Engineers from Google Research have developed an artificial intelligence capable of guessing the weather within 5 minutes. This innovation could revolutionize the way we study the weather.
Known for its ambitious work, the Google Research laboratory has just published the results of its new weather experiments. The company’s engineers have developed an artificial intelligence called ML for Precipitation Now casting from Radar Images, capable of processing satellite images at unprecedented speed. According to Google, this AI would be able to predict the weather with an update every five minutes.
To anticipate specified unpredictable natural disasters, Google’s AI could, therefore, be particularly useful.
5 minutes against… 6 hours
Google’s article is particularly rich in scientific explanations. In summary, he explains that the systems currently used by meteorologists take hours to dissect radar images (HRRR, high-resolution rapid refresh). Optical flow predictions consist of analyzing the movement of clouds, while simulation forecasting anticipates the weather based on existing weather phenomena. It takes approximately six hours for agencies like the NOAA (National Oceanic and Atmospheric Administration) to analyze the terabytes of data received each day, which limits the number of updates to the weather map to… 4 times in 24-hour increments. It is too little to predict in real-time disasters like tornadoes, which are particularly unpredictable.
Also, Read About: Google Algorithm Updates
This is where the Google algorithm comes in. The engineers at the company had the funny idea of treating the weather as if it were not the weather. Exit the “a priori” meteorological, door the laws of physics … Google considers that each satellite captures is only a photograph.
You May Also Like>>> How To Search Reverse Image On Google From Mobile
The Google algorithm, therefore, ensures the same operation as that present on individual televisions: it recovers images in maximum definition, compresses them, and tries to reconstruct them by itself. By providing scanned weather images between 2017 and 2019 to its AI, Google taught it to guess the functioning of each cloud and each meteorological phenomenon. For each month, three weeks were used for learning, and the fourth was used for assessments. The result is impressive; Google’s forecasts are very close to reality.
Advantage of this system
The main advantage of this new system is its ability to perform calculations quickly. Google makes a prediction every five or ten minutes over a six-hour window, which lets you know exactly when the rain will stop, for example.
Currently, Google’s artificial intelligence cannot make weather estimates beyond six hours. Its field of action is limited to the near future (to know if going out to eat on the terrace is a good idea, for example) and not to predict several days in advance. Google says that for its system to be useful, it will have to combine its data with that of meteorologists. The company also admits having trouble measuring the direction of the rain (depending on the winds), which can skew the results. Google hopes to find a solution to these problems soon. Towards a Google Weather application?