I was reading through this twitter/X thread a few days ago about the options for quantifying damages in conflict zones using satellite images. User @yohaniddawela put together a really good overview which is worth reading. The technology is called Interferometric Synthetic Aperture Radar (InSAR) and his overview got me thinking about how we could use this technology for damage prevention. SAR is a satellite imaging technique that’s reshaping the way we assess and protect critical infrastructure. Combined with advanced data processing techniques like deep learning, holds the potential to detect damage and assess infrastructure quality remotely, allowing for proactive steps to be taken before small issues escalate into significant problems. Kind of the same strategy as fibersensing technologies but from satellite imagery. In this article, we’ll explore groundbreaking research conducted by Caltech, which uses SAR data to detect damage in conflict zones and disaster-hit areas, and consider how such technology could transform proactive damage prevention and infrastructure quality assessments in the utility sector.

Understanding SAR and Coherence-Based Damage Detection

Traditional satellite images are photos of the Earth’s surface, capturing what’s visible. However, Synthetic Aperture Radar (SAR) goes a step further by using radar signals to measure the physical structure of objects. Unlike optical imagery, SAR can penetrate clouds and capture data day or night, making it exceptionally useful in situations where conditions may not be ideal for conventional imaging.

One of SAR’s key features is its ability to measure “coherence,” or the similarity between two images of the same location taken at different times. By comparing coherence values (ranging from 0, completely different, to 1, exactly the same), researchers can detect significant changes in the ground surface. When coherence drops sharply, it often indicates structural damage. This technique has been used effectively in various conflict zones and disaster areas, such as earthquake-hit regions in Italy, Iran-Iraq, and California.

Deep Learning for Predictive Damage Analysis

Caltech’s research leverages Recurrent Neural Networks (RNNs) to analyze a time series of SAR images, capturing data from before an event, such as an earthquake. The RNN model predicts coherence levels as if no event had taken place. By comparing these predicted values with actual post-event coherence, researchers can pinpoint areas with significant damage, as indicated by large deviations from the predicted coherence.

This method provides a more refined and accurate assessment of damage than traditional Coherence Change Detection (CCD), which simply compares two radar images before and after an event. Caltech’s approach includes probabilistic forecasts that calculate mean and standard deviation, enabling them to identify abnormal drops in coherence with a high degree of precision. The results from Caltech’s study were promising, showing that this approach correctly identified damage in 72% of flagged areas and correctly identified 56% of all damaged areas. That’s not perfect but it’s a good sign of what we can do with this technology.

Applications in Damage Prevention and Utility Infrastructure Protection

For the utility sector, this technology has immense potential. Utility infrastructure - such as pipelines, power lines, and water systems - often runs through areas at risk of natural disasters or urban development pressures. Damage prevention teams could use SAR data to detect early signs of degradation or external threats, enabling them to address these issues before they result in service disruptions or safety hazards.

Here’s how coherence-based damage detection could support proactive damage prevention:

• Real-Time Monitoring: SAR data could provide ongoing updates, giving damage prevention teams real-time insights into the structural health of both above-ground and subsurface infrastructure. This could be big for pipeline companies who do regular checks on their infrastructure.
• Early Warning Systems: Sudden drops in coherence could trigger alerts, allowing teams to inspect and repair potential damage before it impacts service.
• Resource Allocation: By pinpointing areas at risk, utility companies can focus resources where they’re needed most, increasing efficiency and reducing costs.

Beyond Emergency Response: Assessing Infrastructure Quality in Economic Development

Caltech’s work also showcases how satellite data can play a critical role in economic development, particularly in assessing infrastructure quality in remote or under-resourced areas. A recent study used SAR and optical satellite data to evaluate infrastructure quality in Africa, specifically in areas such as sewage, water supply, and electricity. By combining SAR data with ground-truth data from Afrobarometer surveys and optical data from Landsat 8, researchers were able to estimate the quality of essential infrastructure with surprising accuracy.

For communities and governments focused on economic development, this diagnostic capability highlights where infrastructure investments are most urgently needed, allowing for targeted improvements in quality of life and community resilience. In damage prevention, we’re committed to a proactive approach - anticipating and addressing risks well before digging starts. By identifying areas with vulnerable infrastructure, we can make smarter, more strategic decisions to protect what’s underground. It’s all about connected, forward-thinking planning that strengthens our infrastructure and safeguards our communities.

The Future of Damage Prevention: Integrating Satellite Data and Machine Learning

As we move toward a future where infrastructure assessment can be both proactive and precise, SAR coherence analysis, combined with machine learning, presents a promising way forward. This technology could help utilities and governments to understand vulnerabilities in infrastructure before they become critical, driving a shift from reactive repairs to proactive maintenance and resource allocation.

By harnessing these advances, we could foresee a future where damage prevention is integrated with global satellite networks, offering real-time insights that benefit communities, protect essential services, and reduce repair costs. For the damage prevention sector, adopting satellite data and machine learning represents not only a technological advancement but also a commitment to safer, more resilient infrastructure.

Conclusion

The Caltech research on coherence-based damage detection with InSAR and RNNs demonstrates how a data-driven approach could reshape infrastructure safety. Leveraging SAR data for immediate post-disaster assessments and ongoing infrastructure quality monitoring, combined with deep learning, offers utility companies, damage prevention teams, and government agencies a path to more proactive infrastructure protection. This model of predictive analysis can help identify potential damages before they happen, allowing communities to be safer and more resilient against future challenges.

A key part of this approach is the use of AI, specifically RNNs, which analyze coherence changes over time to grade infrastructure health and suggest areas of concern. By predicting coherence values and identifying discrepancies from expected patterns, these AI models can flag potential weaknesses. However, a common AI challenge, known as "data leakage," can hinder the reliability of these forecasts. If an RNN overfits by memorizing patterns instead of learning generalizable rules, it can produce artificially optimistic results rather than robust predictions.

Anthropic’s CEO discussed this issue on the Lex Fridman podcast, explaining how synthetic data generation could be a solution. By creating augmented or entirely new data sets - similar to how DeepMind trained AlphaGo - AI can avoid data leakage and improve its generalization capabilities. Integrating synthetic data and reasoning-based models with SAR and RNN technologies represents a promising future for infrastructure safety. These innovations enable a truly proactive approach, where critical assets can be monitored continuously and potential damage points identified well in advance, keeping our infrastructure and communities protected.

Check out some of Yohan's work here