INTRODUCTION
Disaster response coordination is critical for minimizing the impacts of natural and man-made
hazards. Timely and effective response relies on rapid collection, validation, and processing of
multi-source information describing disaster occurrences and resource availability.
Traditionally, disaster management systems involve centralized coordination centres which
manually aggregate and verify needs, offers of help, and damage assessments supplied by
affected populations, volunteers, and organizations. However, this centralization often
introduces latency and single points of failure, limiting the scalability and responsiveness of
relief efforts in large-scale or rapidly evolving crises [3][12].
The proliferation of IoT devices, crowdsourced social media data, and satellite remote sensing
offers unprecedented opportunities for real-time disaster situational awareness at scale [11][6].
Social media platforms generate vast volumes of unstructured data containing urgent requests
and offers of assistance, while satellite imagery provides objective damage and flood detection
information [18][25]. Yet, these heterogeneous data streams present challenges of credibility,
trustworthiness, and noise, which impair automated decision making and resource allocation
[7][19]. Furthermore, the volume and velocity of information require AI-driven tools to extract
actionable insights and prioritize resources efficiently [9].
Recent efforts have demonstrated the potential of artificial intelligence and machine learning
to address aspects of automated disaster monitoring, social media mining, and resource
matching [7]. However, existing solutions are mainly siloed, focusing on either image-based
hazard detection or social media text classification, lacking integration into cohesive decision
support systems that quantify report credibility, severity, and urgency [19]. Incorporating
credibility scoring is vital to filter unreliable reports and ensure scarce resources are deployed
to the most critical needs [9]. Likewise, scalable coordination platforms must allow for flexible
interaction across multiple stakeholders including victims, volunteers, NGOs, and
administrators [13].
To address these gaps, we propose SankatSuraksha, an AI-powered multi-hazard disaster
monitoring module that leverages deep learning on satellite image data for accurate flood
detection [25]. Integrated within a scalable Disaster Response Coordination and Monitoring
System (DRCMS), it provides dedicated user and admin portals facilitating disaster reporting,
volunteer and NGO registration, and comprehensive dashboards. Central to this system is a
novel credibility scoring framework quantifying trust, severity, and urgency of all incoming
reports and actors. These scores enhance the resource allocation mechanisms by dynamically
prioritizing response actions based on validated, severity-weighted needs. The modular design
allows integration of additional hazard types and heterogeneous data sources with scalability
to national coverage.
The contributions of this paper are as follows:
We develop a comprehensive AI-driven disaster coordination system that integrates
satellite-based hazard detection with crowdsourced data processing, enabling rapid
needs identification and response prioritization.
We design and implement a credibility scoring mechanism that automatically assigns
trust, severity, and urgency scores to disaster reports, volunteers, and NGOs, facilitating
intelligent resource allocation based on validated priorities.
We realize a prototype with user and admin portals, interactive disaster dashboards, and
geospatial visualization tools, demonstrating state-of-the-art flood detection accuracy
and enhanced response coordination compared to baseline manual systems.
The remainder of the paper is organized as follows. Section II reviews the related works in
disaster management systems and AI applications therein. Section III details the overall system
architecture and methodology. Section IV presents implementation specifics and evaluation
results. Section V discusses limitations and future research directions, and Section VI
concludes the paper.