Irma – an interesting case study in forecasting
As of this morning, Irma has weakened to a tropical storm, but don’t let the downgrade fool you. She’s still packs a punch with sustained winds up to 65 mph and heavy rains. Her tropical-storm-force winds extend outward up to 415 miles from her center, so her reach is still huge. For that reason, tropical storm warnings extend up both sides of the Florida peninsula and along the coasts of Georgia and South Carolina. Irma will be a tropical depression by tomorrow morning, and nothing more than remnants by the end of the week.
Now that the worst of the storm has subsided, I thought it would be interesting to look back at some of the forecast updates from last week. This post will be a bit graphics-heavy, but for those with short memories like mine, the graphics will be helpful.
On Tuesday, September 5, the 11 a.m. advisory graphic showed Irma passing over the northern Leeward Islands, Puerto Rico, the northern coast of Hispanolia, and possibly most of Cuba before reaching the Florida Keys. The forecast cone for 8 a.m. Sunday was actually centered on the Keys as Irma started her expected turn toward the north.
On Thursday, September 7, the 2 p.m. update had shifted the track slightly eastward so that by Sunday morning, Irma was making landfall near Miami. Some feared the worst as Homestead, Florida, which took the brunt of Hurricane Andrew, appeared to have become the target for Florida landfall. By that time, Irma had pasted Puerto Rico and was heading toward the Bahamas. The norther coast of Cuba was under a Tropical Storm Warning.
The 5-day track showed Irma still being a tropical storm by 8 a.m. Tuesday (tomorrow) morning, passing over the western part of the North Carolina/South Carolina border and bringing high winds, heavy rain, and possibly isolated tornadoes to much of NC.
On Friday, September 8, the 8 a.m. forecast update, showed Irma affecting the Bahamas, northern Cuba, and central Southern Florida as a major storm. Truly, the tip of southern Florida was where the center of the cone was located. Irma was expected to take her time moving up the center of the Peninsula, into Georgia as a tropical storm, and then up to central Tennessee as a tropical depression by Wednesday morning.
Jump forward to today, and we see Irma is a strong tropical storm moving through the panhandle of Florida, about to cross into Southern Georgia, and eventually through north-central Alabama, northeastern Mississippi, and into western Tennessee.
If you look at this morning’s wind history map and compare it to the above maps, one fact will likely jump out at you. That forecast track from last Tuesday seems to have been the closest to her actual track and landfall overall. It’s almost uncanny!
Before you rush to judgement, though, and say the National Hurricane Center should have stuck with that forecast all along, let’s think about some of the many things they consider when making these forecasts:
- Multiple numerical weather prediction models with multiple variations on many of them. Some are run hourly, and others run every 12 hours. Each has its own strengths and weaknesses, and each accounts for some variables in the atmosphere, but not others. It’s the forecasters’ jobs to know which tends to be the best for which type of weather.
- Data being added by the Air Force Hurricane Hunters’ flights through and around the storm. Their real-time information help NHC forecasters determine the actual pressure, winds, and steering currents of the storm (among other things). The information they gather is invaluable to forecasters.
- The way hurricanes interact with land and the ocean are very different. Both play a role in the strength of the storm. Land tends to weaken storms while warm ocean waters strengthen them.
- Other low and high-pressure systems in the general region (North America and the northern Atlantic) greatly influence the track of the storm. The models showed a turn toward the north, but the timing of that turn greatly depended upon the timing of the movement of a high-pressure system that was blocking that turn. Some model runs rushed it, and others slowed it down.
- All the things that we know we don’t know. The National Hurricane Center archives and eventually verifies its forecasts. In fact, a page of their website explains their methods for verification and shows their accuracy for track and strength for the years 2012-2016. Verification is part of the scientific method. It helps forecasters track their progress in improving their own forecasts. After all, you can’t get better at something if you don’t have a grasp of what your weaknesses and strengths are. One fact you’ll note from the NHC Official Track Error Cumulative Distribution graph on that page is that the NHC is well aware of the size in nautical miles of their forecast track error 3-5 days out. They even remind readers of the error size in their forecast advisories.
Just thinking about that fifth point, if you know your forecast track can be off by 400 miles or more for the hurricane’s possible location 5 days out, why would you stick with that forecast when it changes the next day in most of the models? In other words, if you know yesterday’s long-range forecast had a high probability for being wrong, wouldn’t you adjust it for what you’re seeing today?
Hindsight can be 20/20 as they say, but it’s unfair to Monday morning armchair quarterback a forecast track for a storm as huge as Irma. Rest assured that meteorologists are probably already looking at the above information and then some, trying to understand whether last Tuesday morning’s forecast was dead-on because it was good or if it was just an interesting coincidence.