With the arrival of artificial intelligence (AI) in weather forecasting, what once seemed insurmountable – predicting the path and intensity of hurricanes – is currently undergoing a huge transformation. Advanced AI models herald the paradigm change in forecastingoffering unprecedented accuracy and potential insight into one of the most difficult phenomena of nature: Atlantic hurricanes.
The journey towards the weather forecasting based on AI began with a key time-accident meeting between two young innovators at Stanford University. John Dean, electrician engineer and Kai Marshland, an IT specialist, began a mission to solve the eternal problem of weather uncertainty. Their ideas, Windborne Systems, introduced a new solution: miniature weather balloons, capable of moving the atmosphere of the Earth for up to 40 days. Using these balloons to collect atmospheric data, Dean and Magshand tried to revolutionize global weather forecasts, exceeding the restrictions on traditional observation methods.
Earlier, the European Center for Medium -range Weather Forecasts (ECMWF) has maintained the data treasurer called Era5. This set of data contains atmospheric, land and ocean data for each day, in points around the world, reaching 1940. The last 50 years of data, especially after the arrival of the global satellite range, is particularly rich. Although Era5 was not initially created for the AI application, it turned out to be extremely useful for training weather forecasting models.
Computers began to seriously use ERA5 data to train AI models to forecast weather in 2022. Since then, technology has quickly made progress. In some cases, the result of these AI models exceeds global weather models that scientists spent decades on design and building. Here is Kicker: These AI models can work on one desktop computer, unlike their supercomputer counterparts.
The transition from theoretical innovation to the actual application was fast. The Weathermesh Windborne, powered by AI algorithms, showed extraordinary efficiency in the forecasting of hurricane, exceeding traditional models based on physics in the accuracy of track forecasting. Similarly, ECMWF adopted the AI revolution by revealing artificial intelligence/integrated forecasting system (AIFS). Thanks to its ability to create “very promising” forecasts in real time, AIFS is a significant leap forward in the possibilities of meteorological anticipation.
In the center of this meteorological miracle lies a fundamental change of approach. Unlike traditional forecasting models, which are based on complex physical equations to simulate atmospheric dynamics, models based on AI, such as Weathermesh and AIFS, work on a different principle. Analyzing huge sets of data, including ERA5, these models notice complex patterns and correlations, enabling them to predict weather phenomena with unprecedented accuracy.
The earth shutter and its weather conditions are shown for the model: values such as temperature, pressure, humidity, winds and much more at different levels of atmosphere. Then it was shown what conditions were six hours later around the earth. The model then “learns” this relationship between the weather and the conditions a few hours later. The process is repeated many times. By consuming all these data, the model is becoming better in recognizing patterns and establishing connections regarding the conditions, say, the low pressure zone over the Northern Atlantic Ocean and what this means for the weather below Europe and Asia in the coming week to 10 days.
One of the most striking advantages of AI models is their performance. Traditional models based on physics require huge computing power, often based on supercomputers. In contrast, AI models, such as Weathermesh, can operate on one high -power stationary computer equipped with a graphic processor. This dramatic reduction of computing resources makes AI models more accessible and scalable.
While the meteorological community impatiently includes the promise of forecasting based on artificial intelligence, a lot of challenges. Ensuring reliability, service of extreme events and solving problems are critical areas of research. The integration of AI models with operational forecasts is logistic and methodological obstacles that require strict assessment and validation. In addition, the coexistence of AI and physics models raises questions about their complementary roles in shaping future forecasting methodologies.
