Resilient Mobility Through AI: Revolutionizing Emergency Planning with Digital Twins

An article from the Center for Secure Water (C4SW)

Written by Hadi Meidani

Extreme weather events and natural disasters, such as floods and hurricanes, are increasingly disrupting critical infrastructure systems worldwide. Urban transportation networks, which are essential for evacuations, delivering emergency resources, and maintaining connectivity during crises, are particularly vulnerable. Recent hurricanes, which have caused widespread disruptions along the southern states and the U.S. East Coast, have underscored the fragility of these systems.

Effectively planning for resilience and managing disruptions to transportation systems requires the use of computationally intensive transportation network models, which are often too resource-heavy and not scalable for real-time multi-scenario decision-making during disasters or for planning across regional networks. AI-powered digital twins for transportation networks offer a viable and effective tool to model and predict dynamic changes in mobility conditions, and support emergency planning and effective response

---

Addressing complexity in emergency planning

During a disaster, transportation networks experience rapid changes in conditions due to flooded roads, traffic congestion, or damaged infrastructure. Traditional approaches for network modeling tasks such as shortest path estimation and traffic assignment have limitations in modeling dynamic changes and considering multiple scenarios. This is because they are traditionally computationally intensive and rely on static data. The lack of real-time adaptability leads to delays in emergency responses, inefficient evacuation planning, and increased risks to public safety.

Key challenges include:

Computational Inefficiency: Traditional algorithms for routing and traffic assignment struggle to scale efficiently, particularly for large urban networks and multiple disruption scenarios. These algorithms often require significant processing power and time to generate solutions, causing delays in decision-making.

Dynamic Disruptions: Transportation networks are subjected to frequent and unpredictable disruptions during disasters, such as road closures from flooding, accidents, or debris blocking routes. These disruptions cause the network topology to change rapidly, sometimes with uncertainty, rendering static models ineffective.

Limited Integration of Real-Time Data: Existing transportation models often lack the capability to integrate live data from diverse sources, such as traffic sensors, drones, satellite imagery, or other remote sensing technologies. Without this integration, they cannot reflect the current conditions on the ground, such as changes in traffic flow, road closures, or newly emerging hazards.

Leveraging AI-based digital twins to support emergency planning

To address these challenges, we developed an AI-powered digital twin for transportation networks, by adopting Graph Neural Networks (GNNs) to model and predict mobility conditions given dynamic changes in network conditions. The digital twin, a digital replica of the physical transportation system, mirrors its real-world counterpart and uses static and real-time data to simulate, analyze, and predict the performance and behavior of that physical system. The digital twin is designed to support asset managers and emergency planners by enabling improved decision-making, monitoring, and optimization and supporting real-time impact assessment and effective response.

The improved system relies on four core capabilities: 

1. Shortest Path Estimation: Using GNNs, the digital twin predicts shortest paths in dynamically changing networks. Unlike traditional methods, GNNs utilize node and edge embeddings to incorporate features such as road lengths and flood levels, enabling real-time updates.

2. Dynamic Traffic Assignment (DTA): The model dynamically assigns traffic based on changing demand and network conditions, providing insights into congestion patterns and alternative routes.
3. Real-Time Data Integration: By integrating remote sensing data (e.g., flood maps) and sensor inputs, the digital twin updates its models to reflect current conditions, enhancing accuracy and responsiveness.
4. Scenario Simulation: The digital twin supports simulation of various hazard scenarios, such as hurricanes and floods, and quick evaluation network resilience and optimize evacuation plans.

 

Building Resilient Infrastructure for Transportation

The AI-based digital twin offers transformative benefits for transportation resilience and emergency planning.

By rapidly identifying efficient evacuation routes during floods or other disasters, the tool reduces delays in reaching shelters and hospitals, minimizing risks to human life and enhancing public safety. The system also allows for optimization of resource allocation as asset managers and emergency planners can prioritize high-risk areas and allocate resources more effectively based on real-time data and predictive analytics. Moreover, reducing downtime in transportation networks minimizes economic losses during disasters, ensuring faster recovery for affected communities and building economic resilience.

Case studies on flood-prone areas in Manhattan and Hillsborough County have shown the model’s ability to accurately and quickly estimate evacuation delays and identify critical bottlenecks. This helps planners quickly test several evacuation or routing strategies towards an optimal real-time response and evacuation plan and enhanced disaster preparedness.

Looking into the future of AI-based Digital Twins

The AI-based digital twin represents a significant step forward in enhancing transportation resilience and emergency preparedness. By leveraging advanced AI techniques, real-time data, and scenario modeling, this tool empowers planners to make faster, data-driven decisions that save lives and reduce disaster impacts.

Building on current advancements, the next steps in this research include several key areas aimed at enhancing the effectiveness of digital twin-based emergency planning tools. One important development is hybrid modeling, which involves combining Graph Neural Networks (GNNs) with other artificial intelligence techniques, such as reinforcement learning. This approach will improve decision-making capabilities, especially under uncertain conditions, where real-time responses are critical. Another area of focus is interconnected systems, expanding the digital twin framework to model the interdependencies between transportation networks and other critical infrastructure systems, such as power grids and water supply. This will allow for a more holistic understanding of how disruptions in one system can impact others during emergencies. Additionally, there is a push for open-source deployment of the digital twin framework. By releasing it as an open-source tool, the research aims to foster collaboration and encourage adoption by a broader range of stakeholders, including researchers, policymakers, and emergency planners. Finally, the model will be extended for global applications, with a particular focus on addressing resilience challenges in developing countries, where infrastructure vulnerabilities are often more pronounced.

The journey toward fully resilient mobility systems is an ongoing one. Addressing remaining knowledge gaps and integrating digital twins into broader policy frameworks will be key to unlocking their full potential. Through continued innovation and collaboration, we can ensure safer, more efficient, and resilient infrastructure systems for communities worldwide.