Earthquake Damage Assessment from Satellite Imagery: Rapid Response from Space

At 4:17 AM local time on February 6, 2023, a magnitude 7.8 earthquake struck southeastern Turkey. Within 6 hours, before dawn had fully broken and while aftershocks continued, the first SAR-based damage proxy maps were being generated from Sentinel-1 data. Within 24 hours, these maps were in the hands of search and rescue coordinators, showing them which neighborhoods had suffered the most structural damage.

That speed — from seismic event to actionable spatial intelligence in hours — represents decades of investment in satellite infrastructure, processing pipelines, and institutional coordination.

The Satellite Response Timeline

Hours 0-6: Event Characterization

Before any satellite imagery is available, seismic data provides the initial assessment:

• Earthquake location, depth, and magnitude from global seismograph networks
• USGS PAGER (Prompt Assessment of Global Earthquakes for Response) estimates potential casualties and economic losses based on shaking intensity and population exposure

This seismic assessment triggers satellite activation.

Hours 6-24: SAR First Response

SAR satellites provide the first imagery-based damage information:

Pre-event archive: Sentinel-1 acquires imagery globally every 6-12 days. Recent pre-event images exist for virtually any location.

Co-event acquisition: The next Sentinel-1 pass over the affected area (within 1-6 days of the event) provides the post-earthquake SAR image.

Coherence change detection: Comparing pre-event and co-event coherence reveals where buildings have collapsed or been severely damaged. Urban areas normally maintain high coherence (0.6-0.9); collapsed buildings produce coherence drops to 0.1-0.3.

The result is a Damage Proxy Map (DPM) — a spatial representation of where coherence decreased beyond the expected natural variation, indicating likely building damage.

Days 1-7: Optical Confirmation

Very-high-resolution optical satellites (WorldView, Pléiades, SkySat) are tasked to acquire imagery over the affected area:

• Sub-meter resolution enables visual identification of collapsed buildings
• Pre-event archive imagery provides comparison baseline
• Cloud cover may delay optical acquisition in some events

Days 7-30: Detailed Assessment

Comprehensive damage assessment combining:

• Multiple SAR and optical observations
• Building-by-building damage grading
• Affected population estimation
• Infrastructure damage assessment (roads, bridges, hospitals)

SAR Coherence for Damage Detection

The physics behind SAR-based damage detection:

Before earthquake: Buildings are stable, rectangular structures that maintain consistent radar scattering properties between SAR passes → high coherence

After earthquake: Collapsed buildings are piles of rubble with completely different scattering geometry → coherence drops dramatically

Key advantage: Works through clouds, at night, and doesn't require optical illumination. This is critical because many earthquake-affected areas experience dust, smoke, or weather that prevents optical observation in the first days.

Damage Proxy Map Generation

The standard DPM workflow:

1. Select pre-event SAR pair (two images before the earthquake) → compute baseline coherence
2. Select co-event pair (one pre, one post earthquake) → compute co-event coherence
3. Compute coherence difference: Δγ = γ_pre − γ_co
4. Apply statistical threshold: pixels where Δγ exceeds the expected natural variation are flagged as potentially damaged
5. Aggregate to building block or neighborhood scale

Accuracy and Limitations

SAR coherence damage detection achieves:

• 70-85% overall accuracy for binary damaged/undamaged classification at neighborhood scale
• Better performance in dense urban areas (many building pixels) than sparse settlements
• Effective spatial resolution of ~100m (due to coherence estimation window)

Limitations:

• Cannot distinguish damage severity levels (partial damage vs. total collapse)
• Low-rise buildings produce weaker coherence signals than high-rise
• Decorrelation from non-damage sources (vegetation change, soil disturbance) creates false positives in mixed urban-rural areas

Optical Damage Assessment

Very-high-resolution optical imagery enables detailed damage grading:

European Macroseismic Scale (EMS-98) Grades

Grade	Description	Visual Indicators from Satellite
D1	Negligible	Not detectable from satellite
D2	Moderate	Not reliably detectable
D3	Substantial to heavy	Partial roof collapse visible at <1m resolution
D4	Very heavy	Major structural damage, partial collapse visible
D5	Destruction	Complete collapse, building footprint changed

Practical threshold: Satellite-based assessment reliably detects D4-D5 damage. D1-D3 typically requires ground inspection.

