To achieve this, scientists rely on high-resolution radar data, which measure the humidity of the air by sending multiple light rays into the lower atmosphere. This technique measures the relative velocity of the signal and determines the rate at which the water vapor slows down.
Using the last 20 minutes of high-resolution radar data, researchers at Google Deep Mind have developed a machine learning approach called Generative Modeling to develop a device that can predict average to heavy rainfall over the next 90 minutes.
“AI has the potential to answer some of the most complex scientific questions in ecology, such as climate change,” said Shakir Mohammed, one of the scientists involved in the study.
Another method used in the medium and long term
So far, meteorologists have used powerful numerical weather forecasting systems to make medium- and long-term forecasts based on equations that describe the movement of fluids in the atmosphere.
A method that has been very effective for a long time, but is less likely to measure time in the near future in a few hours, the Google Deep Mind study proves this Tuesday, in the science journal Nature.
“These models are really amazing in terms of weather forecast from six hours to two weeks, but there is one area, especially from zero to two hours, in which the models work particularly poorly,” said Suman Raouri, researcher project leader, in a commentary broadcast by The Guardian.