Estimating Road Disruptions in Urban Contexts Due to Earthquakes Using Machine Learning Surrogates

Abstract

The estimation of road disruptions due to building debris in urban contexts requires the availability of exposure data at the building level, which is often not available. In this study, we explore how open global datasets at different scales can be integrated with machine learning algorithms to estimate road disruptions following seismic events, overcoming the need for detailed datasets. Using simulated impact data for the municipality of Lisbon, we train a Random Forest model to predict road disruptions due to building collapses. Then, we apply this model to another urban environment (the municipality of Amadora) to evaluate the performance of the model using input data unseen during the training process. Finally, we employ the surrogate model using information extracted from globally available datasets characterizing the built environment and the road network. The proposed approach allows identifying areas within urban centers where road disruptions are likely to occur, and where risk reduction measures should be prioritized to minimize the impact of destructive earthquakes.