Impacts of Urban Three-Dimensional Morphology on Seismic Vulnerability and its Driving Mechanisms: A XGBoost-SHAP Explainable Analysis in Jiangyou City, Southwestern China

Cities are not just flat maps of streets and blocks; they are three-dimensional landscapes of rising walls and crowded spaces. When the ground shakes, the way a city is built in height and density matters just as much as the quality of the bricks and mortar. For decades, scientists have tried to predict which buildings are most likely to fail during an earthquake, often focusing on individual structures or simple factors like how old a building is. However, these traditional methods often miss the complex, hidden relationships between the overall shape of a neighborhood and how it survives a tremor. They struggle to capture the non-linear ways that a cluster of tall buildings might interact with a cluster of short ones, or how the sheer crowding of a district changes the risk for everyone inside it. Understanding these spatial patterns is crucial for urban planners who want to design cities that can withstand disaster, yet the tools to see these patterns clearly have been limited.

In a recent study focused on Jiangyou City in southwestern China, a team of researchers tackled this challenge by combining advanced computer learning with a deep dive into the city's physical form. Jiangyou is a place where the threat of earthquakes is real and present; it sits along a major fault line and suffered devastating losses during a massive earthquake in 2008. The city is characterized mostly by low and mid-rise masonry buildings, offering a distinct contrast to the towering skyscrapers of megacities. The researchers wanted to understand how the specific three-dimensional arrangement of these buildings—their density, their height variations, and how tightly they are packed together—influences their vulnerability to seismic shaking. To do this, they did not rely on simple guesswork or linear equations that assume a straight-line relationship between cause and effect. Instead, they built a sophisticated digital model that could learn from the data, uncovering the subtle, complex rules that govern how urban shapes respond to stress.

The team started by gathering detailed information on nearly 3,000 buildings in the city, recording their height, floor area, structural type, and age. Because the number of actual buildings was too small to train a powerful computer model on its own, they used a technique to generate thousands of realistic, synthetic examples that mirrored the real city's characteristics. This allowed them to feed a massive amount of data into a machine learning algorithm known as XGBoost, which is designed to find patterns in complex, high-dimensional data. Once the model learned to predict seismic vulnerability with high accuracy, the researchers used a special tool called SHAP to open up the "black box" of the algorithm. This tool acted like a spotlight, revealing exactly which features of the city's shape were driving the predictions and how much each one mattered. They also mapped the results to see if vulnerable areas were clustered together or scattered randomly across the landscape.

The results painted a clear and surprising picture of how city form dictates safety. The most powerful factor influencing a building's vulnerability was not its height or its age, but something the researchers called the "spatial crowding degree." This metric measures how tightly buildings are packed into the three-dimensional space around them. The study found that higher crowding generally reduced vulnerability, suggesting that dense, compact neighborhoods provide a kind of mutual structural support that helps buildings withstand shaking. In contrast, features like the variation in building heights and the ratio of floor area to land area showed complex, non-linear effects. For instance, while having a mix of building heights can sometimes help disperse seismic energy, extreme variations or specific combinations of tall and short buildings could actually increase the risk. The researchers discovered that the relationship between a city's shape and its safety is not a simple rule where "more is better" or "less is better." Instead, it depends on finding a balance; extreme configurations, such as very high density combined with very low volume variation, tended to amplify risk, while more coordinated and balanced layouts tended to lower it.

When the researchers looked at the map of Jiangyou, they saw that these risks were not spread evenly. They identified distinct zones where high-risk areas were surrounded by low-risk ones, and vice versa, revealing a pattern of spatial spillover. Some areas showed a "high-low" pattern, where a dense, intense pocket of development sat next to sparse, low-density neighborhoods, creating a sharp transition that could be dangerous. Other areas were "low-high," where small, simple buildings were surrounded by larger, more complex structures, creating a gradient of risk. The study confirmed that the most vulnerable spots were often those with unbalanced or extreme morphological features, such as areas with very high building density but low structural variety, or regions where the building heights were wildly inconsistent. The computer model they built proved to be far more accurate than previous methods, correctly predicting vulnerability levels with a high degree of precision, which suggests that these complex spatial relationships are indeed the key to understanding earthquake risk.

Ultimately, this research offers a new way of looking at urban safety. It moves beyond the idea that we can only protect individual buildings and suggests that the collective shape of a neighborhood is a critical factor in disaster resilience. The study indicates that by carefully managing how buildings are arranged in three dimensions—ensuring a balanced mix of density, height, and volume—cities can naturally reduce their exposure to earthquake damage. The findings provide a scientific basis for urban planners to design neighborhoods that are not just functional, but inherently safer. By understanding the specific ways that spatial crowding and height variations interact, cities can avoid the extreme configurations that lead to disaster and instead foster layouts that absorb and distribute the energy of the earth's tremors, protecting the people who live within them.