Network News

X My Profile
View More Activity
The new Washington
Post Weather website
Jump to CWG's
Latest Full Forecast
Outside now? Radar, temps
and more: Weather Wall
Follow us on Twitter (@capitalweather) and become a fan on Facebook
Posted at 11:15 AM ET, 12/29/2010

Where did Sunday's forecast go wrong?

By Wes Junker

Why were the "best-bet" forecasts of snowfall amounts for the Boxing day snowstorm (or No-mageddon) so poor in the Washington area? Despite expressing that this storm was a exceptionally tricky one, some readers are miffed or more by the mostly no snow result. The following paragraphs outline our thinking during the storm and where I think we might have done better.

The exact track of the storm was always uncertain as noted in the Dec. 20 blog. Through Dec. 23, we continued to stress there was considerable uncertainty in the forecast.

To get a major snowstorm in the Washington area, two streams of flow needed to phase (merge together) at the right time and, for that to happen, two conditions had to be met: 1) the storm would track far enough west to get us in the wrap around snow band, and 2) the low needed to develop quickly enough to our south to get some easterly flow to tap Atlantic moisture.

Essentially, if either condition was not met, the snowfall in the D.C. area would be minimal.

What made the forecast exceptionally difficult was that the various forecast models could never quite agree on the track and on how quickly the low would develop. To make matters worse, Christmas Eve and into early Christmas morning the models seemed to be approaching a consensus toward the snowier scenario.

The forecast difficulties magnified on Christmas eve morning when the GFS model jumped well west of its earlier forecast giving Washington more than an 10" of snow. The ensemble members (remember ensembles are lower resolution runs with the initial conditions tweaked to help assess the uncertainty) trended to the west and wetter. However, NOAA also issued a statement that the initial analysis of the GFS might have problems making it hard for us to assess what that run meant.

However, the 1 p.m. GFS run event trended wetter suggesting that DC might get a foot of snow and its ensembles again were supportive. The other higher resolution models were much less bullish on snow at that time. At that time, CWG bumped up the probability of an inch to 50%.


Total model precipitation from the GFS (top panels) and NAM (bottom) for the 12/26 storm simulated at the times indicated. Source: WxCaster.com

The real forecast dilemma developed on Christmas eve into Christmas morning as the models appeared to be converging toward a consensus. The 7 p.m. GFS model (to the right, top left panel) gave Washington between 0.75 and 1.00 inch liquid equivalent (approximately 7-10 inches of snow). That same evening the NAM model (top right panel) still forecast a miss but the run 6 hours later (not shown) brought the low track closer to the coast (bottom right) and gave Washington between .25" and .50" (3-5 inches or snow). The European model that night also predicted a 3-5 inch snowstorm over the area.

The 1 a.m. GFS model Christmas day held onto its very snowy forecast from earlier. By 5 a.m. Christmas morning, the National Weather Service issued a Winter Storm Watch for the area east of I-95 and they expanded the area mid-morning. At that point, CWG noted that accumulating snow was now likely. Based on this guidance and a quick peek at the NAM model run from 7 AM Christmas morning, I uttered the infamous quote that I was quite "bullish" on the event. However, within that article Jason noted "there remains a small - but not trivial chance - that the storm just skirts us."

The GFS Christmas morning (see above top right panel) started backing off giving D.C. around 5 inches of snow. In view of the slight eastward shift in guidance, the CWG issued the graphic below (to the right) indicating there was a 1 in 4 chance of the storm missing the area. Still at that time, the GFS and the SREF ensemble guidance were supporting the 3-6 inch "best bet" forecast issued at that time.

first-call.jpg
Snowfall probabilities CWG provided for the D.C. metro region Sunday and Sunday night on the afternoon of 12/25.

Even after the operational models started backing off, the ensemble guidance continued to support snow. However, the models were also indicating the precipitation gradient (how quickly the precipitation changed as you went west to east) was stronger than with many storms. Nonetheless, the SREF ensemble model guidance during Christmas day became even more adamant supporting the idea of getting 5 inches or more of snow. Every member of the SREF ensembles gave D.C. over 5 inches of snow on the 4 PM run (see the trend of the ensembles below). The forecast decisions though dinnertime on Christmas day were as good as one might expect given the model guidance that was available.

sref-bust.jpg
SREF ensemble based probability of getting .50" or more liquid equivalent on 12/26 and how they trended from early morning (left) though late afternoon (right) on Saturday, 12/25.

