Earthquake prediction with artificial intelligence
Published on February 15, 2023
Earthquake prediction with artificial intelligence seemed impossible a few years ago. Today it seems to be closer to reality.
We all know about the recent catastrophe in Turkey. A territory shaped by tectonics, wrinkled and crushed by colossal forces, has once again suffered the relentless effects of the energy that gives life to our planet.
Although we understand that earthquakes are a fundamental part of our planet’s processes, we must also face their consequences and devastation. Unfortunately, we don’t know of a reliable method for predicting when they will occur or with what intensity. But what if artificial intelligence could generate those analytics for us?
Outdated earthquakes
Our main obstacle in understanding seismology is the lack of information. Some of the world’s worst earthquakes, such as China 2008, Haiti 2010, and Japan 2011, occurred in areas that were considered safe. Other devastating earthquakes have come from unmapped fault lines. Clearly the risk maps are imprecise when estimating the affected areas.
Another popular method of calculating the risk of an earthquake occurring is the use of probabilities. Derived from complex mathematical formulas, averages are obtained to determine how often intense earthquakes occur in a certain area. As an example, the last big earthquake on the San Andreas fault was in 1857. Since the average interval between big earthquakes is 135 years, the media commonly say that we are already “overdue” and that the big earthquake is imminent. But the intervals between earthquakes are varied, from 44 to 305 years, so the next big one could happen today or a century from now. The system of probabilities then becomes quite imprecise and even alarmist.
Applying intelligence to tectonics
One proposed method for applying artificial intelligence to earthquake prediction involves neural networks. This system is inspired by the functioning of the human brain, where a network of neurons represents interconnected information. By making connections between the data, the system “learns”, reinforces the useful connections that can change as information is added.
In a method comparable to natural language processing, these models can predict friction on a geological fault and how the fault will break in the future. As? Well, in the same way that language is processed, through sound. A geological fault “speaks”, emits a sound or vibration that is picked up by seismograms for study. For now, these models are in the process of collecting the current sound to generate an accurate picture of the behavior of the failure, before launching into a prediction.
Other models try to find the relationship between earthquakes and other phenomena, such as the number of electrons (electrically charged subatomic particles) present in the atmosphere near a rupture zone. Whichever approach is used, the challenge has two edges: having a network for capturing data through sensors that is wide and dense enough to feed the model with data, and a model capable of learning efficiently.
The implications are powerful. Are we close to achieving the long-awaited earthquake prediction? Can we generate a reliable alert avoiding false alarms or panic? Or will it continue to be the domain of astrologers and seers? It is surely only a matter of time before we have the first reliable prediction.
Do you want to know how much your Artificial Intelligence project costs?