EARTH-GUARD

🛰

Multi-Spectral Imaging

Copernicus/Sentinel-2 MSI with 10m ground resolution. Bands 2–8A processed for fire radiative power, flood mapping, and vegetation stress indices.

🤖

AI Risk Classification

Gradient-boosted ensemble model trained on 12 years of Hellenic disaster data. Real-time inference at 340ms per scene with 97.3% recall on wildfire events.

⚡

Sub-Minute Alerting

Alert latency under 45 seconds from satellite pass to civil protection notification. Redundant communication paths via GEO relay and terrestrial fiber.

🇪🇺

EU AI Act Compliant

High-risk AI system per Annex III, Article 6. Full transparency register, human oversight protocol, and continuous accuracy monitoring per Reg. 2024/1689.

🌊

Multi-Hazard Detection

Simultaneous monitoring of wildfire, flood, air quality degradation, seismic activity, and coastal erosion across the Hellenic region and Eastern Mediterranean.

🔒

Secure Agency Access

Role-based access control for Civil Protection authorities, municipalities, and emergency responders. GDPR-compliant data handling with full audit trail.

EARTH-GUARD · OPERATIONS CENTER · HELLENIC REGION

LIVE

↻ 60s

ACTIVE ALERTS SIM

⚠ Alert scenarios are simulated for demonstration. Real-time data: weather, air quality and seismic feeds are live.

🔴 WILDFIRE · CRITICAL14:32 UTC

Penteli Forest Fire

📍 Penteli, Attica, Greece

Active wildfire detected via Sentinel-2 MSI. Wind-driven spread toward residential areas.

340 ha2,840 MW FRP62 km/h wind

AI Confidence

97%

🟠 FLOOD RISK · HIGH12:15 UTC

Thessaly Plain Flooding

📍 Larissa, Thessaly, Greece

Soil saturation at critical levels. Pinios River approaching flood stage.

94% sat.87% BF48h rain

AI Confidence

84%

🟡 AIR QUALITY · HIGH11:00 UTC

Athens Basin PM2.5 Spike

📍 Athens, Attica, Greece

PM2.5 concentration exceeds WHO threshold. Traffic + Saharan dust event.

162 µg/m³AQI 214NO2 high

AI Confidence

91%

HELLENIC REGION · SATELLITE OVERLAY 37.9755°N 23.7348°E

Wildfire

Flood

Air Quality

Earthquake

EU AI ACT STATUS

⚠️

Risk Classification

HIGH-RISK AI

Annex III, Art. 6(2) · Reg. (EU) 2024/1689
Critical infrastructure protection system

COMPLIANCE CHECKLIST

✓

Art. 9 Risk Management System

↓ tap to expand

Continuous risk assessment protocol active. Hazard identification covers wildfire, flood, seismic, and air quality events across 132,000 km². Updated quarterly per Art. 9(6).

✓

Art. 10 Training Data Governance

Training dataset: 12-year Copernicus archive (2013–2025), 840,000 labeled scenes. Bias audit completed Feb 2026. Data provenance log maintained per Art. 10(3).

✓

Art. 11 Technical Documentation

Full technical dossier maintained per Annex IV. Includes model architecture, hyperparameters, validation methodology, and operational constraints. Version-controlled in secure EARTH-GUARD infrastructure.

✓

Art. 13 Transparency & Explainability

SHAP feature attribution provided per alert. Confidence intervals displayed. Plain-language explanations generated for all emergency notifications. Public transparency register active.

✓

Art. 14 Human Oversight Protocol

Mandatory human review within 4 minutes for CRITICAL alerts. Civil Protection officer authorization required before issuing public emergency notifications. Override capability fully implemented.

✓

Art. 15 Accuracy & Robustness

Live accuracy monitoring: wildfire recall 97.3%, flood precision 89.1%, air quality F1 0.912. Monthly adversarial testing. Degraded-mode operation documented for cloud cover >85%.

Regulatory Framework

Built for EU AI Act Compliance

EARTH-GUARD operates as a high-risk AI system under Annex III of EU Regulation 2024/1689 — the EU AI Act. The system is deployed in critical infrastructure protection, triggering the full compliance regime under Article 6(2).

The compliance-by-design architecture ensures that every AI inference is traceable, explainable, and subject to mandatory human oversight before consequential decisions are made.

⚠️

HIGH-RISK AI SYSTEM

Annex III, Article 6(2) · Reg. (EU) 2024/1689

HIGH

Risk Class

EARTH-GUARD falls under the critical infrastructure management category. This classification requires conformity assessment, registration in the EU database, and ongoing post-market monitoring under Art. 72.

⏳

Art. 113 · Full Enforcement Deadline

Digital Omnibus on AI (COM(2025) 836) moved the Annex III high-risk deadline to December 2, 2027 · 00:00 EET

✓ READY

—

The Digital Omnibus on AI (COM(2025) 836) extended the full compliance deadline for Annex III high-risk systems to December 2, 2027. EARTH-GUARD is already fully compliant — 18 months ahead of the new deadline. Note: Art. 50 transparency obligations were NOT deferred and apply from 2 August 2026; machine-readable marking for pre-existing systems is required by 2 December 2026.

✓

Risk Management System

Art. 9 · EU AI Act 2024/1689

COMPLIANT

Implemented continuous risk assessment covering all operational hazards. The risk management system is reviewed quarterly by an independent safety committee, with findings logged in the transparency register. Risk mitigation measures include confidence thresholds, geographic filtering, and mandatory human review triggers.

✓

Data Governance Framework

Art. 10 · Training Data Quality

COMPLIANT

Training corpus: 840,000 labeled Copernicus scenes from 2013–2025. Bias audit conducted February 2026. Geographic and temporal representation verified across all 13 Greek administrative regions. Data lineage fully documented per Art. 10(3) requirements. Digital Omnibus on AI update: The Digital Omnibus on AI (COM(2025) 836) introduced an explicit legal basis for processing special categories of personal data (Art. 9 GDPR) for AI bias detection and correction. This directly supports the EARTH-GUARD Smart Citizen Network layer, enabling fairness audits using socio-demographic proxy data across all 13 Greek regions without a separate Art. 9 GDPR derogation.

✓

Technical Documentation

Art. 11 · Annex IV Requirements

COMPLIANT

Complete technical dossier maintained per Annex IV. Includes: model architecture diagrams, hyperparameter specifications, validation dataset composition, performance benchmarks across hazard types, operational constraint documentation, and version history. Available for notified body review.

✓

Human Oversight & Override

Art. 14 · Human-in-the-Loop

COMPLIANT

CRITICAL alerts require Civil Protection officer review within 4 minutes. HIGH alerts require acknowledgment within 15 minutes. All AI outputs clearly marked as AI-generated. Full override capability implemented — operators can suppress, modify, or escalate any AI recommendation without system resistance.

✓

Accuracy Monitoring

Art. 15 · Robustness Requirements

COMPLIANT

Live performance dashboards track precision, recall, and F1 per hazard type. Monthly adversarial testing under degraded input conditions. Automatic model drift detection triggers retraining workflow. Current metrics: wildfire recall 97.3%, flood precision 89.1%, AQ classification F1 0.912.

✓

AI Output Transparency

Art. 50 · Active from 2 Aug 2026 · Digital Omnibus on AI

COMPLIANT

Art. 50 transparency obligations were NOT deferred by the Digital Omnibus on AI (COM(2025) 836) and apply from 2 August 2026. All EARTH-GUARD AI-generated outputs are labelled at point of interaction. Machine-readable marking is implemented for all AI outputs; the 2 December 2026 grace period for pre-existing systems does not apply to EARTH-GUARD. Smart Citizen Network layer outputs carry the same Art. 50 marking from initial deployment.

System Architecture

From Orbit to Alert in Under 60 Seconds

Continuous satellite data ingestion, AI inference, and civil protection notification — fully automated with mandatory human oversight.

🛰

Satellite Acquisition

Copernicus/Sentinel-2 MSI data ingested every 5 days per point, supplemented by real-time FIRMS active fire data and Sentinel-1 SAR for flood mapping.

›

⚙️

Pre-Processing

Atmospheric correction, cloud masking, and geometric registration. Feature extraction: NDVI, NBR, MNDWI, LST. Scene tiling at 512×512px with 10m GSD.

›

🤖

AI Inference

Gradient-boosted ensemble (XGBoost + LightGBM) with uncertainty quantification. 97.3% wildfire recall. SHAP explanations generated per prediction for Art. 13 transparency.

›

🚨

Alert & Response

Civil Protection authority notified via secure API. Human review required for CRITICAL events. All actions logged to immutable audit trail. EU AI Act Article 14 compliant.

Performance Metrics

Wildfire Recall97.3%

Flood Precision89.1%

AQ F1 Score0.912

Alert Latency<45 sec

Coverage & Scale

Area Monitored340k km²

Active Satellites3 LEO

Ground Resolution10m GSD

Revisit Time5 days

AI Act Compliance

Risk ClassHigh-Risk

Annex IIIArt. 6(2)

Human OversightArt. 14 ✓

TransparencyArt. 13 ✓

Research Origin

EARTH-GUARD is an AI-powered satellite environmental monitoring system designed for real-time detection and early warning of natural disasters across the Hellenic region and Eastern Mediterranean.

The system integrates Copernicus/Sentinel satellite data streams with machine learning risk classification algorithms, delivering actionable intelligence to civil protection authorities, municipalities, and emergency responders — with full compliance-by-design architecture under EU AI Act Regulation 2024/1689.

Intellectual Property

Creator & IP Owner

Pelagia Eleftheriadou

Independent Researcher & Developer

Research focus: AI regulatory compliance frameworks for satellite-based environmental monitoring systems. Expertise in EU AI Act implementation, space law, and human-centred AI design.

MSc Thesis · ICT Law & Policy · Grade 10/10 · Aristotle University of Thessaloniki · 2026 · Pelagia Eleftheriadou

Built for EU AI Act Compliance

🏛 GSCP — Civil Protection Authority