A CNN-based statistical method for land cover classification to assess urban vulnerability to explosions: Case study of Paris, France

Abstract

Explosions in urban areas pose serious risks to human life and infrastructure, emphasizing the need for accurate vulnerability mapping. While Computational Fluid Dynamics (CFD) is commonly used to model explosions, its complexity and high computational cost limit its use to small areas. This study proposes a novel method combining statistical modeling with land cover data processed by Convolutional Neural Networks (CNNs) to estimate blast and thermal radiation effects. Satellite imagery of Paris is used to classify Buildings, Roads, and Trees, each with a dedicated CNN model achieving up to 95% accuracy. Explosion effects under varying TNT weights are simulated to estimate casualties, structural damage, and costs. The method enables large-scale scenario analysis with minimal computational demand. Applied to Paris, the results demonstrate the model’s value for emergency planning, providing confidence intervals that account for uncertainty. This approach offers a scalable, data-efficient tool to support disaster preparedness and public safety decision-making.