About the weather for tomorrow
At the beginning of spring, the mysterious but realistically existing ChipMan, together with several colleagues, visited the Roshydrometcenter, where he talked with the heads of Roshydromet and its computer center for several hours and looked at the supercomputer racks.
Below is a brief insight into the history of meteorology and how Roshydromet and other modern meteorological services work.
Meteorology has a long history - the first meteorologist, apparently, was Aristotle, who wrote the eponymous treatise on weather changes; in subsequent centuries, observing the weather and the systematization of knowledge about it were engaged in all parts of the planet. However, meteorology as we know it - and this is dynamic meteorology, that is, weather forecasting using mathematical equations of fluid mechanics and thermodynamics - appeared relatively recently.
')
We owe the term “weather forecast” to the hot-tempered captain Robert Hot Coffee Fitz-Roy, who had been swimming with Darwin for almost five years during Beagle’s second voyage in the middle of the nineteenth century. Fitz-Roy discovered that the weather characteristics are related to air pressure, and in his mature years he headed the meteorological department at the Ministry of Commerce, created several models of barometers and arranged data collection from captains who came to the port - in short, he did what all modern meteorologists do.
Modern meteorologists use the equations of the Norwegian Wilhelm Bjerknes, the founder of the Bergen scientific school in meteorology - the solution of the hydrodynamic equations derived by him is nothing more than a weather forecast. These equations found practical application only in the fifties of the last century, when computers began to appear capable of calculating them for a fairly large area - the first weather forecast was considered by American meteorologists on the very first ENIAC computer with the help of John von Neumann - meteorology cannot do without its famous history people.
The first domestic weather calculations began in 1961 on an M20 computer that performed 20,000 operations per second. Opening hours M20 to failure - 15 minutes; the weather model was considered much longer, so the staff of the Hydrometeorological Center had to go for various tricks. The very specifics of weather calculations are such that computer technology is used twenty-four hours a day, seven days a week, all year round — the weather never stops changing, so the Hydrometeorological Center was and remains a serious test area for supercomputers: until the mid-seventies, when domestic computer development Replaced by analogues of machines IBM, Meteorological Center received the first serial numbers of Soviet computers.
Since March of last year, two SGI Altix supercomputer systems with a capacity of 11 and 16 TFlops have been launched at the Hydrometeorological Center - more than a billion times faster than the first M20 computer. Meanwhile, weather forecasts have not become a billion times more accurate since. The accuracy of weather forecasting is related to computing resources, but not linearly. First, the above equations have only approximate solutions. Secondly, literally until recently, these equations had to be simplified, so that the weather forecast was calculated fairly quickly, otherwise an accurate daily forecast could be considered a couple of years. Thirdly, the accuracy and efficiency of measurements of weather stations and probes will always remain an unresolved problem - all instruments have errors, and it is clear that the data from the probes and weather stations do not come all at once - collecting and normalizing data is also a separate challenge.
However, forecast accuracy is increasing. Atmospheric processes are calculated on a grid superimposed on a map of the country. Even 20 years ago, the step of such a grid was 300 kilometers - one could easily miss a small cyclone. This spring, two weather models for limited territories (they are called mesoscale) began to be calculated at Roshydromet, one of which is a 300-km region over Moscow. The step of such a grid is 3 kilometers, so with such a scale and the availability of computing power it is possible to proceed even to the description of individual physical processes, which, of course, favorably affects the quality of the forecast.
Roman Vilfand, head of Roshydromet: “Now the weather forecast in Moscow can be calculated in 3 km steps, but in general we plan to increase this resolution to 1 km in 3 years. For the Olympics in Sochi, we will strive to describe the processes in half a kilometer, because in the complex orography of Sochi, we need an even more detailed description of the weather than on the plain. For us, Sochi will be a special project. It’s too early to talk about an administrative decision, but as part of our research and other programs, we are preparing for the Olympics. ”
25-27 years ago, the quality of the weather forecast one day ahead was the same as it is now on the third day - a noticeable progress. Modern computing resources allow to increase the period of time for which the forecast is made. Now the forecast in Roshydromet is built for a maximum of six days. The success of such a forecast on the first day is 95-96%, in the forecast for each of the next day the success decreases by 2-3%. By the end of this year, seven-day forecasts will appear in Roshydromet, and then with the growth of computing power it will be possible to increase the forecast period. But not forever. There is a notion of the limit of predictability, it was introduced by the American scientist Edward Lorenz (American mathematician and meteorologist, known as the author of the “butterfly effect”) - according to this limit, the weather can be predicted for two, maximum three weeks in advance. This is due to the inevitable errors in the initial data.
