With four basic data collected by any weather station , a team of Swiss scientists is able to anticipate where and when lightning will fall . For this they developed an artificial intelligence system that learned to make their predictions by combining those four parameters with the historical two decades of downloads. Although they still have to fine tune the location, the machine can predict most of the twinkles half an hour before they fall.
Rays are one of the most complex, most studied and least known phenomena of meteorology. In essence it involves discharges of a large amount of energy from electric fields generated during storms. Although they have long ceased to be a problem in populated areas, they still have a great impact on natural environments or for specific activities such as air navigation , wind installations or the distribution of electricity. Until now, lightning prediction systems relied on observations from the satellite, airborne systems or models supported by sensors that record the magnitude of the electrostatic field from the ground. Technologies all very expensive and not excessively accurate.
Now, a team of researchers from Swiss and British universities have designed an artificial intelligence system that only needs the parameters of four data that it records to the humblest weather station: air temperature two meters high, wind speed, atmospheric pressure at the height of the station and the relative humidity of the air.
Atmospheric pressure, relative humidity, temperature and wind speed anticipate lightning strikes
To train the machine, they compiled all the records of 12 stations distributed by Switzerland since 2006 together with the rays detected in their environment. On that huge database, they developed an algorithm that looked for correlation patterns between the four parameters and the lightning strike. Through machine learning, his model was connecting the variations in the four parameters produced every 10 minutes with the immensely lower incidence of electric shocks. Thus the artificial intelligence system learned to anticipate where and when a new spark would fall. And they tested it with two of the stations that had more and better data.
"We consider three time ranges of anticipation: from 0 to 10 minutes, 10 to 20 minutes and 20 to 30 minutes. The average probability of detection between the 12 stations was 78%, 78% and 76% respectively," he says in an email the researcher of the electromagnetic compatibility laboratory of the Federal Polytechnic School of Lausanne (Switzerland) and principal author of the mill, Amir Mosajabai.
Where was less accurate. They designed the system to anticipate the fallen rays within a radius of 30 kilometers around each station. "It is true that, for some uses, the distance is large," Mosajabai acknowledges, but says it is only for lack of more information. "The data of the rays we had were those of the range of three kilometers or those of the range of 30 kilometers" and did not have intermediate distance records.
The authors of the work published in Climate and Atmospheric Science , compared their system with three of the predictive models currently used in two of the stations that had vertical electrostatic field sensors with older records. Except against one of these models and in the anticipation of 10 minutes, in the rest of the combinations, the predictive power of their artificial intelligence was superior.
The system anticipated 78% of downloads 30 minutes before they occurred
"It must be recognized that the performance seems really good (high detection rate, very low false positive ratio, less than 10%), but they achieved it in two stations, one of a mountainous area and another high mountain, respectively," says in an email the researcher of the Ray Research Group (LRG) of the Polytechnic University of Catalonia, Oscar Van der Velde.
For there to be a thunderstorm, Van der Velde recalls that three things are needed: "A strong vertical temperature gradient, enough water vapor and a mechanism to raise the air. In the mountains, the third requirement is met very well by the winds that rise up the slope due to solar daytime heating. In remote areas of the mountains, the yield will probably be lower, since it is more difficult to capture the lifting mechanism using a single weather station. "
For Glòria Solà de las Fuentes, a scientist at the Fulgura high-precision ray location company , "there are still many mysteries around them and the main one is where they will fall." In Fulgura, which have large infrastructure companies among their clients, they use sensor networks placed at high points to capture the number of discharges, their current or power. With that information they create their ray maps. As for the study, he believes that its results could be very useful where there are no radars or sensors, such as in tropical regions, "where they do not have these detection systems but meteorological stations do."
Source: https://elpais.com/elpais/2019/11/14/ciencia/1573713534_577460.html
Rays are one of the most complex, most studied and least known phenomena of meteorology. In essence it involves discharges of a large amount of energy from electric fields generated during storms. Although they have long ceased to be a problem in populated areas, they still have a great impact on natural environments or for specific activities such as air navigation , wind installations or the distribution of electricity. Until now, lightning prediction systems relied on observations from the satellite, airborne systems or models supported by sensors that record the magnitude of the electrostatic field from the ground. Technologies all very expensive and not excessively accurate.
Now, a team of researchers from Swiss and British universities have designed an artificial intelligence system that only needs the parameters of four data that it records to the humblest weather station: air temperature two meters high, wind speed, atmospheric pressure at the height of the station and the relative humidity of the air.
Atmospheric pressure, relative humidity, temperature and wind speed anticipate lightning strikes
To train the machine, they compiled all the records of 12 stations distributed by Switzerland since 2006 together with the rays detected in their environment. On that huge database, they developed an algorithm that looked for correlation patterns between the four parameters and the lightning strike. Through machine learning, his model was connecting the variations in the four parameters produced every 10 minutes with the immensely lower incidence of electric shocks. Thus the artificial intelligence system learned to anticipate where and when a new spark would fall. And they tested it with two of the stations that had more and better data.
"We consider three time ranges of anticipation: from 0 to 10 minutes, 10 to 20 minutes and 20 to 30 minutes. The average probability of detection between the 12 stations was 78%, 78% and 76% respectively," he says in an email the researcher of the electromagnetic compatibility laboratory of the Federal Polytechnic School of Lausanne (Switzerland) and principal author of the mill, Amir Mosajabai.
Where was less accurate. They designed the system to anticipate the fallen rays within a radius of 30 kilometers around each station. "It is true that, for some uses, the distance is large," Mosajabai acknowledges, but says it is only for lack of more information. "The data of the rays we had were those of the range of three kilometers or those of the range of 30 kilometers" and did not have intermediate distance records.
The authors of the work published in Climate and Atmospheric Science , compared their system with three of the predictive models currently used in two of the stations that had vertical electrostatic field sensors with older records. Except against one of these models and in the anticipation of 10 minutes, in the rest of the combinations, the predictive power of their artificial intelligence was superior.
The system anticipated 78% of downloads 30 minutes before they occurred
"It must be recognized that the performance seems really good (high detection rate, very low false positive ratio, less than 10%), but they achieved it in two stations, one of a mountainous area and another high mountain, respectively," says in an email the researcher of the Ray Research Group (LRG) of the Polytechnic University of Catalonia, Oscar Van der Velde.
For there to be a thunderstorm, Van der Velde recalls that three things are needed: "A strong vertical temperature gradient, enough water vapor and a mechanism to raise the air. In the mountains, the third requirement is met very well by the winds that rise up the slope due to solar daytime heating. In remote areas of the mountains, the yield will probably be lower, since it is more difficult to capture the lifting mechanism using a single weather station. "
For Glòria Solà de las Fuentes, a scientist at the Fulgura high-precision ray location company , "there are still many mysteries around them and the main one is where they will fall." In Fulgura, which have large infrastructure companies among their clients, they use sensor networks placed at high points to capture the number of discharges, their current or power. With that information they create their ray maps. As for the study, he believes that its results could be very useful where there are no radars or sensors, such as in tropical regions, "where they do not have these detection systems but meteorological stations do."
Source: https://elpais.com/elpais/2019/11/14/ciencia/1573713534_577460.html
Comments
Post a Comment