The most egregious fault with the forecasts of the event was how long meteorologists (not just CWG forecasters) held onto their initial best bet snowfall forecasts and how slow they were to knock back amounts in the face radar and satellite imagery that suggested a near miss. Without doubt, forecasters were late in making changes.

A simple extrapolation of radar and satellite trends did indeed suggest a near miss yet forecasters stubbornly held onto their forecasts. Why? Because sometimes extrapolation of clouds and weather even in 12 to 24 hour forecast can lead a forecaster astray. Extrapolation of satellite imagery has led to several major snowstorm surprises (busts).

The satellite and radar imagery 12 hours prior to the President's Day storm of 1979, Veteran's Day snowstorm of 1987 and the February 16, 1996 event all come to mind as storms that displayed rather innocuous precipitation field the night before the snow began. Satellite imagery of the President's Day storm indicated all the bright clouds and precipitation associated with the warm advection were shifting out to sea mostly south of DC and that the city might only add another inch of snow to the four that had fallen earlier in the evening. That imagery prompted one veteran forecaster to state that there would only be "another inch with the vort. [upper level energy]" There was, but it occurred every 15-20 minutes for several hours.

While no one thought Sunday's storm President's Day storm potential, the CWG crowd knows that sometimes the models have problems handling the smaller scale precipitation banding when mid-level circulation center closes off to our south. Hence the hesitancy to quickly change the forecast and instead to assert that the amounts probably would be on the lower end of the range shown in our graphic.

By late Saturday evening, the radar was still suggesting that the heaviest snow might stay just to our east. The conflicting guidance presented us with a dilemma, should we lower amounts and then risk having to raise them if a band of heavier snow developed over us. With such uncertainty, we opted to hold steady with our amounts.

Our biggest fault during the event was being a little slow to cut back amounts based on radar. However, extrapolation of radar doesn't always work. Obviously, in this case with the more rounded upper trough (dip in the jet stream) and with a low that deepened a little late to get us in the game, radar and satellite told the story. There were differences between this event and some of the events noted earlier in the discussion. Also, we probably should have given the NAM model, which showed the least precipitation leading up to the event, a little more weight as it does have higher resolution. However, it typically scores a little worse on average than the SREF ensemble mean.

In summary, we probably should have relied a little more heavily on the radar and satellite trends especially during the day of Dec 26. Even better, we should have paid better homage to the weather gods, they are a fickle bunch.

By Wes Junker  | December 29, 2010; 11:15 AM ET
Categories:  Latest, Recaps, Winter Storms  
Save & Share:  Send E-mail   Facebook   Twitter   Digg   Yahoo Buzz   Del.icio.us   StumbleUpon   Technorati   Google Buzz   Previous: The No-mageddon D.C. snow hole from space
Next: Is Siberian snow, eastern U.S. cold link legit?

Comments

You might want to sit down for this one... it's 77°F in Santo Domingo...

http://www.intellicast.com/Local/Weather.aspx?location=DRXX0009

Posted by: BobMiller2 | December 29, 2010 10:19 AM | Report abuse

Nice post, Wes. It would be interesting to find an old radar or satellite loop of a storm in which forecasting just based on radar/satellite trends 12 hours in advance would have led to a busted forecast. The reason being that when a coastal low finally takes shape and begins drawing on Atlantic moisture, the precipitation shield expands greatly, appearing almost out of thin air. So the critics who have been saying that CWG should've just seen what the satellite/radar showed and obeyed that guidance, instead of the models plus forecaster experience etc., weren't necessarily correct.

Posted by: afreedma | December 29, 2010 10:35 AM | Report abuse

Wes,

Has there ever been a storm that gave heavy snow to western NC and eastern TN that also gave heavy snow well west of NYC and Boston? And totally bypassed DC? If so, I can't remember it.

The mind boggling thing was that the storm kept swinging out to the east even while the 500 mb flow was rapidly backing. In some ways, this storm defied common sense.

