AI and Earthquake Predictions

- Overview

According to the U.S. Geological Survey (USGS), no one has ever predicted a major earthquake, and scientists don't expect to be able to in the near future. Scientists can calculate the probability of a significant earthquake occurring in a specific area within a certain time, but they can't predict exactly when or where it will occur, or how large it will be.

Some people claim to predict earthquakes, but their predictions are not based on scientific evidence. For example, they might predict a M4 earthquake somewhere in the U.S. in the next 30 days, but their predictions are often so general that there will always be an earthquake that fits. 

Seismologists can estimate where earthquakes may be likely to strike by calculating probabilities and forecasts, including:

• The average rate of past seismic activity in a region
• Digging trenches to examine the geological record for earthquake ruptures that occurred in ancient history

While most scientists are pessimistic, some hold that prediction might be possible if you take into account non-seismic precursors and have enough resources to study them extensively. 

The U.S. Geological Survey's National Seismic Hazard Map shows the strength of ground shaking that has a 1 in 50 chance of being exceeded in a particular place in the lower 48 states over a period of 50 years.

- AI in Earthquake Predictions

According to a seven-month trial in China, an AI-driven earthquake forecasting tool is 70% accurate at predicting earthquakes a week in advance. The AI is trained to identify statistical differences in real-time seismic data that have been paired with previous earthquakes.

The AI-powered earthquake forecasting tool uses a combination of advanced sensory technology and extensive training from AI and human earthquake experts. The tool successfully predicts 14 earthquakes within 320 kilometers of their estimated location and at the calculated strength. However, it also issues eight false warnings and misses one earthquake. 

AI-based techniques used in earthquake prediction include:

One machine learning (ML) method uses only the vibrations coming from the fault, unlike many methods humans use to forecast earthquakes. What human researchers had discounted as low-amplitude noise turns out to be the signal that allowed ML to make its predictions.