AI strips out city noise to improve earthquake monitoring systems

Chris Potter-Walker

It is possible that a deep learning algorithm can remove city noise from earthquake monitoring tools.

Gregory Baroza says that earthquake monitoring in urban settings helps us understand the fault systems that underlie vulnerable cities.

It is difficult to discern the underground signals that indicate an earthquake is happening because of the noise of cities.

To improve our ability to locate earthquakes, Baroza and his colleagues trained a deep neural network to distinguish between earthquake signals and other noise sources.

Over 80,000 samples of urban noise and 33,751 samples of earthquake signals were combined to test the neural network. The earthquake signals were taken from the rural area around San Jacinto, California, while the noise samples came from Long Beach.

The signal to noise ratio was improved by 15 decibels when running audio through the neural network.

The research is very useful for the field, says Maarten de Hoop at Rice University in Houston, Texas.

The neural network was trained on data labelled by humans, but the readings were all from one area. It is less likely that the model will work when presented with noise from other places.

De Hoop says that the holy grail in this field is supervised learning.

Baroza is unsure of how well the model would work in other places.

Science Advances is a journal.