Wow, those are large differences. The site has an archive back to the end of August. It's hit-or-miss which cycles are available, but the 12Z 24 August cycle is there for example, and it does not display the same 2-m temperature behavior. The EMC FV3 parallel is continually changing, so it seems a change was recently introduced that is causing these differences. I've asked the developers if there is a change log that can be shared with the group.

I've also attached a plot of differences between the GFS and FV3-GFS terrain I made to investigate the temperature differences. Like other fields, the FV3 terrain appears smoother than the GFS terrain. It is unlikely the smoothness is causing large temperature differences, but it is still an oddity that I'd like to understand. I've directed a question about the terrain smoothness toward the FV3 developers.

A little bit of follow-up: there were some changes to EMC's FV3 runs on 7 September, but they affected primarily cloud water and stratospheric temperature and winds. I think this 00Z behavior may have been happening near sundown this entire time, but sundown is now closer to 00Z so we're noticing it more. I've attached an image from the FV3/GFS plume comparison for Boise 2-m temperature in the first 72-h of the 12Z 23 October cycle with some obs overlaid. Between 21Z and 00Z the temperature in the GFS (blue line) decreases but the temperature in the FV3 (orange line) either decreases less or even increases. The result is an approximately 3-degree difference each evening at 00Z. The two catch up again by 03Z. Historically, the GFS has decreased temperatures too quickly around 00Z; this appears to be better performance in the FV3. It's something to keep in mind as our evaluation continues.

As for the terrain, it's different because a different filter is applied.

Here's a case where the GFS is COLDER than the FV3, if you look at the forecast 2-meter Temps over the Appalachians for the strong cold front forecast to push through the mid Atlantic late this week.