Google GraphCast AI weather forecasts revolutionised, beats supercomputers

tech.hindustantimes.com:

In a groundbreaking development, Google DeepMind's GraphCast AI, a machine learning algorithm, has demonstrated superior weather prediction capabilities compared to traditional supercomputer-based methods. The model claims to provide more accurate 10-day forecasts in minutes, outshining the High-Resolution Forecast (HRES) system used by the European Centre for Medium-Range Weather Forecasts (ECMWF), currently considered the gold standard in weather simulation. So what is Google Google GraphCast AI is all about.

Published in the journal Science on Nov. 14, the findings reveal that Google GraphCast, running on a desktop computer, excelled in over 99 percent of weather variables across 90 percent of the 1,300 test regions. Despite its success, researchers caution that the model's "black box" nature, lacking the ability to explain its pattern-finding process, suggests it should complement rather than replace existing tools, Space.com reported. Also read: Google Pixel Buds Pro launch confirmed, will take on Apple AirPods Pro!

Traditional forecasting relies on energy-intensive supercomputers, employing complex physical models and granular data for accurate predictions. In contrast, machine learning weather models like Google GraphCast operate more efficiently, utilising less computing power and delivering faster results.

Trained on 38 years of global Earth weather data up to 2017, GraphCast established intricate patterns between variables like air pressure, temperature, wind, and humidity. It then extrapolated 10-day forecasts from 2018 global weather estimates, achieving remarkable accuracy compared to ECMWF's high-resolution forecast. Also read: Beware! Google bans this Google Play Store app; delete HIDDEN Joker Malware from