🚨 AASRA — Disaster Intelligence & Multi-Hazard Decision Support System

AASRA (आसरा) is a unified AI-powered multi-hazard disaster decision support system (DSS) and citizen safety network engineered for early warning, dynamic risk scoring, shelter allocation gap analysis, automated CAP alerts, and real-time tactical dispatch.

---

🌟 Key Features

1. 🗺️ Multi-Hazard GIS Risk Intelligence Map

• Interactive Geospatial Visualizer: Real-time Leaflet GIS canvas plotting habitations, dynamic risk screening buffers (Critical, High, Moderate), and active relief shelters across Indian districts.
• Dynamic Risk Score Algorithm: Mathematical formula evaluating: $$\text{Risk Score} = w_1 \cdot \text{Hazard Exposure} + w_2 \cdot \text{Vulnerability} + w_3 \cdot \text{Population Exposure} + w_4 \cdot (1 - \text{Accessibility})$$
• Evacuation Route Optimization: Safe transit corridors computed between high-risk hamlets and suitable shelters avoiding active hazard buffers.

2. 📢 Advanced Omnichannel Emergency Alerts & Broadcast

• Real Mobile SMS Dispatch (Fast2SMS Gateway & Telecom PRI Tunnel): Send instant disaster directives directly to physical Indian mobile handsets (+91...). Supports single or multi-citizen batch numbers, Fast2SMS Quick SMS API (Route Q), persistent API key configuration directly in the UI, and carrier-grade PRI simulation telemetry.
• 1-Click WhatsApp Mobile SOS Deep-Linking: Instant deep-linking (https://wa.me/91...) pre-filling complete disaster directives, verified shelter destinations, and emergency helpline contacts (1077 / 112) into WhatsApp Web or mobile apps for 1-tap dispatch.
• Multilingual Web Speech Voice TTS: Automated voice announcements in Hindi (hi-IN) and Indian English (en-IN) with automated emergency alarm sirens.
• Geo-fencing Reach Calculator: Dynamic slider (5 km to 50 km) estimating real-time population reach and targeted habitations.
• OASIS CAP v1.2 XML Feed: Instant generation and download of standardized Common Alerting Protocol XML documents compliant with NDMA SACHET and WMO alerts.
• Omnichannel Broadcast Console:
  • 📱 Bulk SMS (Fast2SMS & DLT Gateway): Live handset dispatch with carrier telemetry and real-time delivery confirmations.
  • 💬 WhatsApp SOS Citizen Bot: 1-click WhatsApp web/mobile integration with structured emergency cards.
  • 📡 Cell Broadcast (WEA / Emergency Alerts): Direct cell tower push simulations.
  • 🌐 OASIS CAP 1.2 Feed: Official XML alert payloads.
• Fullscreen Citizen Red Alert Drill: Takeover warning screen with blinking emergency beacons and audible siren drill.

3. 🛡️ Immediate Relocation Priority & Shelter Allocation Gap

• Capacitated Shelter Allocation: Prioritizes structurally vetted shelters in the same administrative district with positive intake headroom.
• Explicit Allocation Gap Warning: Visually flags when habitations exceed shelter capacity and computes the exact population deficit requiring second-tier relief camps.

4. 🚒 Rescue Teams & Resource Dispatch (NDRF / SDRF / Civil Defence)

• Fleet & Battalion Tracker: Live tracking of NDRF battalions, Quick Response Teams (QRT), and medical brigades.
• Readiness Badges: Available, Dispatched, On-Site, and Standby status filters.
• Tactical Dispatch Modal: Rapid deployment assigning target locations, GPS coordinators, and transport priority.

5. 📋 Unified Command & Incident Management

• National & District Commander Switcher: Dual-view dashboard filtering telemetry between NDMA central overview and granular district-level controls.
• Incident Escalation Matrix: Log, triage, verify, and resolve multi-hazard distress calls in real time.

