 "Can we predict earthquakes? China says it has found an AI-powered solution")
After decades of trial and error in earthquake prediction around the world, China claims it has made a significant breakthrough by developing its first comprehensive artificial intelligence model, designed to analyse extensive datasets and predict earthquakes, according to a report by South China Morning Post (SCMP).
The model, named ‘DiTing’, utilises one of the world’s largest datasets in its field. It can detect earthquake signals, monitor seismic activity, and aid in rapid earthquake response, according to its creators.
DiTing will use vast data from the China Earthquake Observation Network and using cutting-edge AI technology, it will substantially enhance the accuracy and speed of detection of seismic signals, the report said citing the deputy director of the China Earthquake Administration’s Institute of Geophysics, during the model’s unveiling in Chengdu on Sunday.
Incidentally, the launch coincided with the 48th anniversary of the catastrophic Tangshan earthquake in northern China, which resulted in at least 300,000 fatalities and destroyed the city.
This initiative is collaboration between the National Supercomputing Centre in Chengdu, the Institute of Geophysics, and Tsinghua University.
The model is named after a mythical creature in Chinese Buddhism known for its ability to detect signals from across the universe.
More From This Section
What have been the challenges of earthquake prediction?
Since the 1960s, geologists have been attempting to predict earthquakes using modern scientific methods, but success has been limited. But the complexity of the Earth’s fault systems and the constant background seismic noise make it challenging to identify clear signals. The United States Geological Survey has noted that accurate earthquake predictions require knowledge of the event’s location, timing, and magnitude—none of which can currently be predicted with certainty.
Instead, geologists create ‘hazard maps’ that estimate the probability of an earthquake occurring within a specific timeframe, which aids in planning and improving building standards. However, these maps do not provide the early warnings necessary for immediate evacuation or shelter. Moreover, not all residents in earthquake-prone areas can afford the infrastructure needed to withstand significant seismic activity.
Researchers continue to seek more precise prediction methods, exploring various indicators such as animal behaviour and electrical disturbances in the Earth’s upper atmosphere. Recently, the focus has shifted to the potential of artificial intelligence to detect subtle signals missed by humans. Machine-learning algorithms can analyse vast amounts of past earthquake data to identify patterns that might predict future events.
How China is aiming to break the seismic prediction barrier?
The National Supercomputing Centre announced that this marks the first time China has compiled seismic data on such a large scale for AI training, surpassing the capabilities of traditional models. The DiTing model currently evaluates 100 million parameters and is expected to reach a billion by August, SCMP said.
A model with more parameters can capture greater detail and improve overall performance. DiTing has been trained on seven years of seismic reports from the China Earthquake Networks Centre, as detailed in a paper published in Earthquake Science, the Seismological Society of China’s flagship journal.
The research team believes this information will serve as a foundation for various earthquake-related studies, setting a benchmark for machine learning model development and data-driven seismological research, the report added. Potential applications include earthquake detection, magnitude prediction, and ground-motion forecasting.
Beyond earthquake prediction, the model has potential uses in mine seismic monitoring, shale gas exploitation, urban underground structure detection, and underwater earthquake monitoring. The deputy director of China’s supercomputing centre said that DiTing can learn the waveform characteristics of oil storage areas to determine if the underground contains oil and gas.