When Tropical Storm Melissa hovered south of Haiti, Philippe Papin at the U.S. National Hurricane Centre had a strong feeling that it was about to strengthen rapidly.
As the lead forecaster at the time, he issued a bold early alert. Within a day, Melissa would shoot up to a Category 4 storm. Of course, it turned toward Jamaica.
It was one of the agency’s clearest and most decisive rapid-intensification warnings in years. A big part of that confidence came from a brand-new tool. Google’s DeepMind hurricane-forecasting AI.
This AI had only recently gone into real-world use. It was showing a sharp, sudden jump in the storm’s strength. Papin trusted what he saw. As we know, Melissa ended up proving the AI right. It moved through Jamaica with a huge force.
Setting a New Bar for Forecast Accuracy
DeepMind’s model is the first AI system designed specifically for hurricane prediction. It is already outperforming the traditional numerical models that meteorologists have used for decades.
Across all 13 Atlantic storms this year, it consistently delivered the best results. In many cases, its storm – track predictions even beat those of forecasters.
Melissa eventually hit land as a Category 5 hurricane. It was one of the strongest storms Jamaica has seen in about 200 years. Papin’s early call gave people on the island precious time to prepare. This likely lowered both the damage and the number of injuries.
How This AI Actually Works
Google has spent years building machine – learning weather tools. This model is based on DeepMind’s larger forecasting system. The AI does not go through a long list of equations like traditional models.
It simply studies a huge amount of past storm data. It then picks up patterns that these older systems might overlook. One of its biggest advantages is speed.
Former NHC forecaster Michael Lowry points out that the AI can deliver forecasts in minutes. It runs on regular PCs and does not need pricey supercomputers. After just one season of use, he says it has already shown it can match, and sometimes beat, the long-trusted physics-based models.
Still Some Gaps
Despite this feat, the DeepMind system isn’t perfect. Retired NHC forecaster James Franklin notes that the AI has struggled with a few extreme cases. It struggled with the sudden intensity spikes during Hurricane Erin and Typhoon Kalmaegi.
Even so, he thinks the model’s success is real, not a fluke. He hopes to work with Google in the off-season. This is to help make the tool even more practical for forecasters.
One of the big issues is improving transparency. The model still operates like a “black box,” with limited insight into how it makes decisions.
A Shift in the Weather-Forecasting World
It’s unusual for a private company to lead this kind of work. Most top weather models are government-built. They are also mostly open to researchers.
But Google does publish DeepMind’s forecast outputs. It is its private property. Thus, the details behind the system remain mostly under wraps.
Meanwhile, U.S. and European weather agencies are racing to develop their model. These will be AI-based models just like DeepMind. Early results suggest they are making some progress.
Source from Gizchina
Disclaimer: The information set forth above is provided by gizchina.com independently of Alibaba.com. Alibaba.com makes no representation and warranties as to the quality and reliability of the seller and products. Alibaba.com expressly disclaims any liability for breaches pertaining to the copyright of content.
