Here are some higher resolution images that illustrate the differences between the GFS-only initial conditions and the GFDL/GFS Merged initial conditions (please see my first post for background). The first 4 animations are the models’ representations of SLP (mb), simulated radar reflectivity (dBz), and low-level potential vorticity (PVU), as described in the captions. These plots display the output from the 3-km grid length inner nest for the model run initialized at 09/10/08 at 12Z. Only 72 hours of forecast are shown here, as Ike approaches the edge of the parent domain at that time and starts to leave the inner nest as a result.
Below the model simulations, I’ve included the radar imagery for Ike during landfall, as well as the satellite imagery for Day 1 of these model runs (9/10/08 12Z to 9/11/08 12Z) and Day 2 of these model runs (9/11/08 12Z to 9/12/08 12Z). The radar, satellite images, and model simulation images do not line up with each other exactly in time, because the data were observed at different intermediate times than those for which model output was generated. However, I’ve included the real data so you can get a qualitative sense of what Ike was really doing vs. what the models display during roughly the same time period.
In particular, on 9/10 and 9/11, Ike was undergoing an eyewall replacement cycle; however, the models did not capture that, as it is very difficult for dynamical models to do so. In addition, Ike was a very large storm with distinct asymmetries, many of which were not simulated well, especially by the bogus vortex.
Going forward, I’ll be trying to quantify the differences between the GFS-only and the merged simulations of Ike, so we can figure out what missing factors in our initial condition are contributing to forecast error (possibly asymmetries, winds, PV, moisture, etc.) and work toward a better IC to fill in those gaps and improve future forecasts.