Advancing Urban Resilience: Machine Learning Models on Liquefaction Prediction in Earthquake-Prone Regions

Researchers at Shibaura Institute of Technology have developed machine learning models to predict soil behavior during earthquakes, specifically targeting liquefaction risk. Their study, published in Smart Cities, utilizes artificial neural networks to create detailed soil layer maps, significantly improving prediction accuracy and aiding in urban infrastructure planning. This innovative approach aims to enhance resilience in earthquake-prone regions, supporting safer and smarter city development.

As urban populations continue to grow, the risk of natural disasters presents a significant challenge to city planners and disaster response agencies. In seismically active nations such as Japan, an urgent concern is liquefaction. This phenomenon occurs when intense seismic shaking leads loose, saturated soils to lose their strength, transforming them into a liquid-like state. As a result, liquefaction poses threats to infrastructure, including sinking buildings, compromised foundations, and the disruption of critical utilities. Historical events, such as the 2011 Tōhoku earthquake, which caused damage to 1,000 homes due to liquefaction, underscore the urgency of addressing this risk. Other incidents include the 6.2 magnitude earthquake in Christchurch, which devastated 80% of the water and sewage systems due to liquefaction and the recent Noto earthquake in 2024 that impacted 6,700 residences. To enhance urban resilience against liquefaction impacts, Professor Shinya Inazumi and his student Yuxin Cong from the Shibaura Institute of Technology have been innovating machine learning models designed to predict soil behavior during seismic events. These models leverage geological data to create comprehensive three-dimensional soil layer maps, distinguishing stable areas from those more susceptible to liquefaction. This advanced approach surpasses traditional soil testing methods, which are often limited in geographic scope and resolution. Their recent study, published in the journal Smart Cities on October 8, 2024, utilized artificial neural networks (ANNs) in conjunction with ensemble learning methods to accurately estimate the depths of foundational soil layers, a vital metric for assessing soil stability and liquefaction potential. “This study establishes a high-precision prediction method for unknown points and areas, demonstrating the significant potential of machine learning in geotechnical engineering. These improved prediction models facilitate safer and more efficient infrastructure planning, which is critical for earthquake-prone regions, ultimately contributing to the development of safer and smarter cities,” states Professor Inazumi. By effectively predicting zones with deep, stable bearing layers, the researchers can pinpoint areas where soil is likely to provide adequate support for structures during seismic events, including liquefaction occurrences. Through rigorous data gathering at 433 locations in Setagaya-ku, Tokyo, employing standard penetration testing and mini-ram sounding tests, the study documents crucial data such as geographic coordinates and elevations. The ANN was trained to predict bearing layer depths at an additional ten locations. The actual ground measurements were employed to validate prediction accuracy. To enhance this process further, the researchers implemented a bagging technique (bootstrap aggregation) that trained the model repeatedly on various subsets of the data, yielding a 20% boost in prediction accuracy. Using these predicted values, the team created a contour map displaying the depths of bearing layers within a kilometer radius of four selected sites in Setagaya Ward. This map serves as a vital resource for civil engineers in identifying suitable construction sites with favorable soil conditions while also assisting disaster management professionals in evaluating areas more likely to experience liquefaction, enabling improved risk assessments and mitigation strategies. The researchers foresee their modeling approach as a key asset for the advancement of smart cities and stress the need for data-centric strategies to guide urban development and infrastructure planning. “This study provides a foundation for safer, more efficient, and cost-effective urban development. By integrating advanced AI models into geotechnical analysis, smart cities can better mitigate liquefaction risks and strengthen overall urban resilience,” affirmed Professor Inazumi. Future enhancements to their model will incorporate additional environmental conditions and specialized models for varying coastal conditions, considering the impact of groundwater—an essential element influencing liquefaction.

Liquefaction is a critical concern in earthquake-prone regions, notably Japan, as urban expansion increases the vulnerability of infrastructure to natural disasters. The phenomenon occurs when the seismic forces induce saturated soils to behave as fluids, leading to significant structural damage, as evidenced by past earthquakes. This necessitates advanced prediction methods to mitigate such risks effectively, highlighting the importance of innovative approaches like machine learning in geotechnical engineering. The study by Professor Shinya Inazumi and Yuxin Cong introduces a novel technique for enhanced soil behavior prediction, aiding in the development of safer urban environments.

The introduction of machine learning models to predict soil liquefaction enhances the resilience of urban areas in earthquake-prone regions. Professor Inazumi and Yuxin Cong’s work marks a significant leap forward in using data analytics for effective infrastructure planning and disaster preparedness. By improving prediction accuracy and mapping soil stability, this research aims to foster safer, smarter cities resilient to natural disasters. Future enhancements will further refine the model, accommodating a broader range of ground conditions and contributing to safer urban development practices.

Original Source: techxplore.com