Before delving deeper into the topic, let’s clarify some terminology: cyclones, hurricanes, and typhoons—what’s the distinction?
In essence, all these terms refer to the same phenomenon! The term "cyclone" is the global scientific classification used for what are known as hurricanes in the Atlantic and typhoons in the North Pacific. These terms describe a large, rotating storm system that develops over warm ocean waters. The heat from the ocean causes warm, moist air to ascend, creating a low-pressure zone that invites cooler air to take its place. Earth's rotation facilitates this cycle of rising and cooling air, and when you combine this dynamic with heavy rainfall and powerful winds, you have a cyclone, hurricane, or typhoon, depending on your geographical location.
Traditionally, how have scientists predicted cyclones?
In the past, cyclone forecasting relied heavily on advanced physics-based models run on supercomputers that could be as large as shipping containers or even three-story buildings. Researchers utilized these enormous machines to simulate the principles of fluid dynamics, thereby forecasting future weather by analyzing current meteorological data from around the world. However, the information from global sensors is frequently incomplete or noisy, and accurately simulating physical phenomena demands significant computing resources, which hampers the forecasting process. Historically, advancements in meteorology led to only about a day's increase in forecast accuracy every decade, achieved by creating improved models, deploying more satellites, and enhancing computational capabilities. By integrating AI with data derived from these physics-based models, researchers have experienced a decade’s worth of progress in a single jump with WeatherNext models.
How is AI transforming cyclone prediction?
While physics models excel at analyzing past and present conditions, AI enhances our ability to forecast future events by identifying patterns from historical data and current observations. WeatherNext harnessed 50 years of weather history to train AI algorithms. This innovation allows researchers to run these swift models on a single Tensor Processing Unit (TPU), as opposed to relying on enormous supercomputers. This results in significantly improved accuracy in predicting both the path and intensity of cyclones prior to landfall. Weather agencies are now aiming to extend their public forecasts from five to seven days, providing communities with essential extra time to prepare for impending cyclones.
What impact can this additional time have on people in vulnerable areas?
The importance of this extra time was clearly demonstrated last year during Hurricane Melissa. Prior to the hurricane season, we collaborated with the National Hurricane Center to enhance their forecasting capabilities. Notably, although Melissa initially seemed weak, the AI-driven predictions provided high confidence and enabled the National Hurricane Center to issue a groundbreaking warning about the storm's rapid escalation from a Category 1 to a Category 5 hurricane. This storm turned out to be one of the most powerful hurricanes recorded, and our models detected this intensification days before it made landfall. The additional day of advance notice proved invaluable, allowing authorities in Jamaica to implement crucial life-saving emergency measures.
