Google AI beats supercomputers for fast, accurate weather forecasts

The next time you roll your eyes at a weather forecast that was wrong, remember that predicting weather is one of the most complex problems in science. Now, Google has put artificial intelligence into the role of weather forecaster and demonstrated that it can make accurate predictions up to 10 days in advance in just one minute. This task would typically take a roomful of supercomputers several hours to complete

The famous butterfly effect hypothesis states that whether a storm brews may be affected by something as small as the flapping of a butterfly's wings on the other side of the world. The weather forecast's job is to turn information about these well-known butterflies into accurate models that tell you whether you should go ahead and plan next Saturday's picnic

Doing so involves so-called numerical weather prediction (NWP), which uses current weather observations from around the world as input data and operates through complex physics equations run on supercomputers. But now, Google has launched an artificial intelligence system called "GraphCast" that can process data faster on less powerful hardware.

This artificial intelligence was trained on 40 years of weather reanalysis data collected from satellite imagery, radar and weather stations. GraphCast will obtain the weather conditions and current status six hours ago, and use this valuable data to predict the weather conditions six hours from now. Based on this, it is able to forecast forward in 6-hour intervals and build weather forecasts up to 10 days

GraphCast does this on more than 1 million grid points on the Earth's surface, each with a latitude and longitude of 0.25 degrees. At each point, the model takes into account five variables at the surface, such as temperature, pressure, humidity, wind speed and direction, and six variables in the atmosphere at 37 different altitudes.

In tests, GraphCast was run on a Google TPU v4 machine and compared to the current gold standard for weather forecasting: a simulated system called High-Resolution Forecasting (HRES), which runs on Super run on the computer. GraphCast is able to predict weather for the next 10 days in under a minute and is more accurate than HRES on 90% of the tested variables and forecast lead times. When these models focus on the troposphere, GraphCast outperforms HRES 99.7% of the time. The troposphere is the lowest layer of the atmosphere and accurate predictions are very useful and applicable in daily life

Even more impressive among the capabilities demonstrated by GraphCast is that, despite not being specifically trained to do so, it identified severe weather events earlier than HRES. In one real-world example, AI was able to accurately predict where a hurricane would make landfall nine days in advance, whereas traditional forecasts were only confirmed six days in advance

Google said that GraphCast’s code is open source, allowing scientists around the world to experiment with it and incorporate it into daily weather forecasts. This kind of number crunching feels like the perfect job for artificial intelligence so they can leave art and writing to us humans.

The results of this study were published in the journal Science

If friends like it, please pay attention to "Knowing the New"!

The above is the detailed content of Google AI beats supercomputers for fast, accurate weather forecasts. For more information, please follow other related articles on the PHP Chinese website!