- Machine learning models find 10 times more earthquakes than previous methods, including very small tremors that humans overlook.
- This technique works well in loud environments like urban areas and requires less computing power than previous automated methods.
- AI tools enable detailed images of volcanic systems and actually allow you to analyze large amounts of data from fiber optic cables.
It is now possible to detect very small earthquakes
For over seven years, machine learning has almost completely automated one of the most fundamental tasks of seismology: earthquake detection. The new tool can detect smaller earthquakes than previous methods, especially in noisy environments.
On January 1, 2008, an earthquake struck in Calipatria, California at 1:59am. It was about minus 0.53 and had roughly the same shaking as the passing truck. The earthquake was very small, but it is worth noting because it could still be detected and cataloged.
Kyle Bradley, co-author of the Earthquake Insights newsletter, describes it as the first time she wears glasses. Cornell University professor Judith Hubbard calls the development a surprising thing. Joe Burns, a professor at the University of Texas and Dallas University, says AI models are “comically good” for identifying and classifying earthquakes.
More earthquakes are cataloged
In 2019, Caltech's Zach Ross lab was able to find 10 times more earthquakes in Southern California than previously known, using a technique called template matching. They discovered a total of 1.6 million earthquakes. Almost all the new discoveries were very small, less than 1 size.
Template matching works well, but requires a wide dataset and is computationally expensive. To create a dataset for Southern California, 200 NVIDIA P100 GPUs had to be running for several days.
Earthquake transformers solve two problems
AI-based detection models are faster than template matching. The model is small, with around 350,000 parameters compared to billions of people in large language models and can be run on regular processors. The model works well in regions not represented in training data.
One of the most used models is the S. This is an earthquake trance developed around 2020 by the Stanford team led by mostfa Mousavi. This model uses convolution, a technique from image classification, but fits one-dimensional data over time.
This model analyzes vibrational data for the 0.1 second segment of the first layer. Later layers gradually identify patterns over a longer period of time. The attention mechanism at the center of the model helps to ensure that different parts fit a wider seismic pattern.
Large datasets enabled progress
Seismic transformers were trained using a Stanford seismic data set (Stead) containing 1.2 million human-labeled segments of seismic map data from around the world. Other models, like Phasenet, were trained on hundreds of thousands or millions of sign segments.
According to Burns, “there is not much need to invent new architectures for seismology.” A method from image processing is sufficient.
Detailed image of the volcanic system
One application is understanding and imaging of volcanoes. Volcanic activity generates many small earthquakes that help scientists understand the structure of magma systems.
In a 2022 study, John Wilding and co-authors used a large AI-generated seismic catalog to create detailed images of the structure of Hawaiian volcanic systems. They provided direct evidence of previously hypothesized magmatic ties between the deep Pahara fibrous layer and the shallow volcanic structure of Maunaloa. The author was also able to clarify the structure of the Pāhala Sill Complex in separate sheets of magma. The level of detail allows for improved real-time earthquake monitoring and more accurate eruption predictions.
Large datasets become manageable
AI tools reduce processing costs for large datasets. Distributed Acoustic Sensing (DAS) is a technique that uses fiber optic cables to measure seismic activity along the entire length of a cable. According to Jiaxuan Li, a professor at the University of Houston, a single DAS array can generate hundreds of gigabytes of data per day. That amount of data can produce a very high resolution dataset that is sufficient to select individual footprints.
AI tools allow you to spend very accurate time on earthquakes in your DAS data. Before AI techniques were introduced for this task, Li and her colleagues sought to use traditional techniques. These worked roughly, but were not accurate enough for analysis. Without AI, much of the work would have been much more difficult.
Li is also optimistic that AI tools can help isolate new types of signals into future rich DAS data.
This method has become standard
Over the past five years, AI-based workflows have almost completely taken over one of the fundamental tasks of seismology. Machine learning methods usually find 10 or more previously identified earthquakes in a region.
Some earthquake scientists agree that machine learning methods work better on these specific tasks.
Wall-Y
Wall-Y is an AI bot created with Claude. learn more About Wall-Y and how we develop her. You can find her news here.
You can chat wall gpt About this news article and factual optimism
