Divination, physics and computer modeling. How weather forecasts became more accurate Earth

Divination, physics and computer modeling. How weather forecasts became more accurate

People have been trying to predict the weather since ancient times. Lacking precise tools, they were guided by symbols, superstitions and omens. It was more an attempt to guess the future than to foresee. However, this analysis of a series of patterns was, perhaps, the first prediction that a person formulated. It was not accurate, but it was used in practice — in agriculture and navigation.

We can find the first weather forecasts among the ancient Greeks and Romans, in Babylonia and during the time of the Chinese emperors. For example, Aristotle in his Meteorology identified the types of precipitation and for the first time said that precipitation is caused by cold. Weather forecasts are found even in the Bible. In the past, people associated weather phenomena with the anger and mercy of spirits and gods, so they prayed and made sacrifices.

Attempts to predict the weather by the behavior of birds and animals were also popular. By the way, some signs have a completely scientific explanation and really work. So, according to a Yandex study, a popular sign: It rains on Samson for seven weeks, then (July 10) comes true in Moscow with a probability of more than 50%. At the same time, Epiphany frosts occur extremely rarely in the capital and are not typical for the whole country. They regularly visit only 4 cities: Omsk, Khabarovsk, Novosibirsk and Irkutsk.

A revolutionary change in forecasting occurred in the XVII century. The idea of creating a weather prediction device was put forward by Galileo Galilei, and his students, Evangelisto Torricelli and Vincenzo Viviani, implemented it. In 1643, the first mercury barometer appeared and the well—known measure of measurement - millimeters of mercury column.

From that moment on, the pressure at a particular point could be much more accurately predicted what the weather would be — sunny (pressure increase) or rainy (pressure decrease). Despite the fact that only the instantaneous pressure value was tracked with the help of a barometer, and for the accuracy of the forecast it is necessary to know the trend of its change, this was the first scientific method of forecasting where the laws of physics were applied. Scientific progress and the development of technology turned the first enthusiasts into researchers who began to create societies of meteorologists.

By the end of the XIX century, meteorologists realized that a mathematical model should be used to make accurate forecasts. The American meteorologist Cleveland Ebbe became a pioneer in this field. His first works date back to 1873, and in 1901 he first attracted mathematics to solve the problem of weather forecasting. In fact, Ebbe proposed using models to predict the weather using the laws of hydrodynamics and thermodynamics.

Immediately after, namely in 1904, the work of the Norwegian meteorologist Wilhelm Bjerknes was published, in which he proposed to divide the process of weather forecasting into two steps — diagnosing the current state and forecasting for a time interval.

Bjerknes was also the first to identify 7 main variables describing the state of the atmosphere: pressure, temperature, density, humidity and two components of the air flow velocity. He also developed the first system of equations, the solution of which gives a weather forecast.

The real innovator and inspirer of all subsequent generations of meteorologists was Lewis Fry Richardson, who in 1922 was the first to apply numerical methods to integrate a system of Berkens equations.

Richardson's calculation gave an absurd result — the change in atmospheric pressure over Munich was predicted to be 14.5 kPa in 6 hours (this is an abnormal increase that did not occur). However, the error was not in the equation, but in incorrect conditions. The experiment was repeated decades later, using more accurate data, and it turned out to be successful. Thus, Ridcharson became the first meteorologist who successfully used a mathematical forecasting model.

The first meteorological observations in Russia began in 1725.

In 1834, a resolution was issued by Emperor Nicholas I on the organization of a network of regular meteorological and magnetic observations in Russia. And in 1837, pamphlets with weather forecasts in French began to be published, which were produced under the leadership of academician Adolf Kupfer. By this time, observations had already been carried out in various parts of our country, but the technological system and management of all observations according to uniform methods and programs appeared for the first time.

In 1849, the Main Physical Observatory was established. The first weather forecast published by her in large circulation appeared in the Daily Meteorological Bulletin in January 1872. This forecast was made on the basis of observations that were received from 28 tracking stations (26 on the territory of the Russian Empire and 2 foreign stations).