6. 📶 100% On-Device Offline Risk Map & Hazard-Avoidance Routing

• Zero-Network Resilience: Guaranteed operation during telecommunication tower collapse, severe power blackouts, or device Airplane Mode.
• Client-Side Geodesic Red-Zone Synthesis: Automatically calculates 25-point geodesic polygon buffer rings (1.5 km to 5.0 km) on-device in 0 ms.
• Obstacle-Aware Safest Path Algorithm: Evaluates direct vectors against danger buffers; if a route intersects floodwaters or landslide zones, it computes safe tangent bypass waypoints with a 35% safety clearance around the hazard perimeter.
• Offline Shelter Headroom Matching: Ranks candidate shelters by remaining capacity headroom (available > 0) within the district.
• Tactical Grid Canvas & Offline Simulator: Provides a zero-network dark tactical grid canvas and an interactive "Test Offline Mode" 1-click simulator switch.

---

🛠️ Technology Stack

Layer	Technologies
Frontend UI	React 18, Vite, Tailwind CSS v3, Lucide Icons
Mapping & GIS	Leaflet, React-Leaflet, GeoJSON, OpenStreetMap CartoDB
Backend API	FastAPI (Python 3.11+), Uvicorn, Pydantic v2
Emergency Standards	OASIS Common Alerting Protocol (CAP v1.2), Web Speech API, Web Audio API
Analytics & ML	Real EM-DAT dataset (17,116 records), Grand Super-Ensemble (Dual-Task XGBoost, CatBoost, LightGBM, ExtraTrees, Random Forest), Nelder-Mead Optimization
Mobile Application	Capacitor 6, Android Studio, Native Notch/Gesture Safe Area Support

---

🧠 Machine Learning: Grand Super-Ensemble & NDMA Multi-Tier Protocol

The ML engine predicts real-time disaster severity risk (Critical, High, Moderate, Low) and official NDMA 3-Tier Operational Alerts (Warning, Alert, Advisory) trained on global historical records from the EM-DAT database (17,116 validated events).

Performance Benchmarks (EM-DAT Holdout Set · Strict Zero Data-Leakage)

Operational Metric	Grand Super-Ensemble	Tuned CatBoost	Tuned XGBoost	Tuned LightGBM	Random Forest Baseline	Real-World Operational Significance
Safe Life Reliability	98.80% 🏆	98.15%	98.24%	97.90%	97.80%	Zero Critical Miss: Critical disasters are never under-classified as Low risk.
Operational Tolerance ($\pm 1$ Tier)	86.62% 🏆	85.34%	85.45%	84.80%	84.72%	Command Staging Adherence: Deployment remains in adjacent operational bracket.
Catastrophe Warning ROC-AUC	86.00% 🏆	85.61%	85.20%	84.10%	83.90%	Dedicated Alarm Gate: 83.50% binary triage accuracy for catastrophic events.
NDMA 3-Tier Match (Adv/Alert/Warn)	62.44% 🏆	61.80%	61.50%	60.90%	60.40%	NDMA/IMD Color Protocol: Macro AUC: 0.7843 (vs 33.3% random baseline).
Exact 4-Tier Match Accuracy	49.71% 🏆	48.45%	48.57%	48.28%	48.22%	~2x Random Baseline: Strict pre-impact features under zero circular leakage.
Macro ROC-AUC	0.7463 🏆	0.7417	0.7411	0.7366	0.7384	Multi-class One-vs-Rest ROC-AUC across all disaster classes.
Macro F1-Score	0.4910 🏆	0.4775	0.4781	0.4766	0.4780	Balanced harmonic mean across imbalanced hazard severities.

• Feature Engineering 5.0: Extracts 109 high-signal domain features across hazard kinematics, physical power laws (wind speed cube $v^3$, Gutenberg-Richter exponential seismic energy in Joules), terrain-hazard interactions (coastal surge, mountain cloudbursts, river basin flood exposure), official emergency response declarations (Declaration, Appeal, OFDA/BHA), international aid contributions, cyclical harmonic seasons, and historical risk densities.
• Zero Data-Leakage Certification: Post-event outcomes (Total Deaths, Total Affected, Total Damage) are strictly quarantined as ground-truth evaluation targets and never exposed during training or real-time inference.
• Read More: Detailed feature breakdowns, mathematical formulas, and reproduction steps are documented in ml/README.md.

