Weather forecasting is something that concerns people every day. However, it is not just the weather of the next few days that meteorologists are trying to find.
In addition to long-term forecasts, weather models often take on the forecast of meteorological conditions within the next few hours, known as "Nowcasting». In the artificial intelligence company supported by Google DeepMind, researchers have now taken an important step forward in the accuracy of Nowcasting forecast.
Short-term weather forecasts are more important than we might think, as they are especially useful for planning big events, transport services, dealing with natural disasters, conserving agriculture and much more.
The new model ranks first in accuracy and utility in 89 percent of cases, compared to two competing methods. It uses a type of machine learning called genetic modeling, which is able to generate new data points after being trained in existing ones.
The new model's primary job, called the DGMR (Deep Generative Model of Rainfall), is to predict the possibility of rain for the next one to two hours and has been approved by more than 50 meteorologists at the Met Office in the UK.
"This collaboration between environmental science and AI technology focuses on the value for disition-making method, opening up new avenues for rainfall Nowcasting and highlighting AI opportunities to support our response to decision-making challenges in an ever-changing environment.Writes the DeepMind Nowcasting team in a blog post.
The researchers behind the DGMR describe the creation of small "radar bands" that create future radar models from the past.
Lots of current tools, including pySTEPS, use numerical weather forecasting (NWP) approaches - essentially deploying mathematical models in current conditions to understand what future conditions will be like. These are powerful models, but they are more expensive in the long run.
The DGMR aims to harness the enormous power of artificial intelligence, while removing ambiguity in existing machine learning-based prediction models, including U-Ne, as these models may have difficulty maintaining accuracy in every part of the process.
Now that DeepMind's new and improved model has been approved by real meteorologists in the UK, researchers can look at integrating it into existing weather forecasting systems.
Of course it will take some more time and further improvement, until the new method is completely accurate.
The survey was published in Nature.