The peer-reviewed journal Science published a study showing a Google AI model named GraphCast outperformed conventional weather forecasting methods. It could predict global weather conditions 10 days in advance in less than a minute! In contrast, PopSci says existing methods took hours to produce the same results.
The world faces unprecedented and erratic weather patterns due to climate change and many other factors. Consequently, we have lost many lives and resources as we struggle to anticipate these unusual phenomena. Fortunately, Google’s GraphCast could potentially save millions worldwide by predicting weather faster than ever before.
This article will discuss how Google researchers created a powerful AI weather forecasting tool. Later, I will cover the other ways artificial intelligence helps predict weather.
How did Google make its AI weather forecasting tool?
The program is a Google DeepMind project led by GraphCast experts. Team member Remi Lam trained their AI model on decades of historical weather information.
Google said its development team used four decades of weather reanalysis data from the ECMWF’s ERA 5 dataset. It contains details like satellite images, radar, and weather stations.
Google also used a traditional Numerical Weather Prediction (NWP) system to recreate a comprehensive global weather record. As a result, it enables the AI system to make cause-and-effect relationships explaining how Earth’s weather evolves.
The AI weather forecasting program predicts five Earth-surface variables: mean sea-level pressure, temperature, and wind speed and direction. The training was computationally intensive, but it enabled the software to yield effective results.
The GraphCast webpage says it takes less than a minute to create 10-day forecasts on a single Google TPU v4 machine. On the other hand, HRES and other conventional approaches take hours using supercomputers and hundreds of machines.
It also provided more accurate predictions on over 90% of 1380 test variables and forecast lead times. Moreover, this unique weather prediction tool only requires two sets of data.
You may also like: AI music study predicts hit songs
The first is the weather records from six hours ago, and the second is the current weather. In response, GraphCast will predict six hours into the future.
Google DeepMind said it could use that data to produce forecasts reaching 10 days in advance. Also, it could identify severe weather events earlier than traditional forecasting models.
Its AI weather forecasting program doesn’t need training on these phenomena. Instead, Google could apply a simple cyclone tracker to predict cyclone movement more accurately than the HRES “gold standard.”
What are the other recent AI weather projects?
Many have been experimenting with artificial intelligence to predict natural phenomena. For example, University of Texas researchers created an AI algorithm that predicts 75% of earthquakes. Here’s how they created it:
- The experts started by giving their artificial intelligence statistical features based on the team’s knowledge of earthquake physics.
- Then, they told the AI to train on a five-year database of seismic data.
- Afterward, it listened for signs of incoming earthquakes.
Sergey Fomel, one of the researchers, said, “Predicting earthquakes is the holy grail. We’re not yet close to making predictions for anywhere in the world, but what we achieved tells us that what we thought was an impossible problem is solvable in principle.”
The director of UT’s Bureau of Economic Geology, Scott Tinker, acknowledged the team’s earth-shattering achievement. “We are very proud of this team and its first-place finish in this prestigious competition,” he said.
The Philippines also partnered with JICA and the Japanese startup Spectee for an AI-powered disaster management system. It visualizes a crisis using data from river/road cameras, climate information, people traffic, social media, and other methods.
You may also like: How AI improves weather forecasting
The JICA system verifies information with image and sentence analysis and credibility scores. Then, human experts recheck and verify the data.
Afterward, the AI program provides reliable notifications through a digital dashboard, smartphone alerts, and emails. JICA chief representative Takema Sakamoto stated, “JICA believes that AI brings benefits to society if we use it well. The Philippines needs and is ready for this kind of technology.”
“We see the potential that more Filipinos would benefit from this digital innovation from Japan, particularly in the area of disaster management,” he added.
Conclusion
Google DeepMind and GraphCast researchers developed an AI weather forecasting tool that is significantly more efficient than traditional methods. The new system only requires a few seconds to provide results, but the latter needs hours.
It may require more research and development before the world adopts this system. Nevertheless, AI-powered weather predictions will become a part of our lives.
Learn more about this AI weather forecasting study on the peer-reviewed Science webpage. Also, check out more digital tips and trends at Inquirer Tech.