Google develops `` a system that can instantly forecast weather ''


by

Angela Compagnone

Accurate weather forecasting is said to be 'very difficult,' but researchers have pointed out that machine learning can improve the accuracy of weather forecasting. Google has embarked on this approach, and a new paper says that Google researchers have developed a technology that can calculate accurate rainfall forecasts up to six hours later, every kilometer, 'per minute.'

Google AI Blog: Using Machine Learning to “Nowcast” Precipitation in High Resolution
https://ai.googleblog.com/2020/01/using-machine-learning-to-nowcast.html

Machine Learning for Precipitation Nowcasting from Radar Images
(PDF file) https://arxiv.org/pdf/1912.12132.pdf

Google says new AI models allow for 'nearly instantaneous' weather forecasts-The Verge
https://www.theverge.com/2020/1/14/21065095/google-ai-weather-forecast-predictions-rainfall-research



Many existing technologies take an hour to create a weather forecast, and more complex data can take even longer. From this, researchers see that Google's new technology is a 'big progress'.

In recent years, many people living on Earth have been swept away by unpredictable weather patterns, and researchers believe that immediately predictable weather forecasts are an essential tool to apply to climate change. That it is. Being able to make predictions in a short period of time enables crisis management and prevents loss of resources and human lives.

Existing weather forecasting technologies include ' optical flow, ' which reads cloud movements, and 'simulation forecasting,' which creates detailed physics-based simulations of weather systems. However, these two, especially simulation prediction, have a problem that the computational load becomes large. For example, the U.S. government's weather forecast processes 100 terabytes of data sent from weather stations every day, but it requires a few hours of supercomputer operation to make the forecast. This limits the number of times a day you can make a forecast, which results in older forecast data being published.

On the other hand, one of the existing technologies, Doppler radar , predicts rainfall in real time, but its accuracy is not perfect. The image below shows the position of the cloud as shown by the satellite above, and the position of the rain as measured by Doppler radar below, but you can see that there is a gap between the two.



Google researchers said that they collected AI radar data from 2017 to 2019 by the United States Oceanic and Atmospheric Administration (NOAA) and trained AI models. When researchers compared the performance of the new method with the performance of the conventional method, they found that the new method performed the same or better prediction than the conventional method. In particular, AI models have shown excellent results for weather forecasts after 6 hours.

This paper has not been peer-reviewed at the time of writing and has not been incorporated into commercial systems, but its future potential is considered 'very promising.'

in Software, Posted by logq_fa