Main menu


Artificial intelligence (AI) predicts micro-earthquakes

Artificial intelligence (AI) predicts  micro-earthquakes

Artificial intelligence (AI) predicts even this... Lightning, hail, and even micro-earthquakes

AI, after matching the current weather with past cases, the lightning location prediction
neural network reads the patterns of the predictors and calculates the probability that the storm will hail.
A large earthquake warning through machine learning and fine earthquakes learned from past data.
A surprise attack from natural disasters that strike without notice

Natural disasters, which have become stronger due to climate change, such as storms, lightning, floods, tornadoes, and hail, are making the global village difficult. Is there really no way to stop the tyranny of this natural disaster?

Experts say that the only way to prevent natural disasters is to predict. By predicting natural disasters and issuing warnings in advance, it saves time to evacuate residents. To do this, clear prediction technology is needed, and AI is at its core.

Statistically, lightning, which takes the lives of about 20 people in the United States each year, is also a representative natural disaster tyranny. However, increasingly sophisticated AI technology has now reached the level of predicting sudden lightning strikes.

On the 9th, Spectrum News 1 introduced a method of predicting lightning using an AI program by a research team at the University of Wisconsin led by Professor John Citino.

A machine learning algorithm developed by Professor John Citino is known to accurately predict the location of lightning up to an hour in advance.

A must-have for this lightning prediction technology is a weather observation satellite. In 2016, the National Oceanic and Atmospheric Administration (NOAA) sent the 'GOES-R' satellite into space, and one of the numerous sensors on this satellite is the Geostationary Lightning Mapper (GLM).

Records of lightning data collected by NOAA's GLM sensors are fed into an AI algorithm, which notifies the computer of what atmospheric patterns occur before and when lightning strikes.

After learning this, the AI ​​program looks at the current weather, matches it to past events, and then predicts where the lightning will strike.

Scientists can now receive lightning information in near real time from data such as cloud images and other weather variables. Meteorologists can use this information to detect severe storms in time and warn the public in advance of a lightning strike.

They also use this data to track severe storms, such as tornadoes, because lightning strength is a key indicator of tornado formation.

The data accumulated in this way is stored in a database and becomes the key to predicting where lightning will strike, the research team said.

However, the research team argues that this technique has limitations. In other words, lightning strikes on specific buildings and on specific trees are unpredictable. The data provided by AI is the lightning prediction area and the rate of occurrence of lightning.

In an experiment conducted near Washington, DC in July of this year, although there was a blue sky overhead, the AI ​​program predicted a 25% chance of a lightning strike in the area 50 minutes before, and raised it to 75% about 30 minutes later. . Then, about 25 minutes later, the GLM sensor observed lightning.

"This program is still experimental, but it could be widely used in the near future," the researchers said.

AI neural network calculates the probability of hail

Have you ever been struck by hailstone the size of a golf ball that is falling from the dark sky at high speed? Scientists say that unless you wear a bulletproof helmet, you can get fatal injuries to the head.

Hail is a phenomenon in which the updraft of a thunderstorm carries water droplets, and when the frozen water droplets become heavier than the updraft, they fall toward the ground.

Hail, falling at a rate of 193 kilometers per hour, causes up to $10 billion in damage each year in the United States. However, if these hail storms can also be predicted, the damage can be minimized.

According to NCAR & UCAR News on May 25, this year, one of the fields of interest of the National Center for Atmospheric Research (NCAR) is the prediction of natural disasters such as hail and tornadoes using artificial intelligence AI .

One problem, however, is that these two weather models are notoriously difficult to predict because they are based on storms.

However, as a result of using the predictive model based on the machine learning neural network, the research team showed much better predictive power than the prediction based on the traditional model calculation.

"Neural networks were able to predict when and where a severe storm hazard could occur more adeptly," said team leader Dr. Ryan Sobashi. Not only that, we better predict whether it will be hail or wind.”

According to the scientists, for a weather model to capture a thunderstorm, it must operate at very high resolution, capable of capturing not only storm air currents, but also microscopic atmospheric phenomena. In addition, an interval of 4 km or less is required between grid (coordinate) points inside the model.

However, at this resolution, the weather model failed to capture storms occurring at much smaller scales, including hail and tornadoes.

Therefore, meteorologists have used a lot of model data called updraft helicity, which measures the rotation of a storm to determine the likelihood of a serious storm.

Nevertheless, helicity also had the disadvantage of not being able to capture the severe weather caused by straight-line storms such as 'derecho', which inflicts enormous damage on the Middle East of the United States.

On the other hand, the AI ​​neural network used in NCAR's new forecast included about 40 other factors as predictors, including updraft helicity, as well as storm location, time, dew point, wind speed, and surface pressure.

This allows the neural network to read patterns in these predictors and calculate the probability that the storm will produce hail, tornadoes, or strong storms.

In the end, the predictive power generated by the neural network shows the likelihood of a storm hazard forming within 40 or 120 km of the individual coordinate points of the model, explains Sovash.

Neural networks are better able to predict the risks associated with straight storms than forecasts based on updraft helicity, and have improved forecasts for areas where supercells (thunderstorms) are not likely to form, particularly in the outskirts of the Midwest.

"Neural networks suggest that machine learning can be a useful tool for disaster prediction," Sovash said.

AI catches small earthquakes and warns of large earthquakes

On October 21, last year, Stanford University News reported that AI could detect hidden micro-earthquakes and warn of large earthquakes.

Mousavi, a researcher at Stanford Earth's School of Environmental Sciences, and his co-authors, in a recent paper published in Nature Communications, developed a new way to use AI to focus on the planet's millions of subtle changes.

Microscopic movements in the Earth's outermost layers, they explain, are the touchstone for deciphering warning signs of large earthquakes. As a related example, in October 1989, the Rome Prieta earthquake that severely shook San Francisco and the Gulf of Monterey was mostly micro-earthquakes that occurred in previously unknown faults.

However, as artificial intelligence AI  algorithms become more sophisticated, Dr. Musabi explains that it is possible to learn such fine earthquakes from past records.

“Big earthquakes are hard to miss, but they are rare. On the other hand, earthquakes that are too small to detect always occur, and these micro-earthquakes, which occur in the same fault as large earthquakes, provide reliable information about earthquakes.”

Stanford geophysicist Dr Gregory Veroza, one of the paper's authors, said: "By detecting and locating these small earthquakes, we can more clearly see how earthquakes spread along faults, how they stop, and more. ” he said.

After examining seismographs every day, Dr. Musabi started developing technology for automating seismic detection. He was able to solve this problem by entering Dr. Veroza's lab and refining machine learning.

Afterwards, they developed a new model called an earthquake transformer that detects small earthquakes caught as weak signals, converts them into a database, and uses this accumulated data to determine the exact timing of earthquake phases.

Professor Veroza said, "Through improved monitoring of small earthquakes, the more deep and three-dimensional fault structures we identified, the better we were able to predict future latent earthquakes."

In other words, understanding the patterns of small oscillations accumulating over decades or centuries is a way to minimize surprise and damage when larger earthquakes occur.