On a scale of 1 to 10 with respect to degree of forecasting difficulty, I would rate the storm a 9.9.

During elections, they always talk about the "margin of error" and say races are "too close to call". In this case DC was always within the margin of error and even until the last day, the event was too close to call.

Posted by: frontieradjust | December 29, 2010 10:39 AM | Report abuse

Andy, I can't remember one but suspect it has happened sometime in the past. It certainly was one of the tougher. It ranks back there with the heavy snow forecast from a miller b that produced sunny skies back around 2000 but hammered areas further north. I consider than storm my biggest snow bust. In that one I followed the higher resolution eta and mm5 when the gfs said no snow. lol

Posted by: wjunker | December 29, 2010 10:47 AM | Report abuse

Wes - thanks for your detailed post & all your time watching computer models during a holiday weekend when I'm sure you would have prefered to be doing other things.

I really do not want your head on a silver snowflake engraved platter.
That would certainly discourage other CWG members from making similar "bullish" statements for future events.

This area has always seemed painfully difficult to forecast for snow events. Better luck next time.

Posted by: FIREDRAGON47 | December 29, 2010 10:59 AM | Report abuse

FIREDRAGON47, Thanks, I'm partial to my head, it may not be pretty but it's the only one I have. As to the bullish comment, given the preponderance of guidance suggesting at least 3 inches of snow, I'd probably make the same statement again with similar guidance, lol, though I might temper such a statement a little more.

Posted by: wjunker | December 29, 2010 11:05 AM | Report abuse

@CWG and everyone else -

It was made extraordinarily clear for the duration of the lead time up to the event that this was a lower than low confidence forecast. Models were everywhere. They were on the moon one run, in 3D on another and a hologram the next. Add that to La Nina, the spiked eggnog and a Redskins win and there was no reason to bash anyone for the difficult call that the mets had to make. Look at where the snow fell (ie the Deep South and the Mid-Atlantic beaches). This was clearly a near impossible one to handle. Thanks to Wes and the Gang for making it crystal clear that the forecast was anything but crystal clear. Im the weather go-to guy for my friends and family and I was just shrugging my shoulders and prepping for anything from a foot a few flurries. This one didn't pan out. There will be more. Rant over. Happy New Year!!!!

Posted by: DullesARC | December 29, 2010 11:14 AM | Report abuse

DullesARC, thanks! At least the New Year's eve forecast appears to be one where we can say with considerable certainty that the weather will be pretty nice.

Posted by: wjunker | December 29, 2010 11:18 AM | Report abuse

Beautiful morning today....was reading the Post yesterday and it said DCA has received 2.1" of snow in December, DOUBLE the average for the month. So not quite sure what snow lovers are complaining about, unless you think last year is the new baseline for snow. If so, you are going to be sorely disappointed. DC averages about 16-18" of snow per year, and in my experience we get one big storm every 7-8 years. Thinking otherwise is a fantasy...

Posted by: weatherdude | December 29, 2010 11:22 AM | Report abuse

There's an old saying I'm sure Wes knows: Live by the models, die by the models

In my recollection the worst inccidence of forecasters across the board ignoring radar and satellite in favor of models only was the "Surprise Snowstorm" of Jan 24-25th, 2000.

"What went wrong in the forecast process in this case is well described in a paper (section 2b) by highly respected research and teaching meteorologist, Lance Bosart. Critical mistakes included forecasters taking the models at face value (with virtual disregard for radar and satellite) and relying too much on "deterministic thinking." That is, to forecasters then it was essentially an absolute yes/no for a snowstorm in the D.C. area, and the models said "no" until very late in the game."

Excerpted From:

http://voices.washingtonpost.com/capitalweathergang/2010/01/major_snowstorm_ambushes_washi.html

Posted by: SteveT-CapitalWeatherGang | December 29, 2010 11:37 AM | Report abuse

@weatherdude

Who says we're not allowed to be hopeful? Mother nature doesn't care about records and statistics and all that stuff, if she wants to give us a blizzard, she'll give us a blizzard, if she wants to give us a 100° day, she will.

Records are broken all the time, and while it is rather unlikely we'll se another blizzard this year, it's not impossible, just unlikely.

Posted by: BobMiller2 | December 29, 2010 11:43 AM | Report abuse

Bob, of course Mother Nature does not care. She could also give us no further snow this winter just as much as a blizzard.

Posted by: weatherdude | December 29, 2010 12:06 PM | Report abuse

Steve, but that storm also showed the problem with just extrapolation of the clouds and weather. Just moving it along was fine for the DC area but for North Carolina, the cloud signature was not that great until convective snows broke out near Charlotte if I remember correctly. 24 hours prior to the event, the imagery really was not that helpful. Once the storm really got going, pressure falls, satellite trends and lots of convective snows over RDU suggested DC would get hit hard. The non-linear development of such storms sometimes causes problems when extrapolating what's going to happen based on radar and satellite or on models. The guys at HPC were faster than most to recognize dc was in trouble during the the 2000 storm. Unfortunately, I missed that storm, I was in St Louis giving a talk on snow forecasting.

we certainly should have heeded what we saw on the imagery more and earlier in this case but when you make forecasts, you try to play the odds or at least what you perceive them to be.

Posted by: wjunker | December 29, 2010 12:06 PM | Report abuse

@ weatherdude

You make a good point.

Posted by: BobMiller2 | December 29, 2010 12:21 PM | Report abuse

Wes and the rest of the CWG team: as sad as I may have been, I certainly don't hold it against you all that the storm didn't pan out, and I'm grateful for your always insightful, interesting and, most importantly, informative commentary. It's much appreciated, and I for one would always prefer to error on the storm side and be over-prepared than under.

That said, I did have a question - there are two main models you refer to (GFS, Euro) and several other "lesser" (?) models - Canadian, UKMET, NAM, others. I'm surprised there's no "meta-model" that looks at the other model runs and attempts to create a weighted average based on how the model did previously. The obvious analogy here is Nate Silver's polling model that essentially weighted each pre-poll result by its accuracy in the past. A simple implementation could be looking at forecasts for liquid at DCA 24-48-72-96 hours out compared to actual liquid that fell there from each model run. Is there such a probabilistic "meta-model" around or am I just being overly hopeful?

Cheers, and thanks for the great work!

Posted by: vnangia | December 29, 2010 12:24 PM | Report abuse

Just curious -- Did the Euro model that was predicting snow far in advance show the unusual precipitation gradient? Was it "right" in that respect?

Posted by: Groff | December 29, 2010 12:35 PM | Report abuse

What surprised me was that even thought the storm formed well east places southwest of us (Roanoke, Lynchburg) squeezed out a few inches from this on xmas day and that precip shield looked like it was gonna get us Xmas night but seemed to fall apart once it got to Fredericksburg. What caused that?

Posted by: lobp | December 29, 2010 12:36 PM | Report abuse

Bottom line: DISAPPOINTED! :(

Posted by: sigmagrrl | December 29, 2010 12:48 PM | Report abuse

Here's the explanation...Originally I had no dances scheduled last weekend; that alone should explain the storm bust.

Then on late Friday or Saturday a friend called and noted a Christmas night dance at the Hollywood Ballroom so we made arrangements to reschedule for a dance night on Christmas. Later Saturday afternoon my friend called; travel to/from the dance would be off due to the snow threat...no dance Christmas night after all.

That must have been about the time the snowstorm decided to "bust" locally. To be sure there was continuous light snow Sunday afternoon but no accumulation until very late by which time the storm started winding down. Areas from Philadelphia north got a huge blizzard however.

Posted by: Bombo47jea | December 29, 2010 1:26 PM | Report abuse

Sunday's forecast was actually right for a long time. All week long the OCMs were saying the probabililty of snow was low. The NWS percentages were dropping. Then suddenly on Friday, was it, everyone starting jumping on the latest NAM/GFS bandwagon.

Certainly, paying close attention to the NWS's Mosaic National Weather Loop on Saturday afternoon could have let the air out of the snow balloon by the 6 p.m. newscasts. The Roanoke Rail Cam was another indicator. Roanoke was getting hardly more than a moderate flurry late Saturday. (I checked the next a.m. and snow there was minimal.)

When I hear an OCM use the phrase, "based on the models..." I think, "uh, huh, here we go with the weasal words."

I'd much rather hear a forecaster say (or write), "based on my knowledge and study of historical storm patterns, radar imagery, and model projections, our best chance of snow is ...." This, I can respect.

We're much more apt to get this careful reasoning on CWG, than from some OCM or NWS "cry snow wolf" forecasts.

Posted by: JerryFloyd1 | December 29, 2010 1:44 PM | Report abuse

I recall it was either Snowmageddon or a storm a few years back that people were calling a bust since the storm started off pretty slow and the radar did not look impressive. However, once the low got cranking off the coast, we were slammed overnight.

Hence, even looking at the radar isn't always helpful. Let's just agree that we can never predict the future with 100% accuracy and precision.

Posted by: JTF- | December 29, 2010 2:19 PM | Report abuse

@JTF
You were probably trying to recall Feb 11-12, 2006.

Posted by: ronbcust | December 29, 2010 3:25 PM | Report abuse

Wes, your description lacks specific timelines. I vividly recall beinging at the HPC map disccussion at 11:30AM the morning before the storm and droping in on HPC and monitoring NWS Sterling that afternoon. I recognize how things can evolve such that simple extrapolation can often be misleading. But, in this case, by mid afternoon the radar/satellite strongly signaled that the models through 18Z (1PM) were out to lunch.

That signal was, at least as far as the outside world would know (public), ignored until after 9PM in NWS offical forecasts and TV broadcasters (including Bob Ryan, as I documented). I refered yesterday in a comment to "meterological cancer" - buying into the models at the expense of looking out the window via radar/satellite - and this was a prime example.

I think our differenes, if any, are the specifics of timing when forecaster recognition of reality finally overtook model unreality.

Posted by: SteveT-CapitalWeatherGang | December 29, 2010 4:26 PM | Report abuse

I was referring not to us but farther south towards NC where they got a surprise when the convection broke out. If you went back 24 hours from then, I don't think you would have made the same call but maybe my memory is faulty since I was in St Louis at the time and was looking at stuff on AWIPS during times when I wasn't talking. Certainly for our area, you're right. Bruce Terry called LWX at around 5 because based on Satellite data, radar and pressure falls it was obvious that the models were wrong.

certainly in our recent miss, local forecasters including CWG should have paid more heed to the imagery and the fact that the best pressure falls were off the coast.

Posted by: wjunker | December 29, 2010 4:39 PM | Report abuse

lobp, I think if you look at the 850 mb (5000 ft) winds and temperatures during that period you can see that there was some warm advection across that region as the 850 trough/ almost low shifted east. However, the northerly winds stayed over us longer and then the 850 low reformed or tracked far enough south and east that we never really got into the heavier wrap around. also the low just didn't develop quickly enough far enough south for us to really tap into the atlantic moisture. Farther north they tapped into the atlantic moisture and got into the deformation zone cold conveyor belt associated precipitation (wrap around to the northwest and west of the low) as the low bombed out.

Posted by: wjunker | December 29, 2010 4:48 PM | Report abuse

Sooooo-When is our next real chance of measurable and sleddable snow?

Posted by: sobolasobola | December 29, 2010 6:21 PM | Report abuse

Anybody here old enough to remember a TV meteorologist by the name of Louis Allen? Well he was a very accurate Channel 7 weather forecaster back in the 1960's who drew all his maps by hand from memory live in front of the viewers. I once interviewed him for a college project, and know where he said he got most of his over-water data from? Ships. Miraculously he was able to compose this ship (as well as land station-reported data) in his head into an amazingly accurate forecast which he rarely "busted".

Just thought I'd relay that story so as not to encourage total reliance on computer models all of the time, to perhaps give a small bit of credence to other sources (even gut feeling) particularly in times when the models have so much trouble getting a handle on what's going on such as in this one.

In this storm, I monitored the radar returns from my house near McLean about midnight Saturday night, noting the precipitation was just reaching Fredericksburg. Said to myself...well if the ground ain't coated when I get up, forget it. About 8am next morning after noting the snow still hadn't reached us, the radar now showed a fairly straight precip line about 20 miles SE of DC, and over an hour essentially didn't change its position. Then my brother called from Doylestown PA wondering what my totally non-professional opinion was, and I said it is clearly going to miss us.

