In regions at risk for earthquakes, such as Japan, there is an urgent need to improve the prediction of soil stability to reduce the chances of liquefaction. To address this, researchers have implemented machine learning techniques—specifically artificial neural networks and bagging methods—to develop precise 3D visualizations of supporting soil layers, using information gathered from 433 locations in Setagaya, Tokyo. This innovative method can assist in identifying safe construction sites, improving disaster preparedness, and fostering safer urban development, thereby enhancing city resilience against liquefaction threats.
In regions at risk for earthquakes, such as Japan, there is an urgent need to improve the prediction of soil stability to reduce the chances of liquefaction. To address this, researchers have implemented machine learning techniques—specifically artificial neural networks and bagging methods—to develop precise 3D visualizations of supporting soil layers, using information gathered from 433 locations in Setagaya, Tokyo. This innovative method can assist in identifying safe construction sites, improving disaster preparedness, and fostering safer urban development, thereby enhancing city resilience against liquefaction threats.
As cities continue to grow, the risk of natural disasters poses a serious challenge for urban planners and emergency management officials. In earthquake-prone nations like Japan, a significant threat to infrastructure comes from liquefaction. This is a phenomenon where intense shaking makes loose, saturated soils lose their strength and act like a liquid, resulting in buildings sinking, foundation cracks, and disruptions to roads and utilities.
Liquefaction often occurs during major earthquakes, leading to considerable destruction. For example, the 2011 Tōhoku earthquake caused liquefaction that affected 1,000 houses, while Christchurch’s 6.2 magnitude earthquake resulted in liquefaction that devastated 80% of its water and sewage systems. Additionally, the 2024 Noto earthquake led to widespread liquefaction impacting 6,700 homes.
To bolster cities’ resistance to liquefaction effects, Professor Shinya Inazumi and his student Yuxin Cong from Shibaura Institute of Technology have been creating machine learning models to forecast how soil behaves during earthquakes. These models utilize geological information to produce detailed 3D mappings of soil layers, highlighting both stable regions and those at greater risk for liquefaction. This technique surpasses traditional soil testing methods that can’t reach every location, providing a more comprehensive understanding of soil behavior.
In their latest research published in Smart Cities on October 8, 2024, they employed artificial neural networks (ANNs) and ensemble learning strategies to accurately determine the depths of supporting soil layers, which is vital for assessing soil stability and liquefaction likelihood during earthquakes.
“This study sets a new standard for predicting depths in previously unknown locations, highlighting the tremendous capabilities of machine learning in geotechnical engineering. These refined predictive models allow for safer and more effective infrastructure development, crucial for regions susceptible to earthquakes, ultimately aiding in the creation of safer, smarter cities,” states Prof. Inazumi.
Identifying areas with deeper, more stable soil layers helps indicate where ground can provide adequate support for structures, particularly during liquefaction events. The researchers gathered depth data from 433 sites in Setagaya-ku, Tokyo, through standard penetration tests and mini-ram sounding tests. Alongside the bearing layer depth, they recorded essential details such as longitude, latitude, and elevation for each site.
The data were then utilized to train an ANN to forecast the depth at 10 additional locations, comparing the predictions to the actual measurements for accuracy assessment. The researchers enhanced the prediction accuracy by implementing a bagging technique, which involves training the model multiple times using varied subsets of the training data, resulting in a 20% improvement in prediction results.
Using the anticipated values, the researchers created a contour map displaying bearing layer depths within a 1 km radius of four selected sites in Setagaya Ward. This visual representation serves as a crucial tool for civil engineers, facilitating the identification of locations with favorable soil conditions for construction. Furthermore, it aids disaster management specialists in recognizing areas vulnerable to soil liquefaction, enabling improved risk assessment and mitigation techniques.
The researchers view their approach as pivotal for fostering smart city development, underlining how data-based strategies can effectively guide urban planning and infrastructure execution. “This research lays the groundwork for urban development that prioritizes safety, efficiency, and cost-effectiveness. By integrating advanced AI models into geotechnical evaluations, smart cities can more effectively address liquefaction challenges and enhance overall urban resilience,” notes Prof. Inazumi, expressing optimism about the future effects of their study.
Looking ahead, the researchers aim to refine their model’s precision by considering additional ground conditions and devising specialized models tailored to coastal and non-coastal areas, taking into account the impact of groundwater, a crucial factor in liquefaction.
