Introduction to Ensembles: Forecasting Hurricane Sandy

Today, when I arrived on the watch floor, I received an OTSR (optimal track ship routing) request for service from the skipper of the USNS Apache, one of Military Sealift Command's four Fleet Ocean Tugs. My name is AG2 Shepherd and I'm an apprentice ship router with the U.S. Navy.

The skipper is tasked with towing a barge from the shipyard in Pascagoula, Mississippi, to one in Bath, Maine. He wants to leave 23 October. The transit is about 2500 miles long. At 10 knots, it will take about 10 days to complete.

The barge will be loaded with the superstructure for a new guided missile destroyer.

I have an apprentice forecaster who entered the movement report request, or MOVREP, into the Joint Metoc Viewer.

The limits are rather conservative, but the superstructure is 60 feet tall, offering ample sail area for any strong wind. And while the tug can take a beating, the barge isn't made to take large seas from any direction. This will be a long, slow transit.

Current conditions are absolutely beautiful. Clear skies across the Gulf of Mexico and up the East Coast. If they could be there tomorrow, I'd send them on their way, but it's 2500 miles from Pascagoula to Bath. With a 10 day transit, that's where I come in. As a ship router, I'm expected to make medium to long range forecasts, 3 to 5 days or more, and keep ships out of harm's way.

A quick look at the 500 mb heights shows a low centered over the Great Lakes with a trough extending south to the Gulf. The trough is progressive and should move out quickly to be replaced by a ridge. Looks like nice weather coming.

A broad trough looks to be approaching the West Coast on day 4; something to keep an eye on for now.

A look at the 250 mb heights shows a trough and associated jet leaving the Southeast with another coming ashore on the West Coast on Day 4.

MSLP charts show the surface low currently over the Great Lakes lifting out to the northeast over the next couple of days. At the same time, high pressure builds over the Southeast.

Looking further down the line, the disturbance at 500 mb is accompanied by a surface cyclone emerging onto the Great Plains on day 5. So far it looks like it will stay to our north, but we'll have to keep an eye out to see how this develops. If we need to, we can delay the transit in Miami.

I get on the phone to the skipper. He asks me, "What are the odds that your model is wrong and I get slammed out there in the Gulf? Last time I scheduled a transit, I had to delay at the last second because the forecaster chose the wrong model to trust."

What are the odds? How sure are you? How likely is that? We hear these questions all the time. With those questions ringing in my ears, I start digging deeper.

First, I check other models.

I have access to several models, but tend to limit my views to models provided by the Navy's FNMOC (NOGAPS and COAMPS) and NOAA's NCEP (GFS and the NAM). If they all agree, then I figure that the odds are pretty good that they're right, at least for the short term. The trouble is, the models don't always agree, and they inevitably start to diverge after a couple of days.

Our senior ship router has been after me to start using model ensembles in my forecast. He says that they'll help me answer questions like, "What are the odds?" I go back to my work station and start looking for ensembles.

The first stage of the transit is from Pascagoula to Miami: about 900 miles, or about 3 days. If the tug departs the morning of the 23nd, they should arrive on the 26th. At my work station I find the NOGAPS ensemble, along with a couple of others. NOGAPS is familiar, so I choose it. Looking at winds and seas, I select the "Mean and Spread." That ought to tell me if there's anything wrong with my forecast.

Lesson 1: What are Ensembles?

To learn more see the lesson on Ensemble Statistics.

A similar examination of the ensemble mean wave heights shows that waves should stay well below the 8-foot limits for the transit.

So far, so good. I issue the OTSR Route Recommendation and move on to other shift work.

The barge is loaded and ready for the transit. With less than 24 hours to departure, it's time to check the forecast and update the ship routing as needed.

The satellite images show it's still clear over the Southeast.

The NOAA Weather Map show high pressure over the Southeast, with clear skies and light winds.

We all know that with long lead times, Sandy could go almost anywhere. I decide to look at the ensembles to check out the range of forecast possibilities.

First, I jump to the chase and look at the cyclone tracks for the NOAA Global Ensemble Forecast System, or GEFS.

Most of the tracks rapidly recurve offshore. That's good.

I examine the probability that the significant wave height will exceed 8 feet and see that it surpasses 50% probability for small parts of the Gulf over the forecast period. Unfortunately, those small regions lie close to the transit route.

To learn more, see the Lesson on Bias Correction.

Looking at the spread in the hurricane tracks over the next week, it struck me that the best forecast I could make for the longer periods, like a week or more, is probably the ensemble mean. And if that's true, why not just use the ensemble to make the forecast?

My Senior Ship Router must have wondered the same thing, because in a follow-up call she mentioned that the Canadians actually have deployed that very thing for their new day 6-7 forecasts. What will they think of next?!?!?

To learn more about what the Canadian Meteorological Center is doing to automate its day 6 and 7 forecasts, see the Lesson on CMC Automated Forecasting.

As I watch the satellite loop of the storm growing by the hour, it's clear by now that the transit isn't leaving port for several days. I call the skipper to let him know.

