|Google's new AI system uses satellite images to produce “nearly instantaneous” and high-resolution weather forecasts / Credits: Hadrian via Shutterstock|
Weather forecasting is a difficult job that requires powerful computers and lots of observational data collected from land, sea, and air. It should be noted that weather forecasters have significantly improved their game over the past two decades. They are also much more equipped now in providing advanced warnings of severe weather. This is made possible through the help of artificial intelligence and machine learning.
Recently, Google introduced an AI system that uses satellite images to produce “nearly instantaneous” and high-resolution forecasts. The researchers believe that the tool is important for effective adaptation to climate change and crucial for crisis management and the reduction of losses to life and property. According to The Verge, an American technology news and media network operated by Vox Media, it outperforms traditional models such as optical flow (OF) predictions even at these early stages of development.
“If it takes 6 hours to compute a forecast, that allows only 3-4 runs per day and resulting in forecasts based on 6+ hour old data, which limits our knowledge of what is happening right now,” Google software engineer Jason Hickey said.
Instead of using complex weather systems, this new AI tool makes predictions using simple radar data, which explains why it can produce results in minutes. They said that they were able to generate accurate rainfall predictions up to six hours ahead of time at a 1km resolution from just “minutes” of calculation. Also, the new AI system takes a data-driven and physics-free approach to weather modeling. This means that it can learn to approximate atmospheric physics from examples alone.
According to Google’s researchers, they collected historical radar data between 2017 and 2019 in the contiguous US by the National Oceanic and Atmospheric Administration (NOAA) to train their AI model. The results were better than the three existing methods currently being used for weather forecasting.