The Fearless Girl statue stands in the snow March 14 in New York. (Don Emmert/Agence France-Presse via Getty Images)

Early this week, the National Weather Service badly missed a snowstorm forecast for New York City. It predicted 12 to 18 inches of snow, and seven inches fell.

It was the second errant snow forecast in the Big Apple in three years. In January 2015, it predicted 24 to 36 inches of snow, and 10 inches fell.

But Louis Uccellini, director of the Weather Service, is standing behind his forecasters. He says the forecast was complex and that the Weather Service forecasters communicated critical information that allowed the city to be ready for what turned out to be a nasty storm, despite the lower-than-predicted snow amounts.

In every Northeast snowstorm, there are places that are easier to predict snowfall amounts for and those that are more challenging. The challenging locations are those along transition points in the storm — typically either where the rain-snow line sets up or where the heaviest precipitation starts and stops. Just slight movement in these razor-sharp zones can make the difference between just a little snow and a ton of snow. In both this week’s storm and the 2015 event, the Big Apple shifted in and out of these zones as the storm approached.

When the margin for error in forecasts is so small, forecasting uncertainty becomes critical. After the 2015 event, in which the Weather Service communication of uncertainty was deemed inadequate, Uccellini said: “It is incumbent on us to communicate forecast uncertainty. We need to make the uncertainties clear. We’re going to review this very carefully and assess a different approach as we deal with these types of storms.”

Yet here we are, two years later, and the Weather Service is again facing questions about whether it is doing an adequate job in communicating uncertainty in high-stakes forecasts.

I sent questions to Uccellini about the challenges this storm posed and how the agency is working to improve how it communicates these difficult forecasts. His responses have been lightly edited for length.

Capital Weather Gang: What is causing these large forecast errors?

Uccellini: Predicting winter storms, their track, intensity and rapid intensification, precipitation types and amounts have proved to be one of the more challenging forecasts the entire weather community has to deal with. The predicted snowfalls did accumulate tens of miles to the west of major cities, in some cases with record amounts. The preparations we witnessed for this storm were prudent for the major cities in the storm’s path. The commutes were expected to be dangerous, the impact to transit was well-forecast in advance, and appropriate decisions were made for public safety on the ground and in the air. This, despite the lower than expected snow totals.

[In these] two events, the model forecasts leading up to the events show extraordinary variability in the track and intensity of the storm right up to the time they begin impacting the major metropolitan areas along the East Coast.

I’m incredibly proud of the work of our forecasters at NWS and meteorologists in the whole weather industry for their dedication to public safety during these events as they have to grapple with all of the complexities noted above. I’m not going to second-guess their forecast decisions. But we have to work to understand the nature of the forecast uncertainties associated with these types of storms and insure we are communicating the potential range of scenarios and the more likely outcome. This is especially true with respect to the interaction with government officials at every level who make important lifesaving decisions as they prepare communities for the impacts of the impending major storm systems.

Three out of four blizzard warnings for NYC in the past four years have verified. We may not have met the blizzard criteria for this storm within the New York City metropolitan area, and we will review the use of the blizzard warnings, but we all have to remember that this forecast miss is measured in tens of miles in the storm track and related to the interaction of small scale physical processes that continue to challenge the forecast and research communities alike.

CWG: Is the Weather Service over-relying on the American (GFS) model, which over-predicted snow amounts for Tuesday’s storm?

Uccellini: The forecasters in the NWS and throughout the weather enterprise have access to and use a number of modeling systems, including their ensemble counterparts, from around the globe. The GFS was actually the first to point to the track of this storm coming up the East Coast. But as it turns out, not close enough to the coast (in this case, the Canadian Model was the first global model to track this storm immediately along the coastline, and thus sweep the warmer air aloft further inland from the Carolinas up into NYC and Long Island into New England).