The media attention that Sandy is now receiving is staggering, despite the fact that the storm is only now over the Bahamas. At least the tug captain has stopped calling. There's no point in leaving port until he knows there will be a port to pull into at the other end.

There have been significant developments over the past 24 hours.

Just 36 hours ago, the ensembles were having a difficult time settling on a consensus track. Some tracks recurved eastward, out to sea. Others made a sharp turn to the west with landfall on the East Coast. Unfortunately, landfall ranged from Chesapeake Bay to Nova Scotia. The most recent tracks have been bringing the storm ashore in the New York/New Jersey, but still with plenty of uncertainty. How do we make sense of that?

To learn more about Multimodel ensembles, see the Lesson on Multimodel Ensembles.

To learn more, see the Lesson Interpretation of Mean/Spread Diagrams.

I compare the mean/spread plot to a spaghetti plot of the 5520m contour for the same time on October 30. It appears that the greatest uncertainty arises from widely variable solutions for the trough over the Atlantic Seaboard. In contrast, the contours that define the high over NE Canada are much more tightly clustered.

Not wanting to tie my forecast to any one ensemble system, I try looking at a different multimodel ensemble, the North American Ensemble Forecast System, or NAEFS. This ensemble combines the NCEP and Canadian global ensembles. Just like the other models, NAEFS shows that Sandy will grow to a very large storm with storm-force winds over a very broad region. At its peak, the mean wind field brings winds of at least 30 kt from South of Cape Hatteras nearly up to Cape Cod.

With the barge transit delayed until further notice, our concern now turns to resource protection. The threshold that the Navy is concerned about is a 50 kt tropical storm force wind speed. While the ensemble mean never gets that strong, that's not to say it won't happen. So again, it's a question of probability.

Interestingly, through the entire forecast period up to landfall, NAEFS never has the probability of 50 kt winds greater than 10% along the coast. I chalk that up to the spread in the hurricane tracks.

Between the wind and the waves, the Eastern Seaboard will take a battering for the next few days. With 4 or 5 days to landfall, it's time to pass this up the chain of command so vulnerable sites can start preparations.

Sandy is currently just offshore from Cape Hatteras. Warnings are up along the East Coast.

A stationary front runs up the Eastern Seaboard. When Sandy veers westward toward landfall, she'll merge with that front. That's when Sandy will go extratropical.

My current task is to provide a forecast for the Naval Station Earle at Sandy Hook, NJ, on the south side of New York Harbor. Its one distinguishing feature is a 3-mile pier in Sandy Hook Bay that looks like Neptune's trident. This keeps ammunition being loaded and unloaded from ships at a safe distance from the surrounding area.

Landfall is expected somewhere in New Jersey with a high probability of strong winds over a broad swath of coastline.

All the models have come into agreement with closely clustered tracks among ensembles and ensemble members.

The National Hurricane Center discussion emphasizes the extratropical transition of Sandy as it joins a trough in the next 48 hours.

The spaghetti plot of the 5520m contour of the 500 mb heights clearly shows how the longwave trough absorbs the tropical cyclone. There is remarkable agreement among the ensemble members pre- and post-merger.

On the East Coast, during hurricane season, all vulnerable facilities assume a Hurricane/Tropical Cyclone Condition of Readiness (TCCOR) where they can be prepared for a storm with winds greater than 50 kt within 96 hours. This is TCCOR-V. Yesterday, the Admiral issued a TCCOR-III for the Mid-Atlantic Region. Now it's time for the daily update. In addition to wind, we're also concerned with waves and storm surge.

The NUOPC wave ensemble run at FNMOC consist of Wavewatch III model forced by NUOPC winds. Here we see significant wave height predicted by this wave ensemble around the time of Sandy landfall. As we can see from this image loop, the mean wave height close to shore is predicted to exceed 24 feet for an extended period of time, with a maximum height of over 32 feet located offshore.

For storm surge, I look at products produced by the National Hurricane Center. For Earle, our threshold of concern is 8 feet. At this time, the probability of the surge exceeding 8 feet at Earle is between 20 and 30%.

I find the median forecast by looking at the 50% exceedance height. It shows that there is a 50% probability that the surge will exceed 5 feet.

A more conservative approach is to look at the 10% exceedance height. That is the height that is predicted for the top 10% of ensemble members. In this case, there is a 10% chance that the storm surge will exceed 11 feet in New York Harbor, including at Earle.

So, my forecast for Earle consists of the following:

• Winds rise to 35 kt or more 6z 29 Oct and persist through 6z 30 Oct for the area near New York Harbor. There is a significant chance of winds 50kt winds through that period.
• Seas near shore rise to 24 feet for the same period.
• Storm surge is predicted to be about 5 feet near the mouth of New York Harbor, but there is a 20-30% chance that it will exceed 8 feet, and a 10% chance of exceeding 11 feet.

Sandy made landfall about 00z on October 30th with high winds and waves and heavy rains.

Hurricane force winds and gusts struck a large region before the storm brought heavy precipitation to inland areas.

To learn more and answer the following questions, see the lesson Ensemble Forecast Verification.