whydomyjointshurt

Many thanks, @Paul. A max. difference of 7 Kelvin across the grid doesn't strike you as high? I thought it seemed high for differences between two "identical" initial conditions, but I have zero experience with this. But the std. dev. of differences is only about 0.2 Kelvin. If I could piggyback with one more question about the analysis vs. the 000 hour forecast... in a 12z forecast, for example, what time does the 000 hour forecast refer to? Is it referring to conditions at 12z (i.e., noon UTC)? I'm puzzled because that forecast gets dumped on the FTP site after 12z, so it's more of a hindcast by then. I guess the analysis describes conditions assimilated prior to 12z? Thanks again.

(I also posted this to americanwx.com but thought this may be a better place for model-specific questions.) I'm trying to learn about NWP, and differences between model runs. From this thread: But if I compare an operational run vs. an ensemble control run of the same resolution, I still see differences of a few degrees Kelvin. Specifically, if I compare 12z runs from earlier today (21 April 2017): GEFS control analysis: File: gec00.t12z.pgrb2anl Message: "51:Temperature:K (instant):regular_ll:isobaricInhPa:level 100000 Pa:fcst time 0 hrs:from 201704211200:hi res cntl fcst" GFS analysis: File: gfs.t12z.pgrb2.1p00.anl Message: "227:Temperature:K (instant):regular_ll:isobaricInhPa:level 100000 Pa:fcst time 0 hrs:from 201704211200" I see differences of up to 7K. Why aren't they identical? Is the 1 degree operational run downsampled from something else in a different way than the ensemble control is? Thanks very much!