The global models were struggling with this storm because of the amplification and phasing of separate trough systems; indeed a complex upper air pattern that lends itself to predictability challenges. As we approached the Sunday-Monday time frame, the models seemed to move back and forth between a solution set that would keep the major metropolitan areas in heavy wind-driven snow, and solutions that moved that snow area inland as the coast was marked by a sharp boundary between snow, sleet and then rain.

Forecasting this type of rain/snow line is a classic challenge for forecasters, as well as researchers and modelers. We’d love to be able to get this particular type of forecast pinned down. We knew there would be a sharp gradient. On a broad scale, we were communicating about the potential and uncertainties 8-9 days in advance. Forecasting at the mesoscale ow moved 50 miles further west than expected. And the rest as they say is history.

CWG: Do you think the Weather Service was effective as it needed to be in communicating uncertainty for this event?

Uccellini: There are uncertainties in every forecast. We try to convey a range of possibilities in events such as these, and those uncertainties were communicated in our least/likely/most predictions. We’ve made a lot of improvements in this area over the last couple of years as we’ve learned how our information could be used for impact-based decision support services.

Our core partners base their decisions on a worse-case scenario because they can’t be wrong. And they are beginning these preparations further ahead in time, which means they have to increasingly rely on forecasts at longer lead times and account for the related larger uncertainties in those forecasts, especially for extreme events that depend on certain processes coming together just right to produce the intense storms that threaten communities. In order to account for the range of possibilities, as we approach the onset of a storm system, we provide to them the best, worst and most likely scenarios. Risk assessments are made and decisions are made based on those assessments.

We believe we are getting better at communicating the uncertainties. But clearly there is more work to be done.

And part of our new way of communicating is a focus on impacts. For this storm, we knew there would be major impacts in metropolitan areas. Overall, I believe the preparations and on-the-ground decisions were reasonable, facilitated cleanup efforts, and likely prevented accidents.

CWG: Is it even possible for the Weather Service to communicate uncertainty when it issues a “Blizzard Warning”? On Twitter, the New York City Weather Service did not communicate uncertainty information.

Uccellini: Last year we held off with issuing warnings until the day before the storm because of the large uncertainties associated with the sharply defined northern boundary of heavy snowfall and whether that line separated heavy snow from no snow, or if it would be north or south of New York City. We got some criticism for holding back, with some folks questioning our strategy.

This year, we went out further in advance with a blizzard warning for NYC and Long Island. We’ll be looking at this strategy and thinking about what other approaches we should consider. There is a general push in our forecast offices to lean forward and be more aggressive, to offer more lead time for warnings as requested by our partners. For some events, we are able to do so with certainty. This event demonstrates that sometimes our level of confidence and precision isn’t always where we’d like it to be. We’ll review this and consider strategies — working with social scientists — for deciding on the timing of issuing warnings when we are caught between predictability and uncertainty.

CWG: What is the NWS doing to improve these forecasts and their communication?

Uccellini: Our Eastern Region headquarters is going to review this snowstorm and specifically how we handled the blizzard warnings. The results of this review will be considered by leaders within our agency as the larger policy issues are brought forward.

Again, I think we handled the issuance of the blizzard warning correctly for the 2016 storm by holding off on issuing warnings until the night before because of the uncertainty of the sharp boundary across NYC. For that case, waiting until we were sure the night before the onset of the storm worked for emergency managers. But we have to acknowledge that there is no magic bullet for every storm. Certain types of storms are more predictable than others as even you alluded to in your detailed summary of the forecast challenges for the Washington D.C. area, and in the end, that factor could play a dominant role in how we proceed.

Let me end here by noting after reviewing a hundred or so storms, every event is different. For this event, we succeeded as we successfully alerted the public a week in advance to the potential for a major winter storm along the East Coast. And we should continue to recognize, and remind ourselves, that the mesoscale complexities associated with these major winter storms before, during and after their development and subsequent evolution will pose forecast challenges now and the foreseeable future, for all of us.