THANK YOU FOR SUBSCRIBING
Deep-Learning for Rain Prediction: We are living in the best of times
The University of Exeter in England and the computer scientists at DeepMind are collaborating with meteorologists from the Met Office to build an AI model capable of predicting rains 90 minutes before it arrives

By
Apac CIOOutlook | Monday, October 11, 2021
Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.
The University of Exeter, DeepMind, and Met Office are collaborating to develop a deep-learning-based rain prediction model that is more efficient than traditional forecasting techniques.
FREMONT, CA: The University of Exeter in England and the computer scientists at DeepMind are collaborating with meteorologists from the Met Office to build an AI model capable of predicting rains 90 minutes before it arrives. Unlike traditional forecasting techniques, were a complex equation is solved that includes weather conditions such as air pressure, moisture, and temperature, Deep-Learning models, are better capable of making near-term forecasts, such as within the next few hours. This was backed by the paper publisher Nature. The utilization of AI has several benefits since it doesn't involve thermodynamic equations and require less computational power.
The team at DeppMinds trained a generative adversarial network (GAN) to generate a sequence map that highlights where it will rain. Each prediction shows on the map where the moisture be will more, covering an area of 1536x1280km. Millions of such maps were gathered from radar observation, dating from 2016 to 2018. This model was then fed a sequence of maps, each capturing weather data over five minutes intervals. It quickly learned to pick up patterns during the training period, such as patterns in clouds corresponding to rains. The system was then asked to generate the next series of maps and predict when rainfall would happen.
More than 50 expert meteorologists judged the model’s accuracy, who ranked it first for accuracy and usefulness in 89% of cases. Niall Robinson, the co-author of the study and head of the collaboration, said that the model’s accuracy is challenging to analyze. The team believed that the best way to check its feasibility would be to ask the end-users, the meteorologists. In a blind study case, end users overwhelmingly chose the new approach over their traditional algorithms. The team believes there is still a lot of work to be done, such as the careful deployment and maintenance of the system and developing a user-friendly interface for the meteorologists.