Definitely want more info on this. Fascinating stuff. How does one make a simulation like that, and when it missed the event completely, how far out was it? Wild that someone can put out something like that, sounds really interesting. Was their goal to look at how changes/updates have affected its precision and accuracy?
They took the ERA-5 initial conditions and fed them into the GFS physics, and it still missed. I figure it went about 24 hours out, which is where the error occurred. The conclusion is that something went wrong with the data initialization. His former PhD student, Mike Brennan (now NHC director), found that there was some convective latent heat release that was entirely undermodelled, and all the readings that measured this were so far off from the model that they discarded this data. So nobody knows exactly why the initial readings were so flawed.
