Earthquakes are one of the most destructive natural disasters in human history. Therefore, it is important to predict when earthquakes will occur, but it is also difficult. Geophysicist Paul Johnson has developed a method in this regard. The earthquake is a natural disaster that has caused great devastation throughout human history and is still one of the elements that threaten humanity. Geophysicist Paul Johnson, one of the names interested in the earthquake, focused on a new earthquake prediction study. Johnson’s team uses a machine learning system that tries to illuminate the physics of earthquakes and clarify the warning signs of possible earthquakes. Two years ago, he and his colleagues successfully predicted the earthquake in the model laboratory system, using a model similar to the algorithms behind image and voice recognition and other forms of artificial intelligence.
A survey was published this week at arxiv.org. Johnson and his team reported that they tested their algorithms on slow-shift earthquakes in the Pacific Northwest. According to Johnson, the algorithm was able to predict the onset of a slow-shift earthquake within a few days. Maarten de Hoop, an earthquake scientist at the University of Rice, who was not involved in the study, said it was a truly exciting development, and noted that progress was made for the first time. Mostafa Mousavi, a geophysicist at Stanford University, said he found the results “exciting and motivating.. Hold and release
Although various scientists and geophysicists have explained that earthquakes cannot be predicted, Johnson knows that earthquakes are a physical process, no difference from the collapse of a dying star or the change of wind. He prioritizes his understanding of fault physics and does not shy away from the problem of prediction. More than a decade ago, Johnson began working on la laboratory earthquakes made with floating blocks separated by thin layers of particulate material. Just like in tectonic plates, the blocks are shaking in a beautiful way, and the tangential stress is slipping when it is large enough. This shift in the laboratory experiment releases tension and the hold-release cycle restarts.
Johnson and his colleagues noticed a sharp rise before each release while recording the acoustic signal emitted during this hold-release cycle. This pioneering event was caused by pioneering shocks before the earthquake. However, earthquake scientists were having trouble translating these pioneering shocks into predictions about when the original earthquake would occur. We were in some sort of dead end, I couldn’t see any way forward, Johnson says. Several years ago, at a meeting in Los Alamos, Johnson explained this dilemma to a group of theorists. Theorists recommended that they re-analyze their data using machine learning. Together, scientists formulated a plan. During each experiment, they would take about five minutes of audio recordings and divide them into many small pieces.
For each track, researchers calculated more than 80 statistical properties. These included low signals, types of low signals, and whether the fragment contained information about the pioneering event. Researchers used a machine learning algorithm known as a forest random forest to systematically look at the combination of strongly bound features and the amount of time remaining before destruction. The algorithm then began to estimate the destruction time based on acoustic emission characteristics only. After successful tests and analyzes, Johnson had to experiment with real devastations to show that machine learning can predict real earthquakes. The best place for this was the Pacific Northwest.
