UK-based AI technology developer DeepMind, a subsidiary of Google’s parent company Alphabet, has partnered with the Met Office to develop a model that improves the accuracy of precipitation nowcasting.
Published in the journal Nature, DeepMind’s paper details how the company worked with the Met Office to apply machine learning to weather forecasts, particularly for rainfall.
According to DeepMind, current numerical weather prediction (NWP) systems are adept at solving physical equations to provide predictions several days ahead; however, they struggle to generate high-resolution predictions for short lead times under two hours, which is where nowcasting is used.
A critical tool for sectors such as water management, agriculture, aviation, emergency planning and outdoor events, nowcasting has benefitted from advances in weather sensing that make high-resolution radar data available at a high frequency (every five minutes at 1km resolution).
DeepMind believes that it is this availability of high-quality data that provides the opportunity for machine learning to make its contributions to nowcasting and improve on existing methods.
In the paper, DeepMind uses generative modeling to make detailed and plausible predictions of future weather based on past weather. “Conceptually, this is a problem of generating radar movies,” commented DeepMind on its blog page. “With such methods, we can both accurately capture large-scale events, while also generating many alternative rain scenarios (known as ensemble predictions), allowing rainfall uncertainty to be explored.”
The study used radar data from both the UK and the US and was particularly interested in the ability of the generative models to make predictions on medium to heavy rain events.
DeepMind’s model was able to demonstrate statistically significant improvements in these regimes compared with competing methods. “We conducted a cognitive task assessment with more than 50 expert meteorologists at the Met Office, who rated our new approach as their first choice in 89% of cases when compared with widely used nowcasting methods, demonstrating the ability of our approach to provide insight into real-world decision makers,” said DeepMind.
It later added, “By using statistical, economic and cognitive analyses we were able to demonstrate a new and competitive approach for precipitation nowcasting from radar. No method is without limitations, and more work is needed to improve the accuracy of long-term predictions and accuracy on rare and intense events. Future work will require us to develop additional ways of assessing performance, and further specializing these methods for specific real-world applications.”
To view the full version of ‘Skilful precipitation nowcasting using deep generative models of radar’, click here.
