Why Long-Range Model Forecasts for the Tropics Can't Often Be Trusted: A Case Study

Weather forecast model data is now widely available over the internet from both government agencies and the private sector.

That doesn't mean it's a good idea to post a model forecast of a landfalling hurricane a week or more from now to your social media account. One current example is a perfect case study.

One Model: A Large Difference in Just Three Days

Forecast models – the complex algorithms run on government supercomputers multiple times daily – account for variables, including temperature, pressure, wind and precipitation.

The animation below depicts eight consecutive iterations, or model runs, of the Global Forecast System (GFS) model run every 6 hours from the morning of Aug. 22 through the morning of Aug. 24. Each forecast is for the same time, 2 a.m. EDT on Aug. 31.

A black circle surrounding deep reds highlights where the model forecasted a more developed tropical cyclone. You can see what should be Hurricane Gaston in the top right of some frames. 

Notice the sometimes massive difference in placement of the tropical system with each model run.

Here is a recap of the eight GFS model run forecasts shown above. Again, all these forecasts are for the exact same time, Aug. 31 at 2 a.m. EDT:

- Aug. 22 morning: A weak tropical wave between the Bahamas and Florida moving westward. - Aug. 22 afternoon: A weak tropical wave near Andros Island in the northern Bahamas. - Aug. 22 evening: A tropical wave in the northern Bahamas.- Aug. 23 early morning: Potentially a weak tropical cyclone near Apalachicola, Florida.- Aug. 23 morning: A diffuse tropical wave approaching south Florida.- Aug. 23 afternoon: A tropical cyclone south of Louisiana heading west-northwest.  - Aug. 23 evening: A tropical cyclone approaching the First Coast of Florida north of Orlando.- Aug. 24 early morning: A low pressure system near the Nature Coast of Florida.

See the problem here?

Understanding Meteorology's Limitations

Notice each of those model runs was a forecast for 7-9 days in the future.

While numerical modeling and meteorology have advanced the past few decades, any model forecast beyond 5-7 days should be taken with a large grain of salt. This is particularly true in the tropics, where data to feed the forecast model's initial guess of the current state of the atmosphere is typically more sparse.

Forecasts for tropical cyclones at any stage of their lifetime extend out only to five days because the science hasn't advanced enough to be greatly accurate beyond that time frame, although we are getting there. 

In this particular case, a tropical depression hadn't formed yet in the Antilles. Instead, the general area of interest, known as an invest to meteorologists, featured a weak low pressure system and a large area of thunderstorm clusters.

(MORE: Things You Should Know About Spaghetti Models)

Which one of these clusters would eventually dominate, setting off the cascade of tropical cyclogenesis? Forecast models may not pick the right area, and there's no guarantee it would happen either.

A basic principle of modeling is that initial errors grow with time. How exactly would a model forecast the position, much less intensity, of a potential future tropical cyclone if the initial position is unsure?

The NHC Facebook post also mentioned "the historically poor and erratic performance of guidance models with weak disturbances" to emphasize that the state of the science cannot forecast impacts to specific locations that far in advance.

(MORE: Hurricane Central)

In general, many of the other models are also showing various situations over time without much consistency beyond seven days.

Complicated, huh?

This isn't simply a problem in the tropics during hurricane season. Often details, such as subtle jet stream-level features several thousand miles away, can make the difference between a crippling February Nor'easter and simply a strong, dry Arctic cold front blasting well offshore.

Follow a Trusted Source

Meteorologists have the training to examine and interpret an avalanche of data on the current state of the atmosphere and a number of forecast models, one of which is the GFS model discussed above.

(WATCH: Weather Geeks on Social Media-rology)

Meteorologists look for trends in forecast models and have a working knowledge of which models tend to perform better in certain situations. Ensemble model forecasts, a method of running a single model multiple times with slightly different initial conditions of the atmosphere, is an excellent tool to gauge uncertainty in a forecast.

With years of experience, most meteorologists are well aware of the uncertainty in forecasts, and communicate that uncertainty in a responsible manner to the public or to companies paying for that expertise.

"Detailed model forecasts now go out very far in time, are available electronically and easily so, and can be further distributed just as easily and without sufficient qualification," said Stu Ostro, senior meteorologist with The Weather Channel.

"Viral 10-day hurricane forecasts hurt us. They eat into the credibility of a field pushing the bounds of predictability of the physical world," said Michael Lowry, hurricane specialist with The Weather Channel. "If there's one thing I've learned from the folks that study human behavior, it's that a conflicting message paralyzes public response."

(DEEP DIVE: Dr. Jeff Masters...Which Model Should You Trust?)

If you see a forecast model image via social media, before sharing or retweeting it, notice who originally posted that image. Is it posted without any context or explanation? Is it depicting a forecast beyond five days, particularly of a landfalling hurricane, major winter storm or severe weather outbreak? 

If your answer is "yes" to those last two questions, check first with a trusted source, such as a local TV meteorologist, the National Weather Service, National Hurricane Center, or private forecast entity such as The Weather Channel, weather.com or Weather Underground before sharing that image with others.

In general, a model forecast beyond fve days has a very low chance of verifying, and should be used as entertainment value only. 

(EXPERTS: Ostro | Lowry | Parker | Norcross | Postel | Masters)

Put it this way, if you sustain an injury, would you simply take home your x-ray, hold it to the light and diagnose a torn ligament without consulting a doctor?

In essence, that's the same thing as posting extended weather model images to social media without qualification or context.

"[Mis]perceptions can spread remarkably quickly. We have to acknowledge the reality, and address it responsibly and as effectively as possible," Ostro said. "We live in a new era of communication technology – a 'viral' age – and like it or not, it's not going away."

MORE ON WEATHER.COM: Images and Photos of a Hurricane's Eye