover to become dominant and the curvature is believed to be produced by the tornadic thunderstorm to its northwest. Combinations of the thermodynamic soundings and vertical wind profiles as well as manual modifications to the wind profiles were used to delineate the cause and effect. Because the curvature only appears in a 2-3 km layer below 5 km, standard bulk Richardson number and storm-relative helicity parameters are incapable of distinguishing the storm behaviors although the general theories can be used for the interpretation of these behaviors. The implications to storm-scale data assimilation of such sensitivities of model storms will be discussed. Further work will involve the assimilation of radial velocity data from the Fort Worth WSR-88D radar which hopefully can correct the mid-level wind profiles for a better prediction of the splitting storm and therefore an improved prediction of the later evolution of the tornadic thunderstorm.