A new method makes it possible to assess, in near real time, the extent of damage to buildings in war zones without resorting to costly commercial satellite imagery. The technique, developed by Daniel Racek and his colleagues at Ludwig-Maximilian University of Munich (LMU) and the Technical University of Munich (TUM), also makes it possible to track the progression of such damage across a region during a conflict using only freely available radar data.
This new method relies on an algorithm that analyses images from the Synthetic-aperture radar (SAR) on the European Space Agency’s Sentinel‑1 mission to identify destroyed buildings in conflict zones over 12-day periods. It statistically evaluates the visual changes observed at different sites over time. What makes it unique? It requires neither training data nor expensive imagery. Radar is particularly well suited to such an application because, unlike optical satellites, it can operate through clouds and even in the dark.
An interferometric technique
Researchers compare SAR images of the same area obtained on several occasions during each 12-day period. They then analyse these different images to calculate what is known as the coherence measure. This coherence indicates the degree of similarity between the radar signals: a sudden drop in coherence suggests that buildings have probably suffered structural damage.
To ensure that these changes can be distinguished from random fluctuations, the researchers used two statistical analysis techniques: non-parametric median regressions; and outlier-robust estimates. For each pixel in an image, they calculated what a “profile” or “pattern” of normal variation should look like, then quantified the deviations from this pattern using probabilities known as “P-values”. Lower P-values indicate a higher probability that a building has been destroyed.
“Distinguishing between different patterns is one of the main challenges facing this type of change-detection technique,” explains Daniel Racek. These techniques are already used to identify changes to the Earth’s surface and have applications in fields such as urban development, agriculture and land-cover monitoring, he adds.
Three case studies
To test their approach, he and his colleagues applied it to three case studies: the 2020 explosion at the port of Beirut; the 2022 invasion of Mariupol in Ukraine by Russian forces; and the ongoing war between Israel and Hamas in Gaza, which began in October 2023. Three situations that are very different from one another.
Their algorithm achieved an accuracy rate of 86% for the Beirut explosion and correctly identified a total of 361 destroyed buildings. This figure represents more than 5% of all buildings in the analysed area. Although not the result of an armed conflict, the explosion had similar effects, with considerable damage concentrated in a densely populated area.
Their algorithm achieved an accuracy rate of 86% for the Beirut explosion and correctly identified a total of 361 destroyed buildings.
The algorithm also revealed that more than 2,400 buildings had been destroyed in the Zhovtnevyi district of Mariupol. This represents more than 20% of the buildings in that area. Unlike in Beirut, explains Daniel Racek, the damage in Mariupol unfolded over several weeks, which allowed us to test our method’s ability to detect it over a prolonged period.
“During the first four days of the invasion, the centre of Mariupol remained virtually intact,” he says. Over the following weeks, however, we observed that the destroyed areas were shifting from the north-west towards the south-east of the city centre, as the Russian army destroyed most of the city through bombardment.”
As for Gaza, their calculations were consistent with the UN’s satellite analyses, although the latter were slightly more conservative. “Here, we tracked the progress of the war over several months,” explains Daniel Racek. “We have shown the P values for pixels representing buildings for the period from 18th September to 11th December 2023. Following the initial attack on 7 October, our results clearly show widespread air strikes across the Gaza Strip. The subsequent calls for evacuation from northern Gaza reflect the fact that the damage was mainly confined to that area. Then, as the Israeli ground offensive unfolded, the damage closely followed the movements of the troops, first in Gaza and then in Shuja’iyya and Khan Yunis.”
An objective assessment of damage
Prior to this study, such analyses were only possible using high-resolution commercial imagery. Not only is this expensive, but it is not always immediately available for war zones, particularly at the outset of a conflict. In such cases, the necessary pre-conflict imagery is lacking, making any comparison or reliable detection of damage impossible.
As previous research also highlights, large-scale automated detection of damage caused by wars and conflicts makes it possible to investigate issues that were previously difficult, if not impossible, to address, he adds. It can provide information on the types of buildings that have been targeted and destroyed, such as schools, hospitals and other essential infrastructure. It can also indicate when these buildings were targeