Stumbling about in the snow

Let's talk a bit about the forecasting of this blizzard event in the northeast yesterday and today.  A lot of people are writing about what went into this forecast and why New York and Philadelphia did not get nearly as much snow as was forecast (thought Boston and points east did!).  The common theme?  Forecasters didn't adequately communicate the uncertainty regarding this forecast.

Just how uncertain was the forecast?  Here's an example of the predicted snowfall accumulation at JFK airport in New York City from several of our model systems---the NAM (the reds and greys), the GFS (the blues) and the Rapid Refresh (green) starting at 12Z yesterday (yesterday morning).

The numbers drop over time after reaching their peak because this forecast factors in melting and compaction of the snow which will decrease its depth over time.  You'll notice that all of the higher forecasts (20-34") are all red, orange or grey---these estimates all were based on forecasts from the NAM model.  On the other hand, the lower forecasts (6-12") are all based on the GFS (or RAP) forecasts.  In total, this is a lot of uncertainty---as few as 6" or as much as 34".  6" is a normal, everyday winter snow.  36" will keep a city shut down for days.  And this is the morning before the event, when leaders and other interested parties have to make decisions about committing resources, closing things down, etc.  What is a forecaster to do?

It turns out that the GFS had the low pressure center moving further east.  Here's the 30 hour GFS surface pressure forecast (the black contours) from the 12Z run, so this this a forecast for where the low would be around the middle of the day today:

Note the center of low pressure is forecast to be southeast of Martha's Vineyard, with a longitude east of Cape Cod.  This GFS forecast kept the heavier snow bands east over Long Island, but not over New York City, hence the lower snow totals that we saw above.  Here's the exact same forecast, but from the ECMWF model (thanks to Cliff Mass for grabbing these):

Here the low pressure center is further to the west---centered straight south of Cape Cod at this time.  The ECMWF had snow bands further west as well---over New York City.  The NAM happened to agree more with the ECMWF as to the position of the low.

From reading the National Weather Service forecast discussions, this seemed to be what led to the decision to go with the higher snow totals.  This is from the 630 AM EST forecast discussion yesterday at the New York Weather Service Office:

"THE 00Z ECMWF REMAINED VERY CONSISTENT WITH ITS
PREVIOUS RUNS. IT DID LOWER QPF SLIGHTLY...BUT THE OVERALL EVOLUTION
OF THE SYSTEM REMAINED SIMILAR TO ITS 12Z RUN. DID NOT WANT TO MAKE
DRASTIC CHANGES TO THE EXPECTED SNOW AMOUNTS AND HEADLINES WITH
JUST ONE MODEL CYCLE. IN FACT...THE LATEST 06Z NAM HAS COME INTO
CLOSER AGREEMENT WITH THE 00Z ECMWF."

Earlier in the discussion, though, they suggest that the spread in the models, particularly at such short forecast lead times, was more than they would like.  And again in the 1:34 PM EST discussion after the 12Z models shown above had come in:

"THE STORM APPEARS ON TRACK. THE 12Z GFS IS AN OUTLIER FROM THE 12Z
NAM AND 12Z ECMWF. HAVE UPPED SNOWFALL TOTALS ACCORDINGLY WITH
MOST AREAS IN THE 20-30 RANGE FOR THIS EVENT."

So that was the logic.  And, in the past, this logic has served us well.  The European model (the ECMWF) has long been considered the "gold standard" of our weather prediction, usually surpassing the US GFS model in terms of forecast skill.  Indeed, it was the ECMWF's superior performance to the GFS in predicting the path of Hurricane Sandy that recently led to a large investment from Congress to improve the GFS.

What's more, even by 00Z yesterday evening, it was hard to distinguish which model is doing better.  The 12-hour forecasts from the GFS, ECMWF and NAM had the low in virtually the same place at that time, so it was hard to determine which way it would go.  It's also challenging when the low is off the coast, as it was here, to get an accurate fix on what's going on...there are limited surface observations over the ocean, so we have to rely on satellite data which can be tricky to interpret.

So what ended up happening?  The low actually moved to the east, as the GFS had been predicting.  Here's a map put together by @anthonywx showing the analyzed position of the low today at 15Z (the black L) and where the ECMWF forecast it would be 3 hours later (the red L, same as the position shown in the maps above):

So the GFS solution ended up playing out---the low was further east as was a lot of the heavy snow.  This didn't stop Boston and Massachusetts from getting heavy snow, but it definitely pulled things away from New York and Philadelphia.

Note the difference in the position of the low---120 miles.  At a quick glance, as far as a 30-hour model forecast goes, that really isn't too bad.  It just so happened that there was a strong gradient in snowfall on the western edge of this storm, at that gradient happened to pass through the largest city in the country.

There was a lot of uncertainty, but forecasters used knowledge that has served them well in the past to refine their forecast in support of decision makers.  Personally, I think that this is a good case of "better safe than sorry".  Would giving the public better estimates of the actual forecast uncertainty (which was large in this case) have helped people make better decisions?  Should we have put up graphics that said the snowfall could be anywhere from 6"-36" in New York City?  What was the actual forecast uncertainty? Again, I think that it was right of the forecasters to use their experience to try and refine that number, even if it may not have been the best refinement here.  Could an automated system have done better?  Much will be debated in the days to come.

And, for what it's worth, the GFS brings another Nor'easter through the area next Monday.  Who (or what) will we trust then?