close
2d8e907c-3787-11ea-8cda-aaaa00151f43.jpg

Editor’s note: This article is from the “love child norm”, author: Li Chen, 36 krypton release authorized.For human beings, time has always been the greatest enemy, and transcending time has always been a human dream. “Predicting the future” is a way that humans want to transcend time.Weather forecasting is the most common type of “predicting the future” in life, but as I said just now, forecasting the weather is also very difficult.Now, AI may be a tool that can greatly enhance the functionality of weather forecasting.▲ Picture from: British Council Learn EnglishGoogle recently shared a new study in its official blog claiming that Google has achieved “near real-time” weather forecasting.However, this work is still in its early stages and has not yet been integrated into any commercial system, but early research results still show great hope.In this paper, which has not been reviewed by industry experts, Goolge researchers describe how they achieved accurate rainfall predictions in just a few minutes, six hours ahead in a one-kilometer range.The calculation time of several minutes is a huge improvement compared to the current one. According to the existing technology, it may take several hours of calculation to generate predictions, although they take longer to generate more complex data.Researchers said that rapid prediction has great practical significance, which will effectively adapt to climate change, especially under extreme weather conditions. Rapid prediction will be a very important tool.Short-term forecasting is of great importance for certain crisis avoidance, and proper application can effectively avoid loss of life and property.Google’s biggest advantage in forecasting is speed, but how does this speed come about?The researchers compared their prediction methods with two current mainstream prediction methods: the optical flow method (by observing the movement of phenomena such as clouds) and the simulation method (creating a physical weather system simulation).The problem with these traditional methods is that the amount of calculation is extremely large, especially the simulation method needs to calculate a large number of physical effects.Like the US federal agency’s simulation for weather forecasting, it can process up to 100TB of data from different weather stations every day, and it can take hours to perform simulations on expensive supercomputers.Based on a calculation of 6 hours at a time, it can only be calculated at most 3-4 times a day.In contrast, Google’s methods only take a few minutes, because instead of trying to model complex weather, they do computational predictions on simple radar data.The researchers used historical radar data collected by the National Oceanic and Atmospheric Administration (NOAA) near the United States between 2017 and 2019 to train their AI models.Researchers say their method is as good or even better than the three existing methods that use the same data.However, the AI ​​model does not perform well when forecasting forward forecasts over 6 hours.At present, this is the best choice for machine learning in weather forecasting: make short-term predictions quickly, and longer-term predictions are handed to more powerful models, like NOAA can create 10-day weather forecasts.Although we have not yet seen the practical application of AI in weather forecasting, many companies are already working in this field, including some companies that we are familiar with, such as IBM and Monsanto.Like Google researchers say.With the interaction between humans and climate, this prediction technology will become increasingly important in the future.Cover image by pexels

Tags : EntrepreneurshipInternet entrepreneurshipInternet entrepreneurship project