DeepMind’s AI predicts when and where it will rain almost accurately

Texas News Today

First Protein Folding, Now Weather Forecast: London-based AI company DeepMind continues to apply deep learning to hard science issues. DeepMind has worked with the UK’s national weather service, the Met Office, to develop a deep learning tool called DGMR that can accurately predict the chance of rain over the next 90 minutes. This is one of the most difficult challenges in weather forecasting.

Compared with existing instruments, dozens of experts say the DGMR’s prediction is the best with an 89% chance of predicting the location, extent, movement, strength and many other factors of rain. I decided. The results were published today in the Nature paper.

DeepMind’s new tool No Alphafold has uncovered an important biology problem that scientists have been grappling with for decades. But even a slight improvement in forecasts can be a problem.

Predicting rain, especially heavy rains, is very important for many industries, from outdoor events to aviation and emergency services. But it is difficult to do it well. Knowing how much water is in the sky and when and where it falls depends on various meteorological processes such as temperature changes, cloud formation and wind. All these elements are complex in themselves, but in essence they are even more complex.

The best current prediction method uses a large computer simulation of atmospheric physics. These are good for long-term predictions, but not very good at predicting what will happen in the next hour or so, which is called nowcasting. Deep learning techniques have been developed before, but they usually work well for one thing, such as location prediction, at the expense of something else, such as intensity prediction.

Actual radar data for heavy rainfall over the eastern United States in April 2019 and a comparison of the DGMR with two rival forecasting methods


“The current rainfall now remains a major challenge for meteorologists,” said Greg Carbin, chief of forecast operations at the NOAA Weather Prediction Center in the United States, who was not involved in the task.

The DeepMind team trained the AI ​​with radar data. Many countries often publish snapshots throughout the day of radar measurements that track cloud formation and movement. For example, in the UK, new readings are issued every 5 minutes. Putting these snapshots together provides an up-to-date stop-motion video showing how rain patterns are moving across the country, as seen in weather forecast sequences on television. .

DeepMind’s AI predicts when and where it will rain almost accurately

Source link DeepMind’s AI predicts when and where it will rain almost accurately


Please enter your comment!
Please enter your name here