Machine learning to predict earthquakes

Table of Contents

  1. Introduction
  2. Existing methods of dealing with earthquake events
  3. Machine learning to help scientists better understand earthquakes
  4. Chinese Researchers developing EarthX software for detection
  5. Conclusion
  1. Introduction

Earthquakes represent an extremely dangerous manifestation of Mother Nature in our daily lives. It is a reminder that we live on top of a dangerous system of rock plates that are dictated by complicated plate tectonics.

One of the main reasons that earthquakes remain so scary to experts is primarily since we are unable to get sensing data at those depths. The stresses occur at faults that are hundreds of feet beneath the earth. If we could get that data, perhaps we could do better at predicting the onset of a dangerous plate shift, but for now we are left to trying to forecast these events.

Earthquakes typically happen when rock breaks near a fault – the boundaries between large plates. The rock breaking releases a high amount of energy that then creates the quaking movement of the earth beneath us. Earthquake detection has notoriously been very difficult. A Los Alamos National Lab seismologist and others state that they are much more hopeful that earthquake detection will come to fruition thanks to machine learning.

  1. Existing methods of dealing with earthquake events

While detection is difficult, there are ways that we currently use to try and forecast the events to save lives and property from being damaged.

There is a system in place on the United States West Coast for example called ShakeAlert that can alert civilians that an earthquake is occurring. However, they only send alerts after an earthquake has already started and can’t give warning more than a day in advance. Ideally, we want as much notice as possible to save human lives.

The United States Geologic Service (USGS) can predict the probabilities that an earthquake will occur over a certain time period in a particular location according to their website. However, this is a far cry from being able to state definitively where and when an earthquake will take place. Knowing that an earthquake will occur sometime over the next one hundred years is not exactly useful information – humans will need to carry on with their everyday lives and hope that when disaster does strike, they can clear out of dodge. Clearly, a system for getting better at detecting earthquakes is certainly needed.

Many seismologists are pessimistic about using machine learning for earthquake prediction citing that it is easy to find a trend in the data going backwards but impossible to find a reliable trend going forwards. In a Smithsonian story, Robert Geller, who is a seismologist at the University of Tokyo states that there has been no reliable precursor in the data found to date.

  1. Machine learning to help scientists better understand earthquakes

A wide range of geographic events rely on tracking seismic waves. The waves occur at a variety of frequencies and are often dominated by significant amounts of noise which obscures the signals that the scientists are trying to see.

A new approach by MIT researchers could allow geologists to track low-frequency seismic waves. The scientists trained a neural network on hundreds of different earthquakes that they were able to simulate. In doing so, the model could mimic the different characteristics of the waves that are a result of the earthquake and subsequently predict the waves occurring at low frequencies that had gone missing. This approach allowed the researchers to artificially create the low-frequency waves that are so often missing in the noisy seismic wave data.

The researchers detail that their model has learned the different relationships between all the frequencies present in an earthquake. When the model is fed a partial profile of an earthquake, the model can fill in the gaps which will allow geologists to have access to a more complete picture of earthquakes.

Through simulating nine different earths and thirty earthquakes on each Earth, the researchers have a library of different seismic signatures. After the training phase, the team can provide the model only a partial high-frequency part of the seismic data. The trained model can then attempt to estimate the missing pieces.

4. Chinese Researchers developing EarthX software for detection

Research into using artificial intelligence to predict earthquakes is not taking place just in the United States. China is also significantly investing into technology that will allow them to more preemptively react to earthquakes that have claimed the lives of many citizens in the past.

Chinese researchers have created a system called EarthX that will attempt to automate the entire earthquake detection process. The software will analyze the location, depth, and magnitude of the earthquake. The program was created as a joint effort between the University of Science and Technology of China (USTC) and China Earthquake Administration (CEA). Like the work by MIT scientists, the software uses a convolutional neural network to analyze the waveforms that compose the earthquakes to make their predictions.

The software will be used for a year as a trial in a small number of provinces and if it demonstrates true value, will be rolled out to all provinces in the country.

5. Conclusion

Artificial intelligence is offering help to a field that has a tough task in front of them – earthquake detection. Analysis of the complex waveforms that are generated from plates shifting underground is complicated and offers scientists lots of data. Continued progress into this field is of extreme importance as currently we have essentially no predictive tools being utilized, as least in the United States. Development of this type of preventative software will give us the warning that humans need to clear out of harms way and save lives.

Show More

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button