Dec. 26, 2010 Blizzarad Model Accuracy Reflections

[Edge Weather]

Looking forward to getting everyone's reflections on the various models and how they handled this storm. When did the Euro first catch this storm? Is it true that the Euro has accurately caught every major East Coast blizzard 6 days out? Also, I want to gather thoughts about HPC throwing out the runs of the GFS that first brought back the blizzard on 12/24. It was the only model showing the threat at that point, yet the runs were thrown out. What are your thoughts on these issues?

[MDstorm]

Looking forward to getting everyone's reflections on the various models and how they handled this storm. When did the Euro first catch this storm? Is it true that the Euro has accurately caught every major East Coast blizzard 6 days out?

The Euro first really showed the coastal bomb on the 0z run Dec 20. I remember watching the play by play courtesy of Earthlight and Tombo. At first the run looked like it would be out to sea. However, at 150 hours, magic happened and the coastal bomb came up the coast. It was a true widespread east coast snowstorm--similar to what actually happened except for the screwing DC part . The Euro continued to show the storm for several runs until losing it for several runs. Unfortunately for me, when the euro brought the storm back, it was further east than originally portrayed. The end result--not much change for places to the east and northeast, but huge implications for DC/Northern VA/ Central MD. I'll leave it to Wes to illustrate how small changes in the trough orientation/position led to the changes seen in the final solution.

MDstorm

[hawkeye]

The Euro was the best of the bunch. period.

[Edge Weather]

So, was this 7 days out?

The Euro first really showed the coastal bomb on the 0z run Dec 20. I remember watching the play by play courtesy of Earthlight and Tombo. At first the run looked like it would be out to sea. However, at 150 hours, magic happened and the coastal bomb came up the coast. It was a true widespread east coast snowstorm--similar to what actually happened except for the screwing DC part . The Euro continued to show the storm for several runs until losing it for several runs. Unfortunately for me, when the euro brought the storm back, it was further east than originally portrayed. The end result--not much change for places to the east and northeast, but huge implications for DC/Northern VA/ Central MD. I'll leave it to Wes to illustrate how small changes in the trough orientation/position led to the changes seen in the final solution.

MDstorm

[Edge Weather]

Also, wanting to find out from those knowledgeable about the models why the majority of the models showed an out to sea solution and even the Euro flipped to an out to sea solution at one point. The UK model was hundreds of miles out to sea, yet the UK is supposed to have the 2nd best accuracy scores. Also, very curious why the Euro lost the storm when we got to around 3 days out. If you remember the Euro also brought us the previous blizzard the week before that did not materialize in about that same 2-3 day time frame. Is the Euro actually better in some regards at days 6, 5, 4, than it is at days 2 & 3 for some reason? What do you all think. Thanks for sharing your thoughts.

[Harbourton]

Although the Euro had it first about 5 or 6 days out, it was the GFS on about the Friday 0Z run that brought it back to a coastal. The other models took at least 3 or 4 runs to catch up. Clear GFS victory in my opinion.

[Edge Weather]

It was the 12z GFS that brought it back on 12/24 but the very next run of the EC brought it back also. I agree, though, bringing it back by the GFS was huge and a victory for first bringing it back, but the Euro was still the first to show the threat like 6 or 7 days out.

Although the Euro had it first about 5 or 6 days out, it was the GFS on about the Friday 0Z run that brought it back to a coastal. The other models took at least 3 or 4 runs to catch up. Clear GFS victory in my opinion.

[OSUmetstud]

It was the 12z GFS that brought it back on 12/24 but the very next run of the EC brought it back also. I agree, though, bringing it back by the GFS was huge and a victory for first bringing it back, but the Euro was still the first to show the threat like 6 or 7 days out.

The GFS had the threat first. It just had the day wrong...it was showing a Christmas Day storm. This is like 8-9 days out. The ECMWF was first to pick up on the threat being later...more of a Boxing Day-27th deal.

[nj2va]

Euro was the first to show it as a Miller A IIRC. GFS originally showed it as sort of a Miller B/bowling ball type of event. Then GFS started showing it diving south into the GOM and up the coast. So it seemed the Euro worked best in the medium range since it was pretty consistent showing a HECS - albeit, a bit too far south with the bombing/development.

NAM did best with the precip, at least in my area.

[thegreatdr]

We took a good look at this last night with forecasts valid 12/27 12z and 12/26 12z. Within 96 hours, both the GFS and ECMWF showed good verification with the surface and H5 low. After that, the GFS showed greater variability than the ECMWF and tended to be quick and to the right with the low's track. Similarly, the ECMWF went through 1-2 days of runs where it was slow and to the left roughly in the 4.5-6 day time frame, as well as day 7 verifying on the morning of the 27th, when the ECMWF became too fast like the GFS. The timing issue at longer ranges was due to too much interaction with an upper low in northern Canada, which never merged/phased with this system, even though older runs of the GFS did so. To some degree, both pieces of guidance showed their medium range biases. It turns out a simple compromise of the GFS and ECMWF for those days (in runs out to six days) would have verified best with the surface low timing and track most days/runs. HPC tended to be slightly too far right with the low track in the final issuances and to the left in the early issuance, and beyond day 4, too quick. For instance, the final day 7 forecast from HPC was about 15 hours too quick as the low tracked near eastern Nova Scotia, while the early issuance was similarly quick but near the Maine coast. None of the ensemble means got the timing right beyond day 4.

