Deepmind’s AI Can Predict Where it is Going to Rain in the Next 90 Minutes

DeepMind’s AI Can Predict Where it is Going to Rain in the Next 90 Minutes

January 7, 2022

Scientists at the Google-owned AI research lab, DeepMind, have developed a deep learning predictive model that they claim can accurately predict exactly where and when it is going to rain in the next two hours. The London-based firm, in partnership with the Met Office – UK’s national weather service – is addressing one of the toughest challenges of weather forecasting. 

The tool called DGMR can accurately forecast the likelihood of rain in the next 90 minutes. The tool outperforms the existing tools across different criteria including prediction of location, extent, and intensity of the rain. The study has been published in the journal Nature

Forecasting heavy rain is extremely crucial as it influences a number of things right from farming to aviation. In fact, heavy rain can lead to a number of hazards from damage to infrastructure to human life. It is extremely crucial to be able to forecast the critical storms and floods accurately in order to take the appropriate precautionary measures to minimize damage. “Extreme weather has catastrophic consequences, including loss of life and, as the effects of climate change suggest, these types of events are set to become more common,” said Niall Robinson, Head of Partnership and Product Innovation, Met Office.

Current forecasting techniques mostly rely on computer simulations of atmospheric physics. While these are very good for forecasting the conditions in the future (say six houses), they are less effective at predicting what’s going to happen in the next hour or so, also known as nowcasting. There are some existing deep learning techniques to address this challenge, however, they are primarily good at one thing such as predicting location but fail to consider other factors such as the intensity of the rain or vice versa. Niall Robinson stated, “One forecast gets precipitation in the right location but at wrong intensity, or another gets the right mix of intensities, but in the wrong place and so on. We went to a lot of effort in this research to assess our algorithm against a wide suite of metrics.”

DeepMind’s DGMR, on the other hand, uses “statistical, economic and cognitive measures” to provide “improved forecast quality, forecast consistency, and forecast value” for short-term predictions at times when “existing methods struggle”. The DeepMind team trained their model on radar composites which is released every five minutes from 2016 to 2019. The radar measurements track the formation and movement of clouds and the snapshots together present rain patterns in the form of a stop-motion video. 

The approach was tested by 56 weather forecasters at the Met Office and a whopping 89% of them preferred the results presented by DGMR over the rival tool in a blind comparison across different factors. 

The study also notes that there are issues associated with probabilistic nowcasting which still need to be addressed. For instance, although the tool seemed to be very effective as compared to the other solutions, the prediction of heavy rain at lengthy lead periods is still difficult to anticipate for all techniques. However, the scientists believe that the work serves as a “foundation for new data, code and verification methods”, along with a greater scope for integration of machine learning  and environmental sciences in “forecasting larger sets of environment variables – that makes it possible to both provide competitive verification and operational utility.”

DeepMind’s collaboration with Met Office is what sets this project apart as the input from the experts at Met Office has pushed the model development in the right way. As Suman Ravuri, research scientist at DeepMind stated, “Otherwise we might have made a model that was ultimately not particularly useful.” 

DeepMind is Tackling Some of the Most Challenging Science Problems

It appears as if DeepMind is finally able to address some of the most challenging science problems posed to humanity. The Science team is also working on accurately predicting the 3D shape of proteins – a feat regarded as one of the biggest challenges in the world of science and biology. The scientists working on this problem are anticipating that the work done in this field could help in creating treatments for rare diseases or even find ways to break down plastic waste. 

Leveraging AI to provide societal and economic value has endless opportunities. From addressing climate change to treating cancer, there is a lot of research work going on in the field. Perhaps in future, AI will be able to address some of the most pressing challenges of humanity.


Leave a Comment

Your email address will not be published.