Supercomputers are driving a revolution in hurricane forecasting

Back in 1998, the European Center for Medium-Range Weather Forecasts housed the 27th most-powerful supercomputer in the world, with 116 cores providing a maximum performance of 213 gigaflops. Today, the ECMWF forecasting center has the world’s 27th 28th most powerful supercomputers, each with 126,000 cores and 20,000 times the computing power of its machine two decades ago.

This dramatic increase in computing power at the European center—as well as similar increases at US-based and other international numerical modeling centers—helps to explain the dramatic increase in hurricane-forecast accuracy over the same time period.

Based upon new data from the National Hurricane Center for hurricanes based in the Atlantic basin, the average track error for a five-day forecast fell to 155 nautical miles in 2017. That is, the location predicted by the hurricane center for a given storm was just 155 nautical miles away from the actual position of the storm five days later. What is incredible about this is that, back in 1998, this was the average error for a two-day track forecast.

In fact, the annual “verification” report released Wednesday shows that for the hyperactive 2017 Atlantic hurricane season—which included the devastating hurricanes Harvey, Irma, and Maria—the National Hurricane Center set records for track forecasts at all time periods: 12-hour, 24-hour, and two-, three-, four- and five-day forecasts.

Super Euro

This achievement represents both a testament to the forecasters at the Miami-based National Hurricane Center, which provides official track and intensity forecasts for Atlantic and Eastern Pacific hurricanes, and also to the increasing power of sophisticated computer models, the precision of which has driven overall forecast accuracy scores higher.

When they attempt to determine where a hurricane will go, human forecasters generally consider three different kinds of models. There are models developed specifically for hurricanes, such as the US-based HWRF. There are “global” models, such as the ECMWF’s European model, and the US-based Global Forecasting System model. Finally, there are “consensus” models that combine the input of several models to make an average track of sorts.

Readers may recall the superlative performance of the European model during Hurricane Harvey, and the new report helpfully includes an analysis of the best computer models for 2017. Overall, the European model fared by far the best among the individual “global” models but below most of the consensus models for the entirety of the 2017 Atlantic hurricane season.

However, over a longer period, from 2015 to 2017, the European model showed its class. During the last three years, the European has actually outperformed the official track forecast from the National Hurricane Center at three-, four-, and five-day forecasts. Only a weighted consensus model, HCCA, which relies most heavily on the European model for its track forecast, was more accurate than the European model itself.

Hurricane Mitch

Forecasts matter. A good example of this can be found back in 1998, when four- and five-day forecasts were so experimental that the National Hurricane Center didn’t even make them publicly available. In August of that year, Category 5 Hurricane Mitch moved westward across the Caribbean Sea, threatening Central America.

In the days leading up to landfall, the National Hurricane Center’s track forecast called for a slow, mostly northwestward motion that would bring the hurricane toward the Yucatan Peninsula. (This reflected the computer modeling available at the time). Instead, Mitch moved westward and then southwestward, making a landfall in Honduras and Nicaragua.

After its surprise landfall, Mitch would cause widespread devastation. It became the Atlantic’s deadliest hurricane in more than two centuries, causing an estimated 11,000 fatalities—eclipsing the Galveston hurricane of 1900. Today, residents in Honduras and Nicaragua would probably be better warned.

[ufc-fb-comments url=""]

Latest Articles

Related Articles