I think it's a computing power issue. It seems to take forever to get the operational output. If you took the operationals at their current resolution and reran them many times over, it would probably take a very long time. So what is done is that they rerun the operational at a lower resolution for the entirety of the time scale. Then they perturb the initial conditions as many times as a particular suite calls for (for the Euro, that would be 50 times - hence 50 individual ensemble members, based off of a control run with 50 slight initial condition variations).
I think we, in the weather forum and social media world, quite often see a failure to use the ensembles for the purpose for which they are intended. For instance, they could be used to validate 7-10 day pattern that might be conducive to a winter storm. Conversely, they are not intended to nail down the specifics of a 7-10 day snowstorm. They are also not intended to be used on an individual member basis. So, when you see a Euro operational model not showing a D7 threat but the Euro control is showing a D7 snowstorm, your point is quite valid: Why would we believe a control run with only one eye looking through a milk jug over an operational run with two eyes with scratched up glasses with the wrong prescription? Answer? We shouldn't. Is it fun to look at? Yes. Is it worth anything, not really (when used like that).
We could have a lengthy discussion about use cases for ensembles. But professionals generally use them to determine the uncertainty around short term specific events (storm formation, rainfall totals, etc.) and the validity of medium and longer range patterns. If the ensembles generally agree with the operational, then you can have higher confidence.
When it comes to winter storms threats, IMO (and others may feel differently), but you should not look at an individual ensemble member to validate anything at all, i.e. the Euro control. However, there is value in observing trends and run to run changes. There is value in observing how many ensemble members are showing a system vs. not (and a note here, as we have all seen, a few big members can skew the mean, so looking at the mean by itself may not be all that valuable in some instances).
Everything gets less skilled out in time, particularly the lowest-skilled solutions like individual ensemble members. Use the ENS to validate the OP and the individual members to see if there are wild or minor differences among members, which will give you an indication of how volatile the atmosphere is. Ensembles are best used to detect the level of uncertainty. Use them against themselves and against other ensemble suites for that purpose.
GFS trying. More organized and slower this run but failed.
