With the increasing frequency and potential for natural disasters, particularly earthquakes, the question of limiting the human losses and material damages associated with earthquakes is among the biggest challenges for countries. The 2022 International Day for Disaster Risk Reduction, held on October 13th, focused on early warning mechanisms and the need to increase awareness of their importance, given the current lack of such mechanisms. The United Nations Office for Disaster Risk Reduction (UNDRR) has indicated that, in general, early warning systems cover no more than half the world’s countries.
The World Health Organization estimates that earthquakes claimed approximately 750,000 lives worldwide between 1998 and 2017, which accounts for more than half of all deaths caused by natural disasters. There is international resolve to invest in early warning systems to help save lives and reduce the resulting damage. The earthquake that struck Turkey and Syria this past February, which killed more than 52,000 people, has revived the conversation about the importance of earthquake early warning systems for minimizing the number of victims.
Prediction Difficulty
Although scientists have identified certain patterns and behaviors that may indicate an impending earthquake, such patterns and factors do not confirm the occurrence of an earthquake and do not lead to a clear and reliable prediction. Some scientists describe earthquakes as largely chaotic and unpredictable, thus making it difficult to develop a reliable early warning system for them. Predicting earthquakes requires identifying their location, time, and magnitude, and geologists grapple with the difficulty of doing so reliably and with certainty. Although some scientists expect earthquakes to occur in specific geographic locations, it is difficult to predict when they will occur. Moreover, the models developed by scientists to understand the behavior of earthquakes are limited by the quality and quantity of the available data, as well as the complexity of Earth’s systems.
Thus, geologists resort to what are known as "earthquake hazard maps," by calculating the probability an earthquake will occur in a specific area during a multi-year timeframe. These maps help in the planning process, though not in prediction. Human and field noise also disperses and distorts seismic noise, with human traffic, construction, and other factors often making it difficult to pick up on clear signals of ground movement and the beginnings of earthquakes. Chris Marone, a professor of earth sciences at Sapienza University of Rome, notes the gap between laboratory and real-world capabilities. While some fissures and indicators can be observed during earthquake simulations in laboratories, these indicators may not be observed in the real world most of the time.
The United States Geological Survey (USGS) stresses that accurate predictions of earthquakes are impossible now and in the foreseeable future, and, at most, scientists can predict that a major earthquake will hit a region within a certain period of years. This prompts the emphasis on minimizing earthquake damage and detecting the initial signs and indicators of earthquakes in order to achieve rapid responses.
AI Capabilities
Artificial intelligence (AI) technologies provide many capabilities and advantages that support earthquake early warning systems and may help in early detection, including the following:
1. Providing more time to reduce earthquake losses: The use of AI technologies contributes to the development of earthquake early warning systems during pre- and post-quake efforts. AI technologies also give people and communities time to evacuate and prepare, help reduce the number of victims, and may help minimize damage to property and infrastructure. The precious seconds that early warning systems give to residents may help reduce the number of casualties by up to 30%.
Given its great speed, AI can provide up to a minute of additional time. For example, some estimates suggest that an early warning system in China before the Wenchuan earthquake—which killed 70,000 people and wounded another 370,000 when it hit Sichuan province in 2008—could have given the population 31 extra seconds and reduced the total number of deaths by an estimated 20,000 to 30,000. Some estimates also indicate that, in cases of poorly constructed and low-rise buildings, the few seconds provided by early warning AI could save more lives.
2. Providing better capabilities for picking up accurate signals: For several reasons, it is difficult to detect tremors, earthquake indicators, and ground noise in the field; however, the capabilities of AI and advanced sensor technologies help detect subtle signals not observed by humans. Among the advantages of AI is the ability to quickly extract signals that are drowned out by other noise.
A research paper published in the journal, Science Advances, in February 2018, concluded that AI can be used to better identify earthquakes and distinguish between earthquake signals and normal geological noise, much as intelligent personal assistant systems are able to isolate human voices from background household noise. Testing of an AI system and network called ConvNetQuake, in the US state of Oklahoma, showed it to be capable of detecting earthquakes 17 times better than traditional methods. While it was hard to detect very small earthquakes due to ground noise, the new network can differentiate between noise and actual earthquakes.
3. Using machine learning to detect earthquake patterns: Machine learning algorithms can analyze huge amounts of data from prior earthquakes in order to identify seismic patterns and models that help predict future quakes. The analysis is based on data collected by remote sensor systems, satellites, and weather stations, as well as on the use of historical data for earthquakes in a given geographical region.
