The HRRR will be upgraded to version 4 soon (Rap to version 5). Here are some notes about that change.
1.) HRRRv3 had 36 hour forecasts every 6 hours. That has been updated to 48 hour forecasts every 6 hours.
2.) HRRRv3 struggled with convective spin up in the 2-6 hour range, supposedly this issue is less problematic with the new update but not totally resolved. In this case, spin-up problems refer to convection that is initialized correctly but then quickly disappears within 1-2 hours, etc. This is related to the data assimilation system used. Previously the HRRR began with an interpolation from the RAP, but now it will initialize from the HRRR ensemble and statistics from that ensemble. It will likely reduce the spin-up problem. In the future the HRRR ensemble will pick a subset of the ensemble members to advance out 48 hours. This can already be viewed experimentally.
3.) Some issues with a warm bias were also corrected, apparently related to excessive erosion of low level clouds. Verification of cloud stats are MUCH better, especially for low level clouds. This should have a positive impact during inversion events at retaining the low level cold pools. It should also improve low cloud forecasts in marine environments. Possibly related to this, surface wind speed forecasts seem to be better.
4.) A lake model is now being run which will offer improved forecast of wind and other variables near/over lakes. For this audience, that might mean some improvement for features over the Puget Sound region. Ocean wind speeds are also improved.
5.) The variable density snow algorithm was updated. It should almost always increase the forecast snow totals.
6.) 15 minute total precipitation increments are now added. HRRR-smoke is now operational as part of the HRRR. This can affect surface temperature forecasts, which should result in more accurate summer temperatures when fires are burning.
7.) Version 5 of the HRRR is probably about 2-3 years away, but its dynamical core will be the FV3 whenever it does come out.