The modern Meteorological Service in Russia was established on June 21, 1921, and on January 1, 1930, the Central Weather Bureau of the USSR was formed in Moscow. In the same year, a meteorological probe was launched in Leningrad for the first time in the world, which rose to a height of about 8 km, measured the air temperature and sent a radio signal to earth.

Then scientists received a technology that allowed them to take readings of sufficiently high accuracy at high altitude. Since then, maps with weather parameters obtained at different heights have been used to make forecasts that include pressure, temperature, precipitation and wind direction. With some changes, this technology is still used today.

The first meteorological satellite TYROS-1 was launched in the USA in 1960. In the Soviet Union, the satellite was first used for meteorological observations in 1967. Kosmos-144 became the progenitor of a series of Meteor satellites that are currently used in orbit. With the help of satellite sensing, meteorologists have the opportunity to monitor clouds, which allows them to refine maps and the location of cyclones and anticyclones on them. In 1978, this information was used to visualize weather forecasts in TV news.

Among dozens of meteorological satellites in orbit there are two Russian ones: Arctic-M and Electro-L. This allows you to accumulate a large amount of data and build meteorological models. Since we are talking about terabytes of data, meteorologists are not able to work with such a volume of information manually. Supercomputers that are able to work with global weather forecast models have been used for analysis and forecasting. These models break the Earth's atmosphere into cubes, each of which solves an equation with its own introductory ones.

Over the past century and a half, from the first attempts to use mathematical models to the use of satellites and probes, forecasts have become much more accurate. The increase in the accuracy of forecasts in our country was associated with the restoration of the network of weather stations in the post-war period and the introduction of numerical forecasts in the 60s.

But with a high degree of probability, it was impossible to find out the weather for the coming weekend a few decades ago. Today, you can get a fairly accurate forecast using websites and mobile applications that tell you about the weather both for the next hour and for the month ahead.

Satellite sensing of the atmosphere, faster and more powerful computers and progress in understanding the physics and dynamics of the atmosphere helped in this, so today the weather forecast for the week has become much more accurate than the weather forecast for tomorrow several decades ago.

As already mentioned, improving forecasting depends on increasing the accuracy of data and their quantity. However, there are a number of obstacles. It is difficult to put stations in hard-to-reach areas, and it is expensive to increase coverage due to the growth of their number. But even with this approach, the prediction result will not be 100% accurate.

What dimensions can be added to solve this problem? For example, messages about the weather outside the window of users of weather services.

Yandex first offered anyone to mark places where it is raining on the map with umbrellas, and then introduced Meteum 2.0, a new weather forecasting technology based on machine learning. Its algorithms are the only ones in the world that are trained not only on data from instruments and weather stations, but also on messages from users.

Every day users send to Yandex.There are more than a million precipitation reports in the weather, and up to three million on some rainy days. For comparison, weather stations send about 8 thousand observations per day. These data helped to improve the short-term forecast of precipitation across the country, and especially in the Urals, Siberia and the Far East, where there are very few meteorological radars.

Meteum 2.0 takes into account the forecasts of five different weather forecast models, one of which has its own. Satellite images and radar measurements are also added to them. The information is processed and combined using a CatBoost-based machine learning model and neural networks. It allows you to take into account not only wind speed, air temperature and other weather parameters, but also additional factors — for example, distance from a reservoir or the height of the sun above the horizon.

Such a model is trained to predict user messages about the weather. For example, a user can provide an application with the ability to collect data from their device or report on the real weather at the moment, which makes future forecasts even more accurate. This forecast already works for rain and directly affects the precipitation map. In addition, the company uses messages with a high credit of trust to correct the erroneous precipitation map in real time. That is, a user who has already been caught in the rain becomes a reliable source. Such information is unlikely to help him, but it will be useful for residents of neighboring districts. As a result, the short-term precipitation forecast became 20% more accurate.

In the future, the synergy of machine learning, big data and user experience will improve the quality of weather forecasts, as well as increase the accuracy of long-term forecasts. After all, until recently it seemed incredible to make an accurate forecast for a few days, and today everyone can participate in making a forecast and make it more accurate.