4. Estimating the magnitude of major earthquakes based on gravitational signals: New studies have shown that gravitational signals can be used to instantly estimate the magnitude of large earthquakes. Such signals, which are very weak and spread at the speed of light, result from the disturbance of the gravitational field caused by an earthquake. However, they are approximately six times smaller than seismic waves, which limits their detection using standard techniques.
Rather than relying on seismic waves, AI algorithms can be used to distinguish gravitational waves emanating from a fault, which helps speed up the warning and accurately estimate and track the magnitude of large earthquakes within seconds of their onset. These mechanisms help better estimate the magnitude of the earthquake, which helps estimate the extent of the expected losses and identify appropriate responses. However, the use of this model is limited to earthquakes that are generated from specific faults that are likely to cause large earthquakes, Under the current technology, gravitational signals cannot detect earthquakes that measure less than 8.3.
5. Transmitting and processing satellite navigation data: Some rely on AI to process GNSS data in countries that prohibit the export of data in real time and use AI to instantly transmit that data—and the data derived from it—even in countries with limited communications. The satellite navigation system is used to study earthquakes, monitor ground movements and disturbances in the total electron content in the ionosphere (TEC) that commonly accompany earthquakes, and study changes in the Earth’s ionosphere.
6. Protecting rescuers and survivors from aftershocks: Machine learning technologies and algorithms can help better predict earthquake aftershocks, thus helping keep rescue workers and survivors safe. Harvard University researchers have used deep learning to study aftershock patterns with the goal of predicting them, which will help minimize losses due to the aftershocks of major earthquakes.
7. Using AI to predict impacts: AI may help go beyond the earthquake prediction and warning process itself, to predicting the impact of earthquakes and their potential consequences on the surrounding environment and people, warning the competent authorities and people of those consequences, and taking the appropriate steps, as well as making an early determination of the magnitude of the repercussions. Naturally, this will help governments formulate recovery policies and prepare to deal with future disasters.
International Experiments
Numerous international experiments have incorporated AI technologies into earthquake detection and early warning systems and the sending of the necessary alerts, including:
1. Development of an extensive Chinese early warning system: After the 2008 Wenchuan earthquake, China focused on developing early warning technologies, with Chinese scientists highlighting what they call the "biggest, fastest, and most accurate earthquake early warning system in the world." This system covers 90% of the population living in earthquake-prone areas, and it can pick up seismic signals as quickly as possible and provide early warnings via mobile phones, television, radio, and new media. Reports indicate that the earthquake early warning function includes 800 million television sets and mobile phones.
In a related context, in 2020, China announced the development of a system to monitor earthquakes using AI, which has entered the experimental operational phase in the southwestern provinces of Yunnan and Sichuan. Automated algorithms extract earthquake signals—including the epicenter, magnitude, time, and depth of the earthquake—using seismic wave signals in several countries. In addition to being almost as accurate as manual calculation, the system is also able to detect indicators of the source of an earthquake within one to two seconds and to process vast amounts of data. The Chinese system has already succeeded in identifying earthquake data in several countries, which has prompted Beijing to promote its new system and its use in other countries.
With Chinese scientists stressing the importance of studying atmospheric electron disturbances for the early detection of earthquakes, China has taken steps in this regard. After Beijing launched the Chinese Seismo-Electromagnetic satellite (CSES) to monitor disturbances in the Earth’s ionosphere, Chinese reports indicated last year that the Chinese Earthquake Networks Center observed a reduction in the density of electrons in the ionosphere up to 15 days before the earthquakes that hit China in May of 2021 and January of 2022.
Nevertheless, the ability of this system to predict imminent earthquakes remains small, especially the inability to determine the exact location an earthquake will occur given that large earthquakes are able to create changes in the ionosphere far from the earthquake’s epicenter, which makes it difficult to confirm its location.
2. Tokyo’s seismic movement monitoring network to predict earthquakes: Japan uses an AI-based system to predict earthquakes using satellite images. Those systems are used to detect signs or landslides, monitor aging infrastructure, and detect and repair weak points before natural disasters occur. Tokyo is developing machine learning systems that monitor ground movement to predict earthquakes.
3. Implementation of the ShakeAlert system in the US: The USGS has introduced an earthquake early detection system for California, Oregon, and Washington, known as ShakeAlert. This system employs a network of seismic sensors to detect and assess initial earthquake waves and transmits them to the data center. The system’s algorithms hypothesize the occurrence of an actual earthquake—as opposed to incidental sensor events—if four separate sensors record the occurrence of initial vibrations. The algorithms then estimate the extent, location, and severity of the earthquake, and, via a group of partners, warning messages and notifications are sent to residents, stakeholders, and operators of vital infrastructure.
