DeepMind partnered with the Met Office to advance Nowcasting science and accurately predicting rainfall within a two-hour time window. (Image credit: Pixabay)
This is the first time I’ve been worried about AI taking a job over completely. It’s just a matter of time before meteorologists are out of the job as it sits now.
London-based DeepMind has partnered with the UK's Met Office (national weather service) to develop a system capable of predicting rainfall within a 2-hour window. The DGMR (Generative Approach) system could be used to predict flooding and heavy rains, which are key factors for aviation and emergency services. This kind of short-time weather prediction is known as precipitation nowcasting, which predicts rainfall on very short time scales over traditional radar-based models, which can make predictions in days or weeks beforehand.
Meteorologists typically use radar imagery to grab information about the weather. Radar works by sending a beam into the atmosphere and then timing how long it takes to reflect, which indicates how much moisture there is in the atmosphere. The collected data is then processed to get a picture of how much precipitation is in the air. The more precipitation, the more likely it will rain. DeepMind's deep-learning model was trained on radar imagery from the UK between 2016 and 2018 to advance Nowcasting science.
DeepMind's DGMR system is more accurate than other predictive systems, including PySTEPS and UNet. (Image credit: DeepMind)
DeepMind's system was then used by over 50 meteorologists in the UK and competed with other deep-learning systems, including another Nowcasting platform known as PySTEPS. The DGMR system came in first place when it came to accuracy in 90% of case models. While the DGMR system is incredibly accurate, it won't revolutionize weather forecasting. Instead, it will be used as another tool that utilizes AI to accurately predict future generations' weather.
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell