Tomorrow it will rain, say. Most of the times it is so, because the weather forecast is handled well in stripes of 24 hours. But the weather today don't know the answer to another question more specific: is it going to rain when leaving the house in 15 minutes? Physics can't answer, but the artificial intelligence is here to help.
This advancement is called nowcasting, which is something like the "immediate prediction of the time". Prediction current fastest entails at least six hours of computation and its spatial scope is not less than three miles, although this tends to be higher. So the physical, as much, can prophesy in the morning, the weather will be in the afternoon.
This advancement is called 'nowcasting', which is something like the "immediate prediction of the time -"
How exceeds the technology the limits of physics? Analyzes million photos of clouds of a specific space. By observing how they have evolved in the past those clouds, it is able to predict how it will change the clouds today. Therefore, instead of working with a model that analyzes the complex physics of rain or not, which requires a computational effort that no supercomputer can accelerate, predict how will be the image of a radar after analyzing millions of images previous.
A model of machine learning learn to hit that a cat is a cat because I have been nourished with millions of photos of cats. The machine sees pixels that are in all probability a cat. Do the same process with the clouds: millions of photos of clouds, made by the radars allow the model to predict what will be most likely the cloud in the next few minutes, and so on.
Google has just published a scientific paper on nowcasting of rainfall in the united States: "we Explored the efficacy of treating the nowcasting of precipitation as a problem of image to image," explain the researchers of the company. Analyze the territory divided in grids of a mile and get the results with a probability of rain given minutes after the pictures are taken by radar. The method is new and is not infallible, but there is nothing better for now to know the exact behavior of a storm. In addition, this model serves only for rain, not expected to temperature or other phenomena.
The physical-traditional is not in a position by now to despise this helps
The physical-traditional is not in a position by now to despise this help: "they Are a breakthrough for meteorology, and especially a solution for episodes of complicated and you need a specific resolution and set, as cold drops or storms of summer very local", said Mar Gómez, phd in physical sciences and head of meteorology of the time.is.
The sensors of the sun
The prediction of rain is not the only field where he works nowcasting. Also, with different purposes, in the solar radiation. With a method similar to that of Google, but with radiation data instead of images of clouds, a team of Spanish scientists is able to predict the solar radiation for immediate by sensors: "If we want to predict future values of radiation of a sensor we use passed values for that same sensor and of others around them," says Alberto Torres Barrán, a postdoctoral researcher in the Institute of Mathematical Sciences (ICMAT).
These methods are being refined, but already can have concrete applications. A solar power plant, for example, need to foresee the amount of sunlight that will get your mirrors to succeed in your placement and are capable of generating electricity. Now that work is done with cameras that make pictures in the sky: if there comes a cloud, expect that the radiation will go down. But the cameras are expensive and less effective than the sensors: "A sensor detects the radiation in each moment, no matter why it goes up or down, the cameras change sometimes do not perceive clouds to be weak or dust," says Torres.
If in the future the roofs of the houses have solar panels, such a model will allow you to calibrate your ability
in Addition to the plants, if in the future the roofs of the houses have solar panels, such a model will allow you to calibrate your ability. The electric companies know how much electricity you can count on each group of plates. "The wind power plants already spend a lot of money in predicting how much electricity will be generated and, if not met, power companies may put a fine," says Torres. When you have thousands of plates in a city, to know the radiation that you receive every roof in the next few hours will be essential for the electrical.
The nowcasting is a product of the explosion of big data of about four or five years. The reason of its emergence is twofold: one, the increasing availability of databases, such as images of radars, satellites, or the values of sensors. The original research of Towers are made with open data from the sensors of the airport of Hawaii. And two, by the increasing computing capacity of computers. All of this allows models of machine learning are going to improve and be able to produce results more refined.
The prediction is still not unbeatable. Google does not yet have any period to carry its nowcasting of rain to an application for users. You must adjust and calculate behaviors in other places to make it global and not in all the world there are good pictures of clouds. But it is clear that the innovation has specific purposes up to now unattainable.Updated Date: 18 January 2020, 12:00