Those notifications can be integrated with automation or automated responses to maintain public safety. For example, the transit system in San Francisco automatically delays the trains. Automated systems can also send financial assistance to regions that the algorithms calculate are at a high potential for risk.
4. Testing of a German system to analyze earthquake data: A research team from the German Research Center for Geosciences and Humboldt University is using AI technologies and machine learning to analyze seismic data to achieve faster and more accurate prediction of expected tremors in the vicinity of an earthquake. The system has been tested through datasets from earthquake-prone countries that have a network of earthquake stations, such as Italy and Japan, and it has been tested on a set of thousands of recorded earthquakes.
5. European project for earthquake early warning via AI: The EU gives importance to enhancing earthquake early warning systems. In this framework, the EU-funded EARLI project will use AI to identify weak and early seismic signals in order to accelerate early warning and explore the potential to predict earthquakes. The early warning system will rely on signals emanating from gravitational field disturbances caused by earthquakes, and the developed AI algorithm will be adapted to search for signals that precede major earthquakes.
6. Google and Harvard collaboration to predict aftershocks: Google and Harvard have developed an AI system that can predict aftershocks, by studying more than 131,000 earthquakes and aftershocks. Reports indicated this system was tested on 30,000 earthquakes, and it succeeded in predicting the location of aftershocks with greater accuracy than traditional methods.
7. Use of mobile phones in early detection of earthquakes: Mobile phones can be used for the early detection of natural disasters. Some earthquake detection systems that rely on mobile phone data have emerged, such as Zizmos, launched in 2015, which is also known for being relatively cheap compared to traditional detection and warning systems. Zizmos can work on a wider network of sensors, via the creation of a network of millions of low-cost sensors. The user downloads an application that monitors several indicators by phone and collects and sends data when tremors occur. The system can monitor the movement of earthquakes and alert users. Calls have been made to maximize the use of such applications by incorporating them into phones when they are manufactured, rather than relying on users to install them.
8. Some phones companies warn users via instant messages: Some companies have introduced AI technologies to help warn of earthquakes, such as China’s Xiaomi, whose phones have an operating system that sends earthquake warnings to alert its users within seconds and before the first tremors are felt. Some systems with which these companies participate identify lists of people vulnerable to earthquake risks, who need that alert to send notifications and messages, given that sending those notifications takes more seconds and slows the notification process.
Data Challenges
A major aspect of the use of AI technologies for the early detection of earthquakes is associated with data, including the collection of accurate, objective, and sufficiently representative data, because the collection of bad or flawed data causes inaccurate processing. This becomes more difficult when there is a need to collect accurate, representative data for very rare and infrequent events like earthquakes. Networks of sensors indeed help in the rapid collection of high-accuracy data, even from areas with difficult and complex topography and terrain where it is difficult to obtain data by other traditional means. Thus, it is important to increase investments in and development of sensor technologies and to increase earthquake monitoring stations. In general, communications infrastructure is important for collecting data on the one hand, and for sending notifications and warnings on the other hand. Given the large volume of data collected, there are challenges associated with processing vast data and calculating complex machine learning algorithms.
Another challenge is that, based on the data from previous records of earthquakes in the past, the potential for changing trends in the phenomenon as a result of natural factors and changes has not been taken into account. By contrast, while the entire earthquake cycle takes thousands of years, the data collected only cover recent periods that may only go back hundreds of years. This limits the efficiency of the designed model and its historical dimension, causing some to charge that the data formulated by machine learning are inadequate or lacking. Some systems also face difficulty in early detection of major earthquakes, given their infrequency and the lack of historical data on earthquakes.
Several reports consider open access to important data and international cooperation in this regard very important. They call for building reliable local, regional, and international datasets. Early warning applications for natural disasters and earthquakes require international data-sharing, especially since earthquakes may not strike just one country, and the indicators provided by smart systems in each country may not be sufficient to launch early warnings in a timely manner, which is consistent with international calls for a global early warning system.
On the other hand, the issue of data collection and sharing raises privacy and security concerns, whether at the level of personal or state data. In addition, it is important to accelerate data transmission and warnings, as reports warn that those very close to the epicenter may feel vibrations before receiving a warning. This calls for reducing the time needed for the sensors to work, detect the signals, analyze the data via AI systems and algorithms, and send alerts.
Finally, some believe that AI technologies can compete with humans in scale and speed of operations, although prediction quality is linked to other determinants that may not make it a certainty. While many challenges and difficulties stand in the way of the development of an earthquake prediction system, there is current international interest in using AI technologies for detecting earthquakes early, quickly determining their location and strength, and sending the necessary alerts in an early and timely manner, in order to mitigate the potential losses and effects of earthquakes, especially with regard to human losses and the number of victims.