Yorkshire Flood Response Digital Twin. 12 live MCPs. Cesium 3D globe. Real-time sensor fusion. BFT governance.
Setting: January 2026. River Aire overflows. Calderdale, Leeds, York affected. 15,000 properties at risk. Multi-agency response required: Environment Agency, Fire & Rescue, Police, Military Aid to Civil Authorities (MACA), NHS.
DEFONEOS role: Provide a single Common Operating Picture (COP) — 3D digital twin with real-time sensor data, AI analysis, and multilateral decision governance. Every action signed on an Ed25519 SIGIL chain.
Demonstrates: Civil-military dual-use. Same MCP infrastructure that serves ISR also serves flood response. Open source sovereignty. BFT-governed decisions.
Yorkshire flood zone with live sensor overlay. Drag to rotate. Click markers for data.
| Layer | Data Source MCP | Visual Representation | Update Rate |
|---|---|---|---|
| Satellite imagery | sentinel-hub-mcp | Pre-flood vs current Sentinel-2 overlay. Change detection in red. | 5 days (Sentinel revisit) |
| Flood extent | data-gov-uk-mcp + OS OpenData | Polygon overlay from EA flood warnings + predicted extent. | 15 min |
| River levels | data-gov-uk-mcp | Coloured markers: green (normal), amber (elevated), red (flood). | 15 min |
| Rainfall radar | met-office-mcp | Animated precipitation overlay. 6-hour forecast. | Hourly |
| Maritime | aisstream-maritime-mcp | Vessel positions. Humber estuary traffic monitoring for port closure decisions. | Real-time |
| Air quality | openaq-air-mcp | AQI sensors — detects contamination if flood water reaches industrial sites. | Hourly |
| Population density | ons-statistics-mcp | Heat map for evacuation priority. Census data overlaid on flood polygon. | Static (2021 census) |
| Camera feeds | rtsp-camera-mcp | Live CCTV from key infrastructure: bridges, weirs, pumping stations. | Real-time |
| News/OSINT | gdelt-news-mcp | Social media reports, news articles geolocated on map. | 15 min |
| IoT sensors | mqtt-bridge-mcp | Temporary flood sensors deployed by responders. MQTT topic → map marker. | Real-time |
| AI analysis | defoneos-sign-mcp + local model | AI-generated risk assessment per zone. Updated as data changes. | On-demand |
| Decision audit | defoneos-bft-mcp | BFT proposals and votes shown as events on timeline at bottom of globe. | Per decision |
Globe loads centered on Yorkshire. Flood polygon visible. River level markers showing red/amber. Satellite overlay shows extent. AI brief auto-generated: "15,000 properties at risk. River Aire peak expected T+6h."
Responder deploys temporary IoT sensors via MQTT. New markers appear on globe within 30 seconds. Real-time water level feeds now augment EA gauges.
DEFONEOS AI (local model on M2 edge node) analyses all 12 data sources. Generates zone-by-zone risk assessment. Identifies a nursing home within the flood polygon not on EA's evacuation list.
AI-generated recommendation: "Priority evacuation: Sunnymead Care Home, 42 residents, Calderdale." BFT proposal emitted to SIGIL chain. 33-agent council votes (demonstrating multilateral governance — no single point of decision).
BFT quorum reached (23/33). Decision ratified. SIGIL signed: "Evacuate Sunnymead Care Home. 42 residents. Priority: IMMEDIATE." This propagates to all connected agencies simultaneously.
AIS data shows vessel MV Arklow Arrow approaching Humber estuary. Flood conditions make port entry dangerous. DEFONEOS auto-generates alert. Port authority notified via CoT message.
Every action, every sensor reading, every AI inference, every BFT vote — all recorded on the Ed25519 SIGIL chain. Demonstrate the audit replay: "Show me every decision made in the last 5 minutes, who made it, and why."
| Stage | Component | Function |
|---|---|---|
| 1. Ingestion | 12 MCP servers (sensor, satellite, government data) | Each MCP wraps a data source in the standard protocol. JSON-RPC calls. Auto-auth. |
| 2. Processing | DEFONEOS Core (Python) | Data normalisation, deduplication, geo-coding. Feeds to Cesium via WebSocket. |
| 3. Visualisation | CesiumJS (browser) | 3D globe. Entities, polygons, billboards, polylines. Real-time updates via WebSocket. |
| 4. AI Analysis | Ollama (M2 edge node) | Local llama3.2 model analyses combined data. Structured JSON risk assessment. |
| 5. Governance | BFT council (33 agents) | Reviews AI recommendations. Votes. Quorum threshold. SIGIL signed. |
| 6. Audit | SIGIL chain (Ed25519) | Every action hash-chained. Replayable. Tamper-evident. Exportable for post-action review. |
| Metric | Value | How Measured |
|---|---|---|
| Sensor data latency | <2s from source to globe | MCP call timing in SIGIL chain |
| AI inference time | ~2s per zone assessment | Ollama timing on M2 |
| BFT quorum time | ~30s (33 agents voting) | SIGIL chain timestamps |
| Total end-to-end | <5s from data to decision | SIGIL chain audit |
| Network dependency | Zero (air-gapped capable) | All local models + cached data |
| Cost per scenario run | £0.00 | All open source + local compute |
| Step | Action | Time |
|---|---|---|
| 1 | git clone https://github.com/CSOAI-ORG/defoneos | 30s |
| 2 | cd defoneos && pip install -r requirements.txt | 1 min |
| 3 | python run_demo.py --scenario yorkshire-flood | 30s |
| 4 | Open http://localhost:8080 in browser | 5s |
| 5 | 3D globe loads with demo data. Interact freely. | — |
Requirements: Python 3.11+, 8GB RAM, modern browser. No GPU required for demo mode. No API keys needed (uses cached simulation data).
This demo is representative of DEFONEOS capabilities. For a live, interactive briefing with real-time data and custom scenarios, contact CSOAI Ltd.
The live system pulls from 30+ MCPs, not 12. The BFT council runs with real agents. The SIGIL chain is a live Ed25519 hash chain, not a simulation. We'll tailor the scenario to your operational context.