Later in the day when we did get some flurries and a dusting, and radar showing some spots of moderate snow a few of which were actually west of us, was thinking well maybe a late-developing snow band will fill in and give us 3-4" after all. But over the next couple hours the returns literally evaporated. Missing that storm fragment appears to have been sheer luck.

Being 63 years old and an off-and-on monitor of weather around here for at least 50 of those years, am wondering maybe the Louis Allen "figure it out in your head" approach perhaps is still a tool to be considered on occasion.

Posted by: kenkimg | December 29, 2010 9:23 PM | Report abuse

Kenkimg, I'm also in my 60s and remember Louis Allen. In this case, your non-professional opinion was better than my professional one. I think most mets try to figure things out in their heads. I know I do but sometimes don't figure things as well as I'd like.

Back when Allen was making forecasts, models weren't very accurate. Allen may have been accurate for the time, but I can remember some huge busts that cause me grief because I didn;t do my homework. Even then I was a weather weenie. All forecasters sometimes have forecast busts. I've certainly had my share.

Posted by: wjunker | December 29, 2010 9:33 PM | Report abuse

I remember Louis Allen had a horrific bust for the Kennedy Innaugurial Storm in January 1961.

The night before that storm, he actually took the surface low on a track north of DC and predicted "snow flurries". DC got 8" which coincided with one of the worst evening rush hours in DC History.

Louis was so upset about the blown forecast that the next night he was in tears.

During January 1962, he said we were going to get back to back storms, 3" from the first storm and a foot out of the second storm. He called it the "double barreled shot gun". We got a total of a half inch from the two storms.

Posted by: frontieradjust | December 29, 2010 10:53 PM | Report abuse

Hoooray!

I'm glad all the forecasters were wrong!

No snow is good.

Posted by: DC_MAN88 | December 30, 2010 12:00 AM | Report abuse

Hoooray!

I'm glad all the forecasters were wrong!

No snow is good.

Posted by: DC_MAN88 | December 30, 2010 12:31 AM | Report abuse

So the forecasts where wrong. I really can't see the big deal here. The Gypsy with the crystal ball is also often (actually always) wrong. If her predictions come true I can stop working, marry a princess and live happily ever after..

I think it would be much worse if it was the other way around. A snow fall like February earlier this year and no warning at all..

I do like this insights given in this article how forecasts are created, but somehow it feels a bit like an apology.

Posted by: zyprexia | December 30, 2010 9:03 AM | Report abuse

Someone I know well and has experience in such things pointed out on Sunday morning (Dec 26th) that IR and visible satellite indicated the heaviest precipitation was just to the east and north of developing low center and snow bands to west pulling rapidly away towards north and east with dry air moving in behind system. He added, looks like a for-get-about-it; and suggested it was time to call off the storm for DC except for possibilities of some flurries. Better luck next time!

It may be just luck that this is exactly the way it turned out, but it re-emphasizes when getting this close it’s time to relinquish one’s dependence on models and ,effectively, look out the window (via radar/satellite).

In a conference room at NCEP, there is a sign pasted on a large window looking out to the scene outdoors. The sign reads, “NOWCASTING” !

Posted by: SteveT-CapitalWeatherGang | December 30, 2010 10:25 AM | Report abuse

My friend read the NOAA website and told me not to worry, the storm was so fast-moving DC was unlikely to get anything much. So - if he got that from NOAA, the official weather people, what were our forecasters reading?

Posted by: AdventurerVA | December 30, 2010 11:19 AM | Report abuse

My friend read the NOAA website and told me not to worry, the storm was so fast-moving DC was unlikely to get anything much. So - if he got that from NOAA, the official weather people, what were our forecasters reading?

Posted by: AdventurerVA | December 30, 2010 11:25 AM | Report abuse

Interesting explanation ... but no need to apologize! I've looked at the various pictures of our 'snow hole,' and I'm astounded at how we managed to avoid it all. Though I'm not a meteorologist, I think it would have been much more difficult to predict the extent of our 'snow hole.'

Posted by: SilverySpringlike | December 30, 2010 11:32 AM | Report abuse

The comments to this entry are closed.

 
 
RSS Feed
Subscribe to The Post

© 2012 The Washington Post Company