In conflicts around the world, entire neighborhoods can fall to rubble long before outside observers understand what has happened. High-resolution satellite images—often costing thousands of dollars per snapshot—have long been the primary way journalists, governments, and humanitarian organizations track destruction. However, those images are expensive, limited, and frequently obscured by cloud coverage.
Now, a new peer-reviewed study introduces a low-cost, near-real-time alternative that could transform how the world monitors war.
The research, published in PNAS Nexus, presents a fully unsupervised algorithm capable of detecting destroyed buildings using freely available radar satellite data from the European Space Agency’s Sentinel-1 constellation.
By analyzing subtle coherence changes between radar images taken just 12 days apart, the method can identify when structures have been damaged—without relying on high-resolution imagery, cloud-free skies, or any pre-labeled training data.
The breakthrough gives humanitarian responders, human rights monitors, and conflict researchers a scalable way to track destruction across entire cities—often as it unfolds.
Unlike earlier approaches that require prohibitively expensive imagery or large volumes of training labels, the new system works “off the shelf” using publicly accessible data. And because Sentinel-1’s radar can penetrate clouds and operate day or night, the tool offers a continuous view into conflict areas where optical satellites routinely fail.
Researchers validated their method across three real-world cases—Beirut, Mariupol, and Gaza—demonstrating that it can reconstruct both the location and timing of destruction events with striking accuracy.
“Automated detection of building destruction in conflict zones is crucial for human rights monitoring, humanitarian response, and academic research,” researchers write. “By statistically assessing interferometric coherence changes over time, unlike existing approaches, our method enables the detection of destruction from a single satellite image, allowing for near real-time destruction assessments every 12 days.”
Traditional approaches to detecting destruction rely on visual satellite imagery. This process can be precise, but also fragile. Cloud cover, smoke, nightfall, or a lack of available photographs can hamper real-time analysis. Worse, many machine-learning methods require extensive training data, which is rarely available in the early stages of war.
Sentinel-1’s synthetic aperture radar (SAR) solves many of these problems. SAR doesn’t need sunlight or clear skies, and it records fine-grained changes in ground structure by comparing returning wave patterns over time.
The researchers leveraged a statistical technique—Interferometric SAR (InSAR)—that measures how consistent an area appears across multiple radar passes. When a building collapses, coherence between images drops sharply.
The team then combined this with a method that allows an algorithm to tell when a building’s radar “signature” breaks down in a way that strongly suggests destruction.
“Our approach is fully unsupervised and hence does not require any labeled training data, making it applicable in scenarios where ground truth data are sparse or even entirely unavailable,” researchers write.
This gives analysts a powerful tool not only for well-documented conflicts, but also for places where international reporting is restricted or slow to emerge.
The researchers first put their system to the test by examining the 2020 Beirut port explosion. Though unrelated to armed conflict, the disaster—triggered by the accidental ignition of 2,750 tons of ammonium nitrate—produced one of the most powerful non-nuclear blasts ever recorded, leveling entire warehouse districts in an instant.
Such a sudden, well-documented event offered an ideal proving ground for assessing whether the algorithm could detect destruction within the correct 12-day satellite window.
According to the study’s visualizations and classification results, nearly all buildings manually annotated as destroyed exhibited extremely low P-values, signaling strong evidence of structural collapse. The algorithm also detected additional damaged structures in a near-perfect radius around the blast’s epicenter.
This precision in identifying not just where but when destruction occurred is one of the method’s most significant advantages over earlier multi-image or supervised approaches.
The second test involved Mariupol, one of the most heavily bombarded cities in Russia’s 2022 invasion of Ukraine. Unlike Beirut’s singular event, Mariupol saw weeks of escalating strikes.
The algorithm captured this progression clearly, detecting how destruction spread from one district to another as the siege intensified. While ground-truth labels for Mariupol are limited, the model still successfully identified thousands of destroyed structures. “Based on the optimal classification threshold, we estimate that 2,437 (22.22%) out of 10,964 buildings were completely destroyed in Zhovtnevyi district, with likely many more damaged,” researchers note.
This level of detail—tracking destruction wave by wave, building by building—has significant implications for documenting alleged war crimes, verifying humanitarian reports, and open-source reporting on conflicts.
The third and most extensive case involved the 2023–2024 Israel–Hamas war in Gaza. Over several months, the algorithm processed 28 radar image pairs to map evolving damage patterns across the entire Gaza Strip.
The results mirror the war’s shifting geography—from widespread initial airstrikes to concentrated destruction following evacuation orders in the north, and later, ground incursions into Gaza City, Shuja’iyya, and Khan Yunis.
The authors compared their damage estimates to those published by UNOSAT and found close alignment—even though their method uses lower-resolution imagery and no supervised training.
Researchers emphasize the cost advantage of their system, noting that high-resolution optical monitoring for Gaza over the same period would cost up to $511,000, whereas Sentinel-1 radar data is free.
Because the method is unsupervised and open-data based, anyone—from NGOs to academic researchers to investigative journalists—can apply it globally.
This broad accessibility is potentially transformative. International observers often face ambiguity or propagandasurrounding destruction claims. With a transparent statistical framework—not just a black-box neural net—the system provides calibrated certainty values for each detected change.
Rather than simply declaring “destroyed” or “not destroyed,” it estimates the algorithm’s confidence, enabling investigative teams to prioritize areas for further verification.
Despite promising test results, researchers acknowledged that the system still has some limitations. Radar’s lower resolution means the method struggles with subtle or partial damage and is more reliable for larger structures. False positives may occur due to repairs, vegetation change, or snowfall.
Nevertheless, the system consistently identifies large-scale destruction—even in complex, fast-moving conflicts. Researchers envision creating a global dashboard that updates automatically every 12 days, providing near-real-time visibility into war zones worldwide.
Ultimately, the study’s findings point to a profound shift in how the world documents and understands conflict. In an age where information travels faster than governments or militaries can contain it, the physical realities of war are becoming increasingly difficult to obscure.
With free, globally available satellite data and transparent analytical tools, the destruction once hidden behind front lines or filtered through propaganda can now be measured, mapped, and verified in near real time.
The researchers’ method does more than identify ruined buildings—it signals a future in which the consequences of conflict are visible to anyone with an internet connection, not just those with access to expensive intelligence assets.
As these systems mature, the ability to independently assess what is happening on the ground grows stronger. In that sense, this study marks another step toward a world where the human and material costs of war can no longer be kept in the shadows.
“Combining remote sensing techniques with robust statistics, our study introduces an unsupervised algorithm that uses freely available Sentinel-1 radar imagery to detect destruction with uncertainty estimates,” researchers write. “Tested across three real-world case studies, our method is able to reconstruct the chronology of destruction events. By leveraging open data, we democratize access to critical tools for conflict monitoring and assessment.”
Tim McMillan is a retired law enforcement executive, investigative reporter and co-founder of The Debrief. His writing typically focuses on defense, national security, the Intelligence Community and topics related to psychology. You can follow Tim on Twitter: @LtTimMcMillan. Tim can be reached by email: tim@thedebrief.org or through encrypted email: LtTimMcMillan@protonmail.com.