An Indian research team was able to develop a solid scientific framework using artificial intelligence, to overcome previous problems that prevented the development of early warning systems for earthquakes, based on the occurrence of a tangible change in the behavior of some animals about 10 to 20 minutes before the devastating effects of earthquakes occur.
Not long ago, there were reports based on popular observations that animals feel earthquakes before they occur, a concept that has not found scientific acceptance by international scientific bodies, because it presents animals as if they possess a sixth sense that reads the unseen.
Sensing, not predicting
Later, scientific studies attempted to develop a scientific framework for these observations, pointing out that what people notice of changes in animal behavior between 10 and 20 minutes before earthquakes is not prediction, but rather sensing, as animals have senses that far exceed the capabilities of humans, making them capable of sensing the initial waves of earthquakes.
These studies assume that when a rock in the ground is destroyed and an earthquake begins, two main types of waves are emitted. The first is the primary waves, which are very fast and pressure waves, and are the first to reach the surface. They are too weak for humans to feel them, but animals sense them and begin to scream or run away. Secondary waves are the slowest and come later, and are responsible for the violent shaking and destruction that humans feel.
Although these studies were published scientifically, they were not sufficient to build a “general warning system” for earthquakes based on animals, for two reasons, the first of which is “inconsistency”, which means that the behavior is not guaranteed. In some historical earthquakes (such as the Haicheng earthquake in China in 1975) animals behaved strangely and contributed to saving the city, but in many other devastating earthquakes, the animals did not show any strange behavior and no one noticed any change.
The second reason is false alarms. Animals may panic or change the pattern of their sounds for various reasons, such as weather changes, the presence of a predator, mating seasons, or even hunger. Governments cannot evacuate cities and disrupt economies based on an indicator that has a very high error rate.
Artificial intelligence filter
To solve the problem of distinguishing “the precise and very specific vocal tone of fear of ground waves prior to an earthquake,” and separating it from normal panic sounds resulting from other causes, which prevented large geological institutions from trusting the behavior of animals as an early warning of earthquakes, researchers from several Indian institutions used artificial intelligence to act as this “strict filter,” and announced the results of their work in the journal Scientific Reports.
Dr. Surya Pavan Kumar Godla from the Aditya Institute of Technology and Management in India, and the study’s co-researcher, told Al Jazeera Net: “We collected and analyzed real and field audio recordings from major seismic disasters, such as the devastating Turkish earthquake, extracting the sounds of dogs, cats and birds in the critical moments that preceded the violent earthquakes, to subject them to intensive training via machine and deep learning algorithms.”
To ensure the purity of vital indicators and isolate them from disaster noise, Dr. explains. Syria stated that they used precise signal processing techniques to extract the spectral and temporal characteristics of audio tones, such as frequencies, energy intensity, and zero-crossing rate. They also introduced noise and artificial modifications to the recordings (such as changing intensity and time dilation) to train the system to accurately detect real “panic cries,” even under the worst surrounding environmental conditions, ensuring the efficiency of the algorithms in real-life environments full of human and climatic noise.
Dogs take center stage
The study produced promising results that changed the balance of early warning systems, as the artificial intelligence model based on “Bi-directional Long Short-Term Memory” (Bi-LSTM) achieved a record test accuracy of 98.87% in distinguishing between natural sounds and those warning of danger.
He says, “Several popular algorithms such as Random Forest, Extreme Gradient Boosting (XGBoost), and Bidirectional Long Short-Term Memory (Bi-LSTM) were tested, and the latter proved its superiority.”
While other algorithms read the audio moment in the abstract, the “two-way long-short-term memory” algorithm is distinguished by its unique ability to process the audio signal in two simultaneous directions together (from the past to the future, and from the future to the past). This “dual perspective” gave artificial intelligence the ability to understand the full context and gradual development of the animal’s tone, which enabled it to capture the most minute changes and hidden spectral patterns that preceded earthquakes, and which other systems were unable to monitor.
Soria says, “Audio analyzes based on the ‘two-way long-short-term memory’ model showed that the barking and howling of dogs constituted the strongest and most clear biological indicator due to the sudden changes it carries in tone, frequencies, and sound intensity, which made it an ideal sound fingerprint that the algorithms captured with an efficiency that surpassed the rest of the animals observed in the study.”
Low cost life jacket
While Surya and his colleagues do not claim that animal sounds can replace traditional geological monitoring devices, they stress that this system represents a “supplemental layer of protection” capable of giving communities a critical window of time of between 10 and 20 minutes before disaster strikes.
He says: “This technology is gaining exceptional importance as a low-cost first line of defense in developing regions deprived of complex seismic infrastructure, in preparation for its future integration with Internet of Things (IoT) technologies to compute data and send immediate warnings that will save thousands of lives.”
He adds, “Our next step will be to develop highly accurate smart audio sensors and build the largest indexed vital audio database for long-term monitoring, to serve as the cornerstone of a hybrid system that combines nature’s instinct and machine intelligence to protect human lives.”
