Ghost hurricanes and weather models

In recent decades, there has been a lot of progress with weather forecasting based on numerical predictions. However, when forecasting tropical storms such as hurricanes, numerical forecast models will occasionally predict the formation of tropical systems that don’t evolve. These are called “ghost hurricanes” or “ghost storms.”

There are many forecast models that are used for weather prediction on a global and regional scale. The most widely used global models are the Global Forecast System (GFS), which is an American product, and the European Center for Medium-Range Weather Forecasts (ECMWF). Both of these forecast models, which are available online, will provide outlooks at least two weeks in advance. There are other computer models that are specific to more regional and specific geographic areas and will have shorter forecast timeframes.

According to an article on CNN, when attempting to forecast more than a week in advance, all forecast models will tend to overpredict or underpredict potential developing storms, especially ones in the tropical regions. The American GFS model is notorious for overprediction of tropical activity, as the model forecast the development of the season’s first named storm earlier this month. However, the storm failed to materialize and was dubbed a “ghost hurricane.”

Many of the models can display or generate different scenarios or forecasts for data that is similar. Despite the fact that some computer models, such as the GFS, will overpredict tropical storm development, programmers designed them to do just that. By contrast, other models are designed not to be as sensitive to potential tropical development and will not display the storm system. In other words, every computer forecast model is designed to react differently from the others. To be fair, most numerical forecasts will struggle with long-term forecasts of tropical storms, especially early in the season.

Tropical storm forecasting is not an exact science. Very small changes in the atmospheric circulation can alter the path of a tropical storm or hurricane, which can make for difficult forecasting of these systems.

As forecasters depend heavily on the technological advancement of numerical models, according to NOAA, a five-day forecast has an accuracy rating of approximately 90%. For a 7-day outlook, the accurate rate is about 80%. Many weather apps and websites have expanded their outlooks to 14 days, or even higher in some cases. Beyond 10 days, the forecasts are about 50% accurate, but certainly better than they used to be.

Artificial Intelligence for weather prediction is making strides to be incorporated into short and long-range outlooks. It was recently announced that Google is working with the National Hurricane Center in Miami, Fla., to test a new model that is AI-driven to better predict tropical storms and hurricanes. Forecasters claim that the new AI model can forecast the development of tropical storms up to 15 days in advance.

Initial testing of the Google AI model has been promising. It uses years of weather data and observations and provides numerous scenarios of the development and track of the storm. If one would like to see this model in action, it is available at https://deepmind.google.com/science/weatherlab.

Although AI models are much faster and considered to be more accurate in many weather scenarios, they have missed some major events. For example, according Physics.org, new research is indicating that these models are not able to forecast “beyond the scope of existing training data.” This may include the omission of major floods, heatwaves or other large events that will occur about every several hundred to a thousand years. However, much progress is being made with AI models as more data is fed into them and training the models to look for new patterns.

There is another model that is available to the public is called www.ventusky.com. This one is very interesting and displays past weather, current and future based on numerous weather models. I have used this one at times and will show current air circulation, rainfall forecasts, satellite data, winds, pressure, snow cover and air quality across the world. A similar model is windy.com. Both are very colorful and user-friendly.

In terms of our local weather, the first half of June was the driest in history in Coeur d’Alene and Spokane. At Cliff’s station and at the Spokane International Airport, no moisture has been measured so far this month. The normal June rainfall for Coeur d’Alene is 1.93 inches.

Speaking of computer models, there are indications that we could see some moisture late this week in the lower elevations, but amounts are not expected to be very impressive. Most of the moisture across the country has been over the eastern sections of the country, while much of the West has continued to be drier than normal.

According to the latest U.S. Drought Monitor, there is extreme to exceptional drought across the southwestern portion of the country extending into western Texas. The Spokane and Coeur d’Alene regions are currently experiencing abnormally dry to moderate drought conditions. With very little rain expected over at least the next several weeks, the region of drought will likely expand.

• • • 

Contact Randy Mann at [email protected].