Thursday, December 9, 2021

Long-Range ECMWF ENS Control Member Shows Frigid January; Will it Verify?

 The latest long-range ensembles from the European forecasting agency came in the other day, and while on the whole they produced a generally milder outlook, we did see a surprise forecast from the 'control' member of these ensembles - a remarkably frigid January.

Before we get into the good stuff, let's review what that means.

The ECMWF agency is known for their namesake weather model, but they also provide a critical forecasting tool with their ensemble system. In an ensemble forecast, an initial forecast model is taken, tweaked ever so slightly, and then re-ran. That's one ensemble 'member'. Then, they take that initial data again, tweak it ever so slightly in a different way, and re-run it again. That's another ensemble member, and it continues. The ECMWF's ensemble system has 51 different members - they are all different versions of the ECMWF weather model, but each one has slightly different starting conditions in order to produce a different forecast

You might ask why one would do that, since it sounds a lot like all we're doing is intentionally changing the initial conditions and letting it run free. That doesn't sound very productive, does it? In reality, the value comes from how similar all of these 51 members are at each point in time. For example, if all 51 ensemble members showed a winter storm in the Midwest ten days from today, that would be a big boost to forecast confidence. Why? Because if all 51 members have that storm happening, it means no matter what the weather conditions are right now (i.e. even if they're tweaked in all kinds of ways), we still end up with a winter storm on the horizon. As a result, ensemble forecasting is a potent part of any forecaster's toolbox.

But each ensemble set - whether it's the ECMWF ensembles, the American (GFS) ensembles, or any other set - has one member that is different from the rest. This member's initial conditions are the best possible data we have, and all the other members are those slightly-tweaked versions of this one. We call this special ensemble member the control. Indeed, the control member is the raw forecast taken strictly using initial data. Nothing is tweaked, nothing is altered, just the raw initial conditions. Every other ensemble member takes that control member data, tweaks it a bit, and re-runs it, like I described above. 

One may argue that this makes the control member more attention-worthy, since it only uses the best initial conditions data we have and isn't artificially tweaked. Whether that argument has merit is something I have yet to verify on my own.


In any event, now that we understand why an ensemble's 'control' member is worth monitoring, let's get down to the good stuff: the reason why this post was made.


Attached above from WeatherBell (all images on this post are from WeatherBell) is a look at the seven-day average temperature anomaly from the ECMWF ensemble control member over the December 25th - January 1st timeframe. We see that the control member anticipates a deep and formidable reservoir of below-normal temperatures in western Canada, which begins bleeding south and east into the United States as December turns to January. Indeed, seven-day average anomalies of more than 25 degrees Fahrenheit below normal are nothing to shake a stick at!

The outlook continues into mid-January below:


From January 11th - 18th, temperatures are shown by the ECMWF ensemble control member to be much colder than normal across almost all of the country, particularly east of the Front Range and especially in the Ohio Valley. With seven-day average temperatures seen at almost 20 degrees F below normal in parts of Ohio, Indiana and Kentucky, it's clear that the ECMWF Control member expects the Arctic gates to burst open in the first month of the new year.

Teleconnections accordingly show a breakdown in the tropospheric polar vortex as judged by a collapse in the Arctic Oscillation in the blue line (control member's forecast)...


... and a similar plunge in the North Atlantic Oscillation (NAO) from the control member allows that frigid Canadian cold to plow eastwards...


Now, that's all good and fun to look at, but this post wasn't written to get all the cold weather fans out there hyped up. Indeed, as the above two teleconnection charts show, the control member is an extreme outlier in the late-December and early-January period, and the average of all ensemble members just barely turns the NAO negative at the end of this month. The mean doesn't even turn negative at all for the Arctic Oscillation!

My takeaway is this: we're seeing some noteworthy cold signals for the end of December and start of January from the ECMWF control and the CFS ensemble system (a different analysis of that coming later today), but we can't rely on models that far out. To me, it's just not smart forecasting to use them for anything aside from large grains of salt, if that makes sense. Instead, I'm focusing on a changing global weather regime as the Madden-Julian Oscillation changes into colder phases for the end of December and through January. That, I feel, is where the real intrigue lies for Arctic outbreak risks as we enter the new year.

Stay tuned for that aforementioned post later today.

To Summarize:

  • The ECMWF ensemble's 'control' member - often seen as the most accurate member of the ensemble group - is expecting much-below-normal temperatures from the end of December through much of January in the U.S.
  • At this stage, while it's intriguing, I would not take this signal as anything more than a large grain of salt
  • There is more value in monitoring the changing global pattern as the new year kicks off, and a post discussing that will be published a little later today.

Andrew