The Silicon Trend Tech Bulletin

Banner ads header
Icon Collap
Home / AI / ML / Google Develops AI-based Nowcast System with Minute Precision

Google Develops AI-based Nowcast System with Minute Precision

Published Tue, Oct 05 2021 04:30 am
by The Silicon Trend



Google Develops AI-based Nowcast System with Minute Precision

Weather forecasts are set to be transformed as Google's DeepMind engineers have made a drastic advancement in nowcast systems. Researchers at the London-based AI Google lab - DeepMind collaborated with the meteorologists at the Met Office for the AI system development. Such innovation had come under the light when the weather forecast sometimes failed to predict the rainfall chance.


Precipitation Nowcasting System

Along with the University of Exeter, experts have created a system known as precipitation nowcast that has high-resolution in forecasting rainfall up to 2 hours beforehand. It is vital and adds the 4 consecutive radars reading of the past 20 minutes to leverage as context for the generator to measure the expected rain prediction from medium to high in the coming 90 minutes.


You may read: A Leap Towards 100% Self-learning Machine with DeepMind


The study report was published in Nature, and the report mentions that leveraging cognitive, statistical, and economic measures, the system will offer enhanced forecast value, consistency, and quality. In addition, when existing approaches struggle, this system will provide a swift and precise short-term prediction at lead times.


MuZero DeepMind-1

Deep Generative Model of Rainfall (DGMR)

Present methods are primarily controlled by robust numerical weather prediction (NWP) that uses mathematical equations to estimate rain probability and other kinds of weather depending on the fluid's movement in the air. The project's co-lead - Suman Ravuri, added, "These models are really amazing from six hours up to about two weeks in terms of weather prediction, but there is an area - especially around zero to two hours - in which the models perform particularly poorly."

The predictions have been identified to be precise in 89% of cases against the existing forecast systems. Dubbed deep generative model of rainfall (DGMR) uses a type of ML - Generative Modeling, which produced new data points after being trained on the previous ones. As a result, researchers expect to foresee the precipitation chance across the next 1-2 hours. This technique has already received approval from more than 50 meteorologists at the UK Met Office.


You may also read: DeepMind AGI Paper Adds Importance to Ethical Artificial Intelligence


System Challenges 

DeepMind said the joining of AI and environmental science aims at the decision-makers' value, opening new opportunities for the rain nowcast, and points to the AI avenues in supporting the challenging responses of the decision in a surrounding under constant transformation. The system leverages radar composites acquired every 5 minutes between 2016 January and 2019 December to estimate and train the models over London. 

However, the studies suggest that these innovative techniques have challenges regarding the probabilistic nowcast. Though the system seems to have a skillful prediction, foreseeing heavy precipitation amid long leads might be complex for any method.




Image source: Deepmind,