Gencast, Google Deepmind’s weather forecasting tool, is said to produce 15-day weather forecasts better than the existing standard, the ENS ensemble forecast of the European Centre for Medium-Range Weather Forecasts. ECMRW forecasts also can forecast up to 15 days.
But GenCast outperforms; in several ways. GenCast is said to have outperformed outperformed ECMWF's system in 97.2 percent of forecasting scenarios tested, with a 99.8 percent accuracy rate for forecasts beyond 36 hours. Most models can forecast up to about a week in advance.
GenCast also demonstrated more precise predictions of tropical cyclone tracks, offering an average of 12 additional hours of advance notice, according to Alphabet.
GenCast also can produce a complete 15-day forecast (using more than 80 surface and atmospheric variables) in eight minutes. in eight minutes, using a single Google Cloud Tensor processing unit v5, while traditional models like ENS typically require several hours.
Weather forecasts are difficult due to the highly chaotic nature of atmospheric systems.The slightest discrepancy in initial measurements can cause weather forecasts to diverge from actual weather over time.
Some will note that chaotic systems are inherently nonlinear. Temperature, pressure, and humidity interact in ways that are not directly proportional or linear. So the system’s behavior cannot be accurately predicted by simply adding or scaling inputs. The variables interact in complex, non-additive ways.
This chaos means that beyond a certain time horizon, weather becomes inherently unpredictable, regardless of the quality of data or models used. Up to this point, the “gold standard” has been predictions generally correct for as much as a week.
Artificial intelligence, especially deep learning, excels at detecting complex patterns in large datasets, which is why it helps with weather forecasts.
No comments:
Post a Comment