In this blog, I will examine this storm and note that it reveals some major issues with modern weather prediction. Issues that are general and reflect problems we had with the Northwest windstorm bust of October 2016.
As noted note above, this forecast could be considered a great success for modern weather prediction, with a real heads up 4-5 days ahead of time. Here are the NWS official surface analyses at 5 AM and 1 PM on Tuesday, March 14th (sea level pressure shown). The low (central pressure of 986 hPa) rapidly intensified (to 978 hPa) as it moved to the tip of Long Island)
Virtually all the major modeling systems (NOAA GFS, European Center, UKMET office, Canadian CMC) honed into a similar solution, with minor (but important) differences in track and intensity.
The implications for the Northeast U.S. were serious. Such a strong storm, with attendant large pressure gradients, would certainly produce strong winds over the Northeast. With cool air over the continent, substantial snow was in the offering for many folks. Where heavy snow and winds came together, blizzard conditions (heavy snow, greater than 35 mph winds, low visibility) would occur.
And predicting this line is not easy. It is dependent on the movement of different temperature air around the storm. It is dependent on the intensity of vertical air motions (which can influence temperature). It is dependent on the amount of precipitation (since the evaporation and melting of precipitation can change the temperature profile). It is dependent on coastal effects (where temperature and surface variations are large). Bottom line: the rain-snow line can be a real challenge for even the best models and forecasters. And in a coastal area, it is critically dependent on the exact track and intensity of a storm. And an inability to diagnose and prediction physical details like cloud processes.
On Friday and early in the weekend, most of the model guidance suggested that the rain-snow line would be just offshore, indicated New York and Boston would have mainly a snow event, with the most probable amounts being 1-2 feet. This guidance included high resolution deterministic forecasts and lower-resolution ensembles of many forecasts (used to define uncertainty).
On Sunday, the NWS SREF (Short-Range Ensemble Forecasting System) showed some uncertainty for the total snow forecast at New York's JFK airport (see below), with an ensemble average of around 12 inches and a range from an inch to 29 inches.
The ensemble-based probabilities of precipitation type (below), were dominated by snow (blue lines)
But by Monday morning, the situation had changed substantially. The morning (8 AM) run of the NWS SREF ensemble had decreased the snow to about 10 inches, with more of the ensembles going for lower amounts.
The probability of rain (green) was gaining on snow (blue) and was higher than snow at 11 AM (18 UTC) Tuesday.
But on Monday, we were closer to the event and forecasters had powerful, more modern tools available.
The future of forecasting such events is high-resolution ensembles, using convection-allowing grid spacing (like 3-km). The National Weather Service does not have such an ensemble system (and it should!), but the National Center for Atmospheric Research (NCAR) does (but a small one with only 10 members). And NCAR ran such an ensemble starting 5 AM on Monday. The average of the ensemble (the ensemble mean) showed a huge coastal snow gradient, with 6-10 inches over Long Island, with heaviest amounts to the west and north. But a slight shift would have huge impacts on NY snow. This kind of situation suggest large uncertainty.
The NCAR ensemble showed lot of uncertain for JFK airport, with a mean of around 8 inches (see below) and a spread from 2-17 inches. It also indicated a change from snow to rain (not shown)
And now the problem. Here is the National Weather Service forecast released 4 AM Monday, one that uses their new approach to presenting uncertainty. The suggested a most probable value for JFK airport of 17 inches and a range from 8 to 22. This was the forecast that was in place for most of Monday morning when a lot decisions were being made. And in addition, there was a winter storm warning and blizzard warning in place over New York on Monday. Nothing like a blizzard warning to get media juices going.
And as uncertainties in the forecast were increasing, the NWS doubled down on the blizzard warning:
By late on Monday, the models shown above were clearly edging towards a lesser event and shorter-range rapid refresh models like HRRR was becoming available (HRRR is initialized every hour and run out for 18 hr). HRRR uses very high resolution (3-km grid spacing), is initialized with lots of regional assets, and the snow output makes use of variable density snow, which gives more accurate totals.
Here is the 18h HRRR snow total starting 5 PM Monday, March 14th, which encompasses pretty much the whole storm in NY. 3-6 inches over the eastern end of LI and the immediate south shore of LI, increasing to the NW, with perhaps 12-15 inches would be expected over the NW side of NY city. Roughly 10-11 inches around JFK.
Still too much, but the pattern is very good. A new version of HRRR, called HRRRx, does even better, with the suggestion that better physics descriptions helps with the overprediction of snow.
Now I have been fixated about snow in the above discussion. Winds were important as well, and the models and the NWS forecasts were very realistic about their strength and duration.
So what is the bottom line of all this? In many ways this was a very successful forecast that shows how far weather forecasting technology has come.
1. The threat of a major Nor'Easter was identified 5-7 days in advance.
2. The large scale prediction was quite accurate, but there were minor but important track and structural errors.
3. The winds were forecast quite well.
4. The general structure of the snowfall was handled reasonably, but the rain-snow line was displaced too far to the southeast by the models, resulting in an overprediction of snow over NY and Boston.
5. There was considerable evidence of forecast uncertainty in the days before, with the possibility of less snow becoming more evident on Monday, March 14. In particular, it become clear by Monday afternoon that a transition from snow to rain was probable over LI and much of NY City.
6. Forecasters held on to the heavy snow/blizzard forecasts on Monday and probably should have backed off that forecast by Monday afternoon, highlighting the change to rain more.
So what are take home messages?
1. The NWS must continue to improve its forecast technology, including high-resolution ensembles that provide uncertainty and probabilistic guidance. The NWS has been delaying too long in building a convection-allowing operational ensemble system. Congress needs to intervene if necessary.
2. NWS forecasts have to transition to a probabilistic framework, where probabilities are given for various outcomes. The old style watch-warning system is really not effective in a new world of uncertainty and probabilistic prediction.
3. Forecasts much continuously evolve as more or better information comes in. Forecasters should not try to second guess users or keep intense forecasts in place to encourage "right" decision making.