Shulou(Shulou.com)11/24 Report--

If the weather forecast says 25% of the sky will be covered by clouds tomorrow, can you judge the weather tomorrow? I'm afraid it's hard. Maybe 25% of the clouds will gather together and bring a torrential rain, or maybe they are just many lovely scattered clouds in the sky to add to your joy on a sunny day. To predict the weather from the clouds, we need to know more.

Clouds affect the weather to a large extent. Surprisingly, the most advanced weather models can only give a very rough description of clouds, such as the 25% cloud cover we just mentioned. The reason is that clouds are often so small that weather models cannot take into account every small area of the sky. And if every area of the sky is taken into account, even the fastest supercomputer cannot do the calculations needed for the weather forecast. Even if computing power has increased sharply in recent years, it is not enough to solve this problem.

"there is a big difference between a huge cloud and many very small clouds," said Devine, a mathematician at the University of Bath. "these two situations will lead to huge differences in weather behavior, but the weather forecasting models currently used do not take this into account.

However, with a slight change of direction, we can see hope. Instead of trying to speed up computing, maybe we can use the power of computers to complete complex tasks by learning a large amount of existing data. This is machine learning, which is a form of artificial intelligence. From online shopping to health care, artificial intelligence is entering all areas of life. If this idea also applies to meteorology, weather forecasts will become more accurate and require less computing power than current weather models.

The traditional weather forecasting model is the result of the movement of the earth's atmosphere and ocean, the movement of water in the atmosphere, and the change of air pressure and temperature. The atmosphere and the ocean belong to gases and liquids, respectively, and they both belong to fluids, and in meteorology, there happens to be a set of equations that describe the motion of fluids: Navid-Stokes equations.

The principle behind the weather forecast is relatively simple. First measure the factors that describe the current weather, such as temperature, air pressure and density, wind speed, and air humidity. Then, the data are provided to a mathematical model based on the Navid-Stokes equation, so that the weather changes can be calculated on the computer in time.

In practice, however, there are several things that can make weather forecasting tricky. First of all, you can't measure the temperature, pressure, humidity and so on at every point on the earth. Second, you can't measure them with infinite precision. The famous butterfly effect means that as the calculation goes on, the inevitable small error may become very large, resulting in a very biased prediction. Third, due to the complexity of Navid-Stokes equation, it takes a lot of computing power to apply it to weather models.

In order to be able to make predictions, weather modelers divide the earth and its atmosphere into a grid, just as television or computer screens divide images into pixels. Just as each pixel is assigned a color, each grid box is assigned only a value for pressure, humidity, temperature, and so on-a value that is accurately measured from a single grid box, which makes the calculation easy. Then we can use techniques such as integrated prediction to reduce the impact of the butterfly effect.

The weather model divides the earth and its atmosphere into a grid. Photo: national Oceanic and Atmospheric Administration. In the most advanced weather models available, the grid is about 1.5 kilometers square in the horizontal direction and about 300 meters high in the vertical direction: even the fastest supercomputers cannot handle higher resolutions. Clouds can of course be much smaller than this. They can do all kinds of wonderful things in a grid box, and many other processes will take place on a scale smaller than the grid box.

In order to take these processes into account, weather models are estimated using mathematical formulas that roughly describe the physical properties of these processes. This estimate is called parameterization.

"Parameterization is a step in modeling that calculates the physical properties of what happens in the grid box and then correlates it with the grid scale," explains Chris Bard, a mathematician and weather forecast and machine learning expert at the University of Bath. The proportion of the sky covered by clouds in a single grid is such a parameterized amount. "in addition to clouds, there are also parameters such as radiation from the sun, fluctuations caused by gravity in the atmosphere, and friction experienced by the wind as it blows across the earth's surface," Bud said.

What can AI do? Machine learning means that computer algorithms learn how to discover the rules in the data, and then make full use of these laws for practical application. Here is a classic example of a computer learning to distinguish a cat from a picture of a dog. To teach machine learning algorithms to do this, first enter a large number of pictures of cats and dogs and tell them the correct answer to each picture-whether it is a cat or a dog.

In a seemingly magical but efficient mathematical process, the algorithm carefully analyzes the pictures and adjusts the internal parameters until a very high accuracy is achieved in the training set. Then you can give him a new picture of cats and dogs, and he can distinguish the animals in the picture with high accuracy.

When it comes to weather forecasting, we hope that machine learning algorithms can learn how to determine some details of what is happening in the grid box from the numbers associated with the grid box by looking at a large number of real-life weather. If possible, these algorithms can be incorporated into weather models, replacing existing parameterized algorithms, and allowing the model to contain more detailed information about the subgrid process-- including more details about the behavior and organization of the cloud.

Trial AI Bard and Devine are both members of a research group called Mathematics in Deep Learning, which explores a range of potential applications of machine learning and the mathematics behind it. They mentored graduate student Coward on a project with the Met Office to test whether machine learning can provide more information about clouds.

The total surface area of these cirrus clouds is larger than that of clouds of the same volume. Photo: for such a test, the first thing we need to do is to determine what information we want the machine algorithm to learn about the cloud. Coward gives an answer based on the results of geometry: when the clouds are all gathered together, the surface area of the whole cloud is often smaller than when it is divided into many small clouds.

Therefore, the surface area of the entire cloud, also known as the cloud perimeter, is a good indicator of what kind of clouds are in the grid box-large cumulus clouds or slender cirrus clouds. It is also a useful parameter to improve other parameterized processes and algorithms, such as algorithms for predicting the transport of radiation through clouds.

The question is whether the machine learning algorithm can estimate the cloud perimeter within a single grid box based on the numbers assigned to the entire grid box. "this is the goal of the Coward project: machine learning to estimate the cloud perimeter based on a range of environmental factors." Devine said.

To train the algorithm, Coward used data sets of clouds recorded in Oklahoma. "they set up a bunch of cameras in their space," Devine explained. "the cameras can read whether there are clouds on a grid scale of one meter." For three years, clouds are recorded every 20 seconds, and using this data, machine learning algorithms produce what Coward calls "a completely unique view of the cloud's life cycle."

Coward uses this data to train two machine learning algorithms. After training them, he compared the cloud perimeter predicted by the algorithm with the cloud perimeter recorded by the camera.

The better error of the two algorithms is 16%. It's not zero, but it's not very big. In fact, the best method of parameterizing the cloud perimeter has an error of nearly 24% without using machine learning. Therefore, in this case, the accuracy of machine learning is more than 1/3 higher than that of non-machine learning.

Proof of concept, Coward's project is one of a series of preliminary attempts to test whether machine learning can be used in weather forecasting. "Machine learning is a very new approach for people in this field," Devine said. "We are in the initial stage, most of the content is experimental, and people are trying different things, trying to come up with new technologies and see how they perform."

It is hoped that machine learning will eventually be able to calculate not only clouds but also other phenomena in weather models. If this method succeeds, artificial intelligence will eventually be applied to the App of weather forecast, and you will know the good news then.

Author: Marianne Freiberger

Translator: Tibetan idiot

Revision: Xiao Cong

Original link: Catching clouds with artificial intelligence

This article comes from the official account of Wechat: Institute of Physics, Chinese Academy of Sciences (ID:cas-iop), author: Freiberger

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.