Extracting building typology parameters from 3D city models for earthquake risk assessment
S.H. Brakelé (TU Delft - Architecture and the Built Environment)
J.E. Stoter – Mentor (TU Delft - Architecture and the Built Environment)
G. Giardina – Mentor (TU Delft - Civil Engineering & Geosciences)
Ihsan E. Bal – Mentor (Hanze Hogeschool Groningen)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Earthquakes pose a significant risk to communities worldwide, damaging buildings and disrupting society. This makes it essential to understand how buildings respond to seismic events. Earthquake risk assessments rely on structural building information, typically based on clusters of buildings with similar structural characteristics. Traditionally, this data is collected through time-consuming and expensive field surveys, because many relevant structural attributes are not openly available on a large scale. Meanwhile, developments in 3D city models offer accurate geometric data that could support automated building analysis. However, many models currently lack information on structural characteristics. Therefore, this research aims to explore how 3D city models can be used to automatically extract and predict building typology parameters at the individual building level to improve earthquake risk assessment.
This approach focuses on extracting and predicting Global Earthquake Model (GEM) taxonomy parameters from 3D city building models. First, the geometric and descriptive attributes are extracted using a computational process. Then, correlation analysis is performed to select a unique set of predictive features. Machine learning techniques are then used to predict non-visual structural characteristics, such as the type of walls, foundations, and floors. This study applies the methodology to the area around the Groningen gas field in the Netherlands as a case study. It uses 3DBAG, an automatically generated 3D model of buildings in the Netherlands, alongside a small dataset of structural characteristics of buildings in Groningen obtained through field surveys.
The results demonstrate that geometry-based attributes of the GEM, such as building height, inter-storey height, structural irregularity, roof shape, and floor plan shape, can be reliably derived from 3D building data. Predictions of non-visual structural attributes (i.e., wall, foundation, and floor types, and Lateral Load-Resisting System (LLRS)) reach an accuracy of around 90%, while full vulnerability class predictions achieve approximately 75% accuracy. However, these results should be interpreted with caution due to the small validation dataset and the probabilistic nature of the models. Nevertheless, these findings demonstrate the potential value of 3D city models as data sources for the automated, large-scale retrieval of individual building typology parameters.