However, with increasing computing power, there are ways to improve the quality of the forecast, even if there are errors in the initial data. If the usual method of making a forecast is called deterministic - that is, once defined, then in recent years the ensemble method of forecasting has gained popularity. To do this, errors in the initial data are artificially “outraged”, and the weather forecast is calculated not once, but 10, 20, 30 - in developed computer centers, ensembles are counted up to 50-60 times. The ensemble forecast allows you to create a "tree of probabilities" - "and what if the weather changes in one or another probable direction?" It is clear that such probabilistic calculations are possible only on the most modern supercomputer equipment, and that's where the margin for growth is absolutely incredible.
Roshydromet employees note that in order to proceed to probabilistic calculations, you need to prepare a consumer who is accustomed to extremely simple, deterministic predictions: “+15, sunny, no precipitation”, leaving no doubt. 40 years ago, probabilistic forecasting methods were introduced in the USA, but the population did not accept them. Now, perhaps due to the informatization of all segments of the population, the reaction to such forecasts is more benevolent.
The competitiveness of domestic meteorology is largely determined by the technical base - in those years when we purchased and installed new equipment, as was the case in the nineties with the Cray Y-MP supercomputer, we were in the top ten of the meteorological services of the planet. As soon as there was a pause in the refurbishment of the Roshydromet computer center, we began to lag behind our foreign colleagues. However, within the World Meteorological Organization, there are a number of standards that provide weather information from around the globe. Each country tries to make its contribution to the Global Telecommunication System, through which meteorologists collect information, because the weather of the neighbors plays a significant role in its own forecast. In the 90s, when, for understandable reasons, the meteorological network began to collapse in our country, American meteorologists instantly felt this - the quality of their forecasts deteriorated noticeably on the 4th ... 5th day - especially if there were no weather data in Siberia. It is clear that the opposite is true: even in the worst years, we always had forecasts from the USA, Great Britain, France and other countries - and there was no sharp jump in quality for the consumer.
This is how Vladimir Antsypovich , head of the Roshydromet computer center, comments on this: “Meteorology is an unusual industry. There are no problems, conflicts and complex relationships between the National Meteorological Services of the World Meteorological Organization. Absolutely everyone is interested in that each national meteorological service works better and the observations are more accurate. ”
Do not hesitate to join the ranks of Intel's readers on Habré.
Thank you for your attention, to be continued.
Successes!
Below is a brief insight into the history of meteorology and how Roshydromet and other modern meteorological services work.
Meteorology has a long history - the first meteorologist, apparently, was Aristotle, who wrote the eponymous treatise on weather changes; in subsequent centuries, observing the weather and the systematization of knowledge about it were engaged in all parts of the planet. However, meteorology as we know it - and this is dynamic meteorology, that is, weather forecasting using mathematical equations of fluid mechanics and thermodynamics - appeared relatively recently.
')
We owe the term “weather forecast” to the hot-tempered captain Robert Hot Coffee Fitz-Roy, who had been swimming with Darwin for almost five years during Beagle’s second voyage in the middle of the nineteenth century. Fitz-Roy discovered that the weather characteristics are related to air pressure, and in his mature years he headed the meteorological department at the Ministry of Commerce, created several models of barometers and arranged data collection from captains who came to the port - in short, he did what all modern meteorologists do.
Modern meteorologists use the equations of the Norwegian Wilhelm Bjerknes, the founder of the Bergen scientific school in meteorology - the solution of the hydrodynamic equations derived by him is nothing more than a weather forecast. These equations found practical application only in the fifties of the last century, when computers began to appear capable of calculating them for a fairly large area - the first weather forecast was considered by American meteorologists on the very first ENIAC computer with the help of John von Neumann - meteorology cannot do without its famous history people.
The first domestic weather calculations began in 1961 on an M20 computer that performed 20,000 operations per second. Opening hours M20 to failure - 15 minutes; the weather model was considered much longer, so the staff of the Hydrometeorological Center had to go for various tricks. The very specifics of weather calculations are such that computer technology is used twenty-four hours a day, seven days a week, all year round — the weather never stops changing, so the Hydrometeorological Center was and remains a serious test area for supercomputers: until the mid-seventies, when domestic computer development Replaced by analogues of machines IBM, Meteorological Center received the first serial numbers of Soviet computers.