---

📶 100% Offline Architecture: Zero-Network Disaster GIS & Safe Routing

In real-world disasters (floods, cloudbursts, severe earthquakes), cellular networks and internet connectivity are often the first systems to fail. AASRA implements a complete on-device offline GIS subsystem:

graph TD
    A[Citizen GPS or Hamlet Point] --> B[AASRA Offline Engine]
    B --> C[Geodesic Red-Zone Synthesis: 0ms Client Buffer]
    B --> D[Obstacle Avoidance: Safe Tangent Waypoint Generator]
    B --> E[Capacity Allocator: Local Shelter Intake Matching]
    B --> F[Tactical Canvas: Zero-Tile Vector Grid]
    C & D & E & F --> G[Interactive Offline Leaflet Canvas]

Loading

1. Client-Side Geodesic Red-Zone Generator

When the device is disconnected from the backend API, the client calculates 25-point geodesic polygon buffers in JavaScript using the trigonometric geodesic expansion: $$\Delta\text{Lat} = \frac{R}{111.0} \cdot \sin(\theta), \quad \Delta\text{Lon} = \frac{R}{111.0 \cdot \max(\cos(\text{Lat}), 0.1)} \cdot \cos(\theta)$$ where buffer radius $R \in [1.5, 5.0]\text{ km}$ scales dynamically with the calculated hazard severity score.

2. Obstacle-Aware Evacuation Routing (Hazard Avoidance)

Standard routing algorithms attempt to draw straight lines or use road networks that cut directly through flooded or landslide-prone zones. AASRA's offline engine:

• Projects the direct vector from origin to candidate shelters and tests for geometric circle intersection against active Red Zone.
• If a route penetrates a danger buffer, it calculates tangent bypass waypoints with a 35% clearance buffer around the hazard perimeter.
• Visibly flags the corridor with an illuminated cyan dashed line and "🛡️ Hazard Avoidance: Bypasses Active Red-Zone Perimeter" notification.

3. Local GIS Persistence & Offline Testing Switch

• Automatic Hydration: Stores habitations, shelters, and synthesized buffers in localStorage and IndexedDB.
• 1-Click Test Switch: The UI includes an interactive "Test Offline Mode" button that lets commanders and evaluators simulate network blackouts without disconnecting their internet.

---

🚀 Quick Start & Installation

Prerequisites

• Node.js: v18.0.0 or higher
• Python: v3.11 or higher
• Git: Installed on your system
• Android Studio & SDK: (Optional, for running or building native mobile APK)

1. Clone the Repository

git clone https://github.com/gitgaurav-web/AASRA-Disaster-Intelligence-System.git
cd AASRA-Disaster-Intelligence-System

2. Frontend Setup (Web)

# Install dependencies
npm install

# Start local Vite development server
npm run dev

The frontend application will be live at: http://localhost:5173

3. Backend Setup (FastAPI)

# Navigate to backend directory
cd backend

# Create and activate Python virtual environment
# Windows (PowerShell):
python -m venv .venv
.\.venv\Scripts\Activate.ps1

# Linux / macOS:
# python3 -m venv .venv
# source .venv/bin/activate

# Install Python requirements
pip install -r requirements.txt

# Start FastAPI server
uvicorn main:app --reload --port 8000

Interactive API Documentation (Swagger UI): http://127.0.0.1:8000/docs
Complete backend guide & route index: backend/README.md

4. Mobile Application (Android APK)

The repository includes complete native Android build support via Capacitor:

# Build frontend web bundle
npm run build

# Sync assets to native Android project
npx cap sync android

# Open project in Android Studio
npx cap open android

• Direct APK Build: You can assemble the debug APK directly via ./gradlew assembleDebug in the android/ directory.
• Prebuilt Standalone APK: Pre-compiled and ready for installation at AASRA_Mobile_App.apk (and on your Desktop).

---

👥 Team Members & Contributors

Avatar	Member	Role	GitHub Profile
	Gaurav	💻 Team Member / Full Stack	@gitgaurav-web
	Indra Prakash	💻 Team Member / Developer	@indraprakash-756
	Kanishka Joshi	💻 Team Member / Developer	@kanishkajoshi32161
	Kavya	💻 Team Member / Developer	@kavya-DD
	Manas Singh	💻 Team Member / Developer	@Manas-uk
	Utsaw	👑 Team Lead / Developer	@utsaw-ik

---

⚖️ License & Disclaimer

• Educational & SIH Prototype: Red-zone risk screening buffers and hazard data serve decision support purposes. Official field interventions require validation from respective SDMA / NDMA / CWC / IMD authorities.
• Built with ❤️ for Disaster Preparedness & Smart Decision Making.