If you have the habit of reading weather reports, you will probably deeply feel the unreliability of the forecast: it is said that it will be sunny all day, but there will always be sudden rainstorms. Although meteorologists know the laws of motion of the atmosphere, the actual calculation is very difficult. Supercomputers are needed to calculate the future weather conditions from the current weather observations. In reality, it is difficult for us to observe the weather conditions everywhere. However, the atmosphere is a "chaos" system, and even if the input values are only slightly different, completely different predictions may be obtained - this is also the "butterfly effect" that we are familiar with.
Therefore, it is still very difficult to accurately predict whether company banner design it will rain in a place, when it will rain, and how heavy it will be. In July 2018, continuous torrential rains in Kansai, Japan caused severe floods in various places, killing more than 200 people, and the victims even had to climb on the roof to wait for help. The Japan Meteorological Agency issued a warning at 2 p.m. on July 5, stating that record-breaking rainfall may be recorded in two days. But it turned out that 14 hours ago, the world-renowned private meteorological company Weathernews had used artificial intelligence to predict the crisis earlier, and accurately predicted that Hiroshima and other places would be the hardest hit areas! AI_5 In Japan, in addition to the official Meteorological Agency, there are also private meteorological operators providing the latest weather information and forecasts, many of which use artificial intelligence to improve the accuracy of forecasts.
Unlike traditional forecasting methods, artificial intelligence does not start from known scientific laws, but learns and finds laws from big data. It analyzes the radar images of rain clouds over the past three years, identifies the activity patterns of different rain clouds, and uses real-time radar images to make predictions. The spatial resolution can even be as high as 250 meters. In traditional numerical models, this resolution requires a huge amount of computation and is difficult to apply to real-time forecasting.