An even better question is: with the GFS/ECMWF having the H5 and surface patterns nearly perfect within 4 days, why was their QPF so variable and flawed? Three to four days out, they were both extremely dry for much of the Northeast but nearly ideal for the DC area. Beyond that, the ECMWF was off by about 1.5 inch liquid equivalent for central MD/DC/northern VA (huge oops), and in shorter time ranges, the GFS showed the same issue (big oops) even if their QPF was better for New Jersey, New York, and Connecticut. In other words, the models that succeeded QPF wise for NY, CT, NJ failed for the DC metro area, and vice versa.

No run we saw verification for was perfect from the GFS and ECMWF QPF-wise, even within the real short range. Sounds like job security to me. =)

DR

[Harbourton]

Mood swings associated with Model Runs

GFS first causes guarded opimism after last weeks OTS semi-disappointment. At one point, Euro brings 964 low east of AC-NJ bringing some gasps from the east coast forums. Then the models all trended OTS prompting calls of I'll never trust the Euro again or the King is dead. Anywho, the GFS starts bringing it back and there is some hope for the weenies staying up all night to listen to Tombo's play by play of the Euro at 1:15 AM/EST (Thank you Tombo) . Models all start jumping on the bandwagon and there is joy in the bleachers. The last twist was the QPFdagger for the DC/Balt. area..

[Edge Weather]

Thank you very much for this detailed response. It is greatly appreciated. I enjoy learning what we can from the event so we can see where we went right and where we went wrong. It is through this type of analysis that we can improve our forecasts in the future.

We took a good look at this last night with forecasts valid 12/27 12z and 12/26 12z. Within 96 hours, both the GFS and ECMWF showed good verification with the surface and H5 low. After that, the GFS showed greater variability than the ECMWF and tended to be quick and to the right with the low's track. Similarly, the ECMWF went through 1-2 days of runs where it was slow and to the left roughly in the 4.5-6 day time frame, as well as day 7 verifying on the morning of the 27th, when the ECMWF became too fast like the GFS. The timing issue at longer ranges was due to too much interaction with an upper low in northern Canada, which never merged/phased with this system, even though older runs of the GFS did so. To some degree, both pieces of guidance showed their medium range biases. It turns out a simple compromise of the GFS and ECMWF for those days (in runs out to six days) would have verified best with the surface low timing and track most days/runs. HPC tended to be slightly too far right with the low track in the final issuances and to the left in the early issuance, and beyond day 4, too quick. For instance, the final day 7 forecast from HPC was about 15 hours too quick as the low tracked near eastern Nova Scotia, while the early issuance was similarly quick but near the Maine coast. None of the ensemble means got the timing right beyond day 4.

An even better question is: with the GFS/ECMWF having the H5 and surface patterns nearly perfect within 4 days, why was their QPF so variable and flawed? Three to four days out, they were both extremely dry for much of the Northeast but nearly ideal for the DC area. Beyond that, the ECMWF was off by about 1.5 inch liquid equivalent for central MD/DC/northern VA (huge oops), and in shorter time ranges, the GFS showed the same issue (big oops) even if their QPF was better for New Jersey, New York, and Connecticut. In other words, the models that succeeded QPF wise for NY, CT, NJ failed for the DC metro area, and vice versa.

No run we saw verification for was perfect from the GFS and ECMWF QPF-wise, even within the real short range. Sounds like job security to me. =)

DR

[SnowGoose69]

DT posted a day or so before the models began to show a big hit that the reason the 1/25/00 event was missed was because the models mistakenly viewed the far northern disturbance as being a kicker as opposed to it pulling the trough negative....this event may have been a result of the same error although I think its more likely the models had bad data on the northern disturbance until it neared the U.S. border.....I really don't believe much error was a result of the southern wave.

[CooL]

It is clear the the EURO/GFS are the best models in the business

they worked almost like a tag team - gfs showed it then handed it off to the euro for 6 runs - then the euro handed it off to the gfs 2 days before the event, and then everyone in the crowd (models) decided to join in on the party. Some of the problems the models had was a "pill" shaped trough that allowed the storm to round it and miss the mid atlantic and nail new england..hmm could have been some data issues with the northern stream

PS: NOGAPS should be retired hahaha

[Coach McGuirk]

6-10 expected tops. Ended with 15 here and 14.2 at Norfolk.

[green tube]

Looking forward to getting everyone's reflections on the various models and how they handled this storm.

i started a similar thread in the nyc/philly forum... and got some good replies.........

http://www.americanwx.com/bb/index.php?/topic/6631-the-models-winners-and-losers-in-the-blizzard/

[Quincy]

there was a 00z run of the ECMWF, possibly 00z DEC 26, in which it showed dry slotting into Long Island and portions of Connecticut. (700mb RH was fairly dry) Even though surface and other upper level features on the NAM were similar, the NAM did not show such dryslotting.