AI Helps Keep Flood-Hit Communities Connected

In Bangladesh, annual floods often cut off millions from communication networks, isolating communities and hindering emergency responses. This disruption leaves people without access to critical alerts, health data, and coordination efforts, exacerbating the impact of natural disasters. The need for resilient communication systems that don't rely on fixed infrastructure is urgent, as traditional networks like cellular and satellite frequently fail under such conditions. A new protocol called AZIZA addresses this by enabling data delivery through mobile devices, drones, and vehicles, ensuring that essential services remain operational even in the most challenging environments.

Researchers developed AZIZA to achieve high delivery rates for messages in flood-affected areas, where connectivity is intermittent and nodes are mobile. The protocol successfully delivered over 92% of messages in simulations, outperforming established s like Epidemic Routing and MaxProp, which achieved 88% and 86% delivery rates, respectively. This improvement means more emergency alerts and sensor readings reach their destinations reliably, reducing the risk of miscommunication during crises. By focusing on real-world scenarios from Bangladesh, the system demonstrates its ability to maintain communication when it's needed most, without depending on fragile infrastructure.

Ology behind AZIZA involves dividing disaster regions into geographic zones and using a combination of AI-driven decision-making and trust-aware routing to forward messages efficiently. Nodes, including smartphones, UAVs, and vehicles, estimate delivery probabilities to different zones and use a decision tree classifier to decide whether to forward, hold, or drop messages based on factors like trust scores, residual energy, and message urgency. For instance, the classifier analyzes a feature vector that includes delivery probabilities, trust levels, and energy states, achieving 92.8% accuracy in simulations. This approach ensures that data moves through the network only when conditions are favorable, minimizing wasted resources and maximizing reliability in fragmented environments.

Simulation using the Opportunistic Network Environment (ONE) simulator show that AZIZA not only excels in delivery ratio but also reduces average delivery delay to 120 minutes, compared to 160 minutes for Spray-and-Wait. Additionally, it maintains an overhead ratio of 6.9, significantly lower than Epidemic's 15.2, meaning fewer redundant messages are sent. Energy efficiency is also enhanced, with AZIZA delivering more messages per joule of energy consumed, and it retains 85% delivery performance even when 10% of nodes are malicious, thanks to its trust-scoring system that blacklists unreliable actors. These metrics, detailed in figures like Figure 6 for delivery ratio and Figure 9 for security resilience, highlight the protocol's robustness and efficiency in disaster scenarios.

Of this research are significant for disaster management, as AZIZA can be deployed using low-cost hardware like Raspberry Pi devices and drones, making it accessible for regions prone to floods and other crises. By enabling continuous communication for emergency alerts and health reports, it helps coordinate relief efforts and keep affected populations informed, potentially saving lives and reducing isolation. The protocol's design allows integration with existing systems used by governments and NGOs, such as disaster dashboards in Bangladesh, ensuring it can be adopted widely without major overhauls. This advancement not only addresses immediate needs in flood zones but also sets a precedent for resilient networking in other disaster-prone areas worldwide.

Despite its strengths, AZIZA has limitations, including the need for predefined zone maps and initial trust bootstrapping, which may not always be feasible in rapid-deployment scenarios. The AI decision logic relies on pre-trained models that aren't updated in real-time, potentially limiting adaptability to unexpected events. Additionally, UAV operations depend on weather conditions, requiring backup plans like vehicle relays to maintain functionality. These constraints highlight areas for future improvement, such as developing more dynamic mapping and trust initialization s to enhance the protocol's flexibility and reliability in diverse emergency situations.