Google DeepMindthe division of Big G that deals with projects onartificial intelligence developed by the technology giant, he presented GraphCast: a new model of AI weather forecast which promises accuracy never seen before.
According to the statements of Googlethis system can guarantee reliable forecasts up to 10 days and with one 90% accuracy compared to traditional models in use.
The company did not hesitate to define it as a “carried out in the weather forecasting sector”, already publishing the first encouraging information in this regard. Here’s what we know.
How GraphCast works
Current models used for weather forecasting (called numerical weather prediction) use principles of fluid dynamicsOf thermodynamics and others atmospheric sciences to simulate ongoing weather changes.
Of course, this is a very complicated process, which requires computer centers a lotpowerful and with a decidedly non-negligible cost; so much so that, generally, they are systems owned by the Air Force.
GraphCast, on the other hand, goes against the trend and instead of using only these complex calculations, it also relies on historical datawith this machine learning model also referring to relative past weather forecastsof course, at the same time frame.
This is not just a look back and the system still requires dozens and dozens of complex calculations to arrive at a logical conclusion but, from the first tests, the reported data already appears to be very precise.
According to the first statements of GoogleGraphCast begins to analyze current weather conditions, then going back six hours and up to the analysis ever-longer time spanscomparing the various data and arriving at your own predictions.
Google GraphCast, the future of weather forecasts
The Google DeepMind team, as already anticipated, has already spoken of an accuracy of 90% on the first tests carried out, going forward over time for up to 10 days and naturally maintaining the same level of reliability.
It is clear that if this system achieved similar results also for long-term weather forecast, it could be a real turning point for meteorology and beyond.
In fact, GraphCast has already had great success in predicting too events extreme weathersuch as tropical cyclones and climate changes.
What is surprising is that these predictions turned out to be accurate even though the system had not yet been trained for this type of detection, a clear sign that the intuition is more than good and that Google’s powerful AI continues to exceed all expectations. .
According to Google scientists, with the new GraphCast the weather forecasting sector is at an epochal turning point, with themachine learning and theartificial intelligence which are already the tangible key to radical change.
This system, therefore, on the one hand will significantly improve the current forecasting methods and, on the other, will give meteorologists new and more powerful tools to carry out their work and also anticipate extreme events which are now increasingly frequent and increasingly dangerous for humans.