AI-based concepts for Crisis Propagation Forecasting and Early Warning in Urban Areas

Abstract

By 2050, over 68% of the global population is expected to live in cities, increasing demands on energy grids, transportation networks, and public services. This urban growth will intensify the complexity of infrastructure systems and heighten vulnerability to natural disasters, cyber-physical failures, and crises. In urban emergency management, unexpected events frequently occur that traditional radar and weather systems fail to detect. Additionally, existing warning mechanisms are often disconnected from predictive models, causing delays in issuing context-aware alerts. Current systems also struggle to deliver tailored warnings that account for the dynamic circumstances of individual citizens. In this large panorama, AI-powered technologies, particularly those using machine learning and natural language processing, offer promising solutions to overcome these limitations and improve urban crisis management. Based on that, this paper proposes an integrated crisis warning system to address these challenges. It presents an AI-centered pipeline that connects an imputation model, a forecasting model, and a crisis advisor, creating a robust and adaptive warning system. The proposal is evaluated through a scenario-based case study conducted in Darmstadt (Germany). Storm scenarios are used to test its forecasting and alert-generation capabilities, highlighting its potential to enhance urban resilience and citizen safety.