European Weather Forecasting Center (ECMWF) has just released a form of prognosis, which the center says is more effective than a modern physics form.
This model has been called an artificial intelligence forecasting system (AIFS) from the launch of the new ECMWF, working at a speed that is faster than physics and using less energy about 1,000 times in the forecast.
ECMWF, which is currently in the 50th year of production, etc., which is one of the world's leading weather predictions. Medium prophecy Including the weather forecast between three days and 15 days in advance, but ECMWF also predicted the weather for a year The weather forecast model is necessary for the state and local government to prepare for special weather events – including daily demand, such as knowing what the weather will be on your upcoming holidays.
Traditional weather prediction model makes predictions by solving the physics equations. The limitations of these models are the estimation of dynamics in the atmosphere. The aspect of the understanding of the model that is driven by AI is that they can learn the complex relationships and change in the form of the weather directly from the data instead of adhering to the previously known equations and documents.
The ECMWF announcement comes from the point of Google Deepmind's. GENCAST model For the prediction of the weather driven by AI, the next repeat of the Google Weather Prophecy Software, which includes Neuralgcm and graph– Gencast is more effective. etc.ECMWF leading weather predictions in 97.2% or different air variable targets With more than 36 hours, Gencast is more accurate, etc. in 99.8% or goals.
But the European center is also creative. The launch of AIFS-SINGLE is the first model of the system.
“This is a great effort to ensure that the model is stable and reliable,” Florian Pappenberger, the director of the forecast and service of ECMWF, said in the opening of the center. “At the moment, the resolution of AIFS is less than our model (IFS), which has received a 9 km resolution (5.6 miles) using physical methods.”
“We see AIFS and IFS as a supplement and part of providing a variety of products to our user community, who decided that what is most suitable for their needs,” Pappenberger added.
The team will explore the creation of models that are driven by hybrid and physics to improve the capacity of the organization to predict the weather with precision.
“The model that uses physics is the key to the current data checking process.” Matthew Chantry, a strategic leader for the learning of the ECMWF and the head of the innovation platform in the email to GizModo “. The process of sucking the same data is important to Start the learning model every day and allow them to predict. “
“One of the next borders for the weather forecast, the learning of the machine is the process of checking this data, which means that the full weather chain may depend on the learning of the machine,” Chantry said. supplement
Chantry is a co -author of the study waiting by friends who describe the information system that is driven by information that does not depend on physics analysis.
Called the graphdop. The system uses observed quantities such as the temperature, brightness from Orbiters “to create a corresponding performance of the global change and the physical process.” The writing team “and can create skills with skills. Five days in the future “
Integration of artificial intelligence methods with the creation of a weather prediction model that is driven by physics is a place that is more likely for accurate predictions. Testing to the present indicates that the AI predictions that are driven by can be done better than historical models. But until now, those models have relying on repeated analysis information Observing the ground is necessary for model training and still have to see the ability to predict the impressive technology when it is forced to leave the script.