Since March of last year, two SGI Altix supercomputer systems with a capacity of 11 and 16 TFlops have been launched at the Hydrometeorological Center - more than a billion times faster than the first M20 computer. Meanwhile, weather forecasts have not become a billion times more accurate since. The accuracy of weather forecasting is related to computing resources, but not linearly. First, the above equations have only approximate solutions. Secondly, literally until recently, these equations had to be simplified, so that the weather forecast was calculated fairly quickly, otherwise an accurate daily forecast could be considered a couple of years. Thirdly, the accuracy and efficiency of measurements of weather stations and probes will always remain an unresolved problem - all instruments have errors, and it is clear that the data from the probes and weather stations do not come all at once - collecting and normalizing data is also a separate challenge.
However, forecast accuracy is increasing. Atmospheric processes are calculated on a grid superimposed on a map of the country. Even 20 years ago, the step of such a grid was 300 kilometers - one could easily miss a small cyclone. This spring, two weather models for limited territories (they are called mesoscale) began to be calculated at Roshydromet, one of which is a 300-km region over Moscow. The step of such a grid is 3 kilometers, so with such a scale and the availability of computing power it is possible to proceed even to the description of individual physical processes, which, of course, favorably affects the quality of the forecast.
Roman Vilfand, head of Roshydromet: “Now the weather forecast in Moscow can be calculated in 3 km steps, but in general we plan to increase this resolution to 1 km in 3 years. For the Olympics in Sochi, we will strive to describe the processes in half a kilometer, because in the complex orography of Sochi, we need an even more detailed description of the weather than on the plain. For us, Sochi will be a special project. It’s too early to talk about an administrative decision, but as part of our research and other programs, we are preparing for the Olympics. ”
25-27 years ago, the quality of the weather forecast one day ahead was the same as it is now on the third day - a noticeable progress. Modern computing resources allow to increase the period of time for which the forecast is made. Now the forecast in Roshydromet is built for a maximum of six days. The success of such a forecast on the first day is 95-96%, in the forecast for each of the next day the success decreases by 2-3%. By the end of this year, seven-day forecasts will appear in Roshydromet, and then with the growth of computing power it will be possible to increase the forecast period. But not forever. There is a notion of the limit of predictability, it was introduced by the American scientist Edward Lorenz (American mathematician and meteorologist, known as the author of the “butterfly effect”) - according to this limit, the weather can be predicted for two, maximum three weeks in advance. This is due to the inevitable errors in the initial data.
However, with increasing computing power, there are ways to improve the quality of the forecast, even if there are errors in the initial data. If the usual method of making a forecast is called deterministic - that is, once defined, then in recent years the ensemble method of forecasting has gained popularity. To do this, errors in the initial data are artificially “outraged”, and the weather forecast is calculated not once, but 10, 20, 30 - in developed computer centers, ensembles are counted up to 50-60 times. The ensemble forecast allows you to create a "tree of probabilities" - "and what if the weather changes in one or another probable direction?" It is clear that such probabilistic calculations are possible only on the most modern supercomputer equipment, and that's where the margin for growth is absolutely incredible.
Roshydromet employees note that in order to proceed to probabilistic calculations, you need to prepare a consumer who is accustomed to extremely simple, deterministic predictions: “+15, sunny, no precipitation”, leaving no doubt. 40 years ago, probabilistic forecasting methods were introduced in the USA, but the population did not accept them. Now, perhaps due to the informatization of all segments of the population, the reaction to such forecasts is more benevolent.
The competitiveness of domestic meteorology is largely determined by the technical base - in those years when we purchased and installed new equipment, as was the case in the nineties with the Cray Y-MP supercomputer, we were in the top ten of the meteorological services of the planet. As soon as there was a pause in the refurbishment of the Roshydromet computer center, we began to lag behind our foreign colleagues. However, within the World Meteorological Organization, there are a number of standards that provide weather information from around the globe. Each country tries to make its contribution to the Global Telecommunication System, through which meteorologists collect information, because the weather of the neighbors plays a significant role in its own forecast. In the 90s, when, for understandable reasons, the meteorological network began to collapse in our country, American meteorologists instantly felt this - the quality of their forecasts deteriorated noticeably on the 4th ... 5th day - especially if there were no weather data in Siberia. It is clear that the opposite is true: even in the worst years, we always had forecasts from the USA, Great Britain, France and other countries - and there was no sharp jump in quality for the consumer.
This is how Vladimir Antsypovich , head of the Roshydromet computer center, comments on this: “Meteorology is an unusual industry. There are no problems, conflicts and complex relationships between the National Meteorological Services of the World Meteorological Organization. Absolutely everyone is interested in that each national meteorological service works better and the observations are more accurate. ”
Do not hesitate to join the ranks of Intel's readers on Habré.Thank you for your attention, to be continued.
Successes!
Source: https://habr.com/ru/post/98506/
All Articles