AI-Assisted Damage Classification

Machine learning models trained on pre/post-earthquake image pairs can automate building-level damage classification:

• Input: Pre-event and post-event VHR optical images
• Output: Per-building damage grade prediction
• Accuracy: 75-85% for binary (damaged/not damaged); 55-70% for multi-class grading
• Speed: Thousands of buildings classified in minutes once imagery is available

Activation Mechanisms

International Charter "Space and Major Disasters"

A consortium of space agencies that provides free satellite data for disaster response:

• Activated by authorized users (national disaster agencies, UN)
• Multiple satellites tasked within hours
• Data delivered to responding agencies at no cost
• Over 800 activations since 2000

Copernicus Emergency Management Service (EMS)

EU-operated service providing:

• Rapid mapping products within hours of activation
• Reference maps, delineation maps, grading maps
• Standardized cartographic products for field use
• Historical archive of all activations

Sentinel Asia

Asia-Pacific regional mechanism for satellite emergency response, coordinated through JAXA.

Accuracy Reference: SAR Damage Proxy Maps vs. Field Surveys

Understanding expected accuracy helps calibrate trust in satellite-derived damage assessments:

Earthquake Event	DPM Accuracy (neighborhood)	Time to First Map	Primary Limitation
2023 Turkey-Syria M7.8	~78% (overall)	12–24 hours	Low-rise informal construction decorrelation
2021 Haiti M7.2	~72% (binary)	18–36 hours	Rural dispersed settlement pattern
2020 Izmir M7.0	~82%	6–12 hours	Dense urban; good S1 archive
2018 Lombok, Indonesia M6.9	~74%	24 hours	Cloudy; SAR was sole immediate source
2015 Nepal M7.8	~68% (first version)	48 hours	Processing pipeline not yet optimized

Common error patterns:

• False positives: Agricultural fields near urban areas; seasonal vegetation change misread as damage
• False negatives: Partially damaged buildings that maintain geometric coherence; low-rise adobe/earthen construction
• Resolution limitation: DPMs at 100–200m pixel scale cannot detect isolated collapsed buildings — they identify damaged neighborhoods, not individual structures

The 72-hour window: In earthquake response, the probability of finding survivors alive decreases from ~90% at hour 24 to ~30% at hour 72. Satellite damage maps produced within the first 24 hours are primarily useful for SAR (Search And Rescue) task force prioritization. Maps produced after 72 hours shift utility toward damage extent reporting, reconstruction planning, and humanitarian logistics.

Case Studies

2023 Turkey-Syria Earthquakes

• Sentinel-1 DPMs available within 24 hours
• Identified most heavily damaged areas in Antakya, Kahramanmaraş, and surrounding cities
• Optical imagery from multiple commercial and government satellites provided building-level assessment
• Over 50,000 buildings classified as damaged through satellite analysis
• DPMs guided search and rescue operations in the first critical 72 hours

2024 Noto Peninsula, Japan

• Dense cloud cover prevented optical observation for 48+ hours
• SAR coherence maps provided the only spatial damage information during the initial response
• Identified concentrated damage in Wajima and Suzu cities
• Demonstrated the operational necessity of SAR for winter/cloudy earthquake events

2015 Nepal Earthquake

• Pioneering use of crowdsourced damage mapping (OpenStreetMap + satellite imagery)
• Demonstrated that volunteer networks can scale damage assessment faster than institutional capacity alone
• Led to improved protocols for integrating volunteer and professional damage assessment

The Future

Higher resolution SAR: Upcoming SAR missions with finer resolution will improve building-level damage detection.

AI automation: End-to-end automated damage assessment pipelines are reducing the time from image acquisition to damage map from hours to minutes.

Real-time SAR: Planned SAR constellations will provide revisit times of hours rather than days, enabling near-real-time damage monitoring as aftershocks continue.

Integration with ground data: Combining satellite damage maps with ground sensor data (seismometers, IoT building sensors) and social media reports for comprehensive situational awareness.

The speed and coverage of satellite-based earthquake damage assessment have improved dramatically over the past decade. What once took weeks now takes hours. The satellite doesn't replace boots on the ground — rescue teams must still reach affected areas physically — but it tells them where to go first, and that prioritization saves lives.