First Protein foldingNow it’s the weather forecast: DeepMind, a London-based artificial intelligence company, is continuing to apply deep learning to hard science problems. DeepMind has developed a deep learning tool called DGMR in cooperation with the National Weather Service of the United Kingdom, which can accurately predict the probability of rain in the next 90 minutes-this is one of the most difficult challenges in weather forecasting.
When making a blind comparison with existing tools, dozens of experts believe that DGMR’s prediction is the best in terms of a series of factors (including its prediction of the location, range, movement and intensity of rainfall), accounting for 89% of the time.turn out Publish In nature papers Nowadays.
DeepMind’s new tools No AlphaFold, solved a key problem in biology Scientists have been working hard to solve this problem for decades. However, even small improvements in forecasting are important.
Forecasting rain, especially heavy rain, is critical to many industries, from outdoor activities to aviation to emergency services. But doing it well is difficult. Determining how much water is in the sky and when and where it will fall depends on many weather processes, such as temperature changes, cloud formation, and wind. All these factors are complex enough by themselves, but if they are put together, they are even more complicated.
The best available prediction techniques use a large number of computer simulations of atmospheric physics. These are suitable for long-term forecasting, but are not very good at predicting what will happen in the next hour or so, namely nowcasting. Previous deep learning techniques have been developed, but these techniques usually do well in one thing, such as predicting location, but at the expense of other things, such as prediction strength.
“Precipitation nowcasting is still a huge challenge for meteorologists,” said Greg Carbin, director of forecast operations at the NOAA Weather Forecast Center in the United States, who was not involved in the work.
The DeepMind team trained their AI on radar data. Many countries/regions often publish all-day snapshots of radar measurements to track cloud formation and movement. For example, in the United Kingdom, a new reading is issued every five minutes. Putting these snapshots together provides an up-to-date stop-motion video showing how rainfall patterns move across the country, similar to the predictive visuals you see on TV.