Using AI to Detect Earthquakes
Earthquakes are one of the most devastating natural phenomena that occur more commonly than we think. In fact, according to the US Geological Service, about 20,000 earthquakes occur every year or approximately 55 per day. Given these numbers, you may be wondering: “If so many earthquakes happen every year, how come we don’t hear about them or feel them”? The answer is that some of the movements are so tiny that we may barely feel them and they usually go unnoticed. However, even these subtle motions can predict more massive earthquakes that can cause a lot of destruction.
Researchers are now using AI to analyze seismographic charts which can better detect the earth’s movement, but it also requires significant data annotation work to be done. Let’s take a look at how AI can be used to predict earthquakes and also how the training datasets need to be prepared.
How Does the Old Method of Detection Compare With the New One?
Before AI came along, researchers needed to take recorded earthquake signals and verify that decades-old algorithms had detected true earthquakes rather than tremors generated by ordinary things like crashing waves, passing trucks, or stomping football fans. This process involves somebody sitting in front of a computer screen and observing all of the waves on the charts. Thanks to AI, not only can this process be automated, but we can also improve our ability to detect and locate very small earthquakes that would go unnoticed by humans researchers. This would allow us to get a clearer view of how earthquakes interact or spread out along the fault, how they get started, even how they stop.
In order to determine the location and magnitude of an earthquake, researchers look for two types of waves, according to Stanford University. First, there are P waves that advance quickly which push, pull, and compress the ground as they move along. Then there are S waves that travel more slowly but can be more destructive as they move the Earth side to side or up and down. Last year, researchers created such an AI system called the Earthquake Transformer which is an attentive deep-learning model for simultaneous earthquake detection and phase picking. In order to train the system, geologists used 5 weeks of continuous waveforms recorded in Japan.
The results were very impressive. The model detected and located 21,092 events – more than two and a half times the number of earthquakes picked out manually. It is also worth noting how massive the training dataset was. One million hand-labeled seismographs recorded mostly over the past two decades were used. Even though the training dataset was massive, researchers will need additional data to train the system which will also need to be annotated. Mindy Support can source and recruit candidates with the necessary knowledge, education, and experience to perform such annotation work and in many other industries as well.
Challenges Presented to the Earthquake Transformer
One of the biggest challenges is to identify false positives. For example, cultural noise, such as large gatherings of people, produces seismic waves that have a similar impulsive nature and frequency range that an earthquake wave would have especially when a short window around the arrival is used. Also, there are some issues in picking up P waves recorded at larger epicentral distances. However, both of these issues could be solved by having more training data which could also be useful in picking up small activities.
The good news is that deep-learning models trained by a dataset in a specific region can generalize well to other regions. This could ease the burden of obtaining new datasets of each region and having to study those separately to predict earthquakes in every part of the globe. This means that researchers can focus more attention on adjusting the network type for example using recurrent instead of convolutional or using a different training process to create a more accurate system.
Mindy Support Provides Comprehensive Data Annotation Services
Mindy Support has extensive experience actualizing data annotation projects in a wide variety of industries and complexities. We are one of the largest BPO providers in Eastern Europe with more than 2,000 employees in six locations all over Ukraine. Our size and location allow us to source and recruit the necessary number of candidates quickly and we can also scale your team quickly without sacrificing the quality of the annotation. Contact us today to find out how we can help you.
February 10th, 2021 Mindy News Blog
Talk to our experts about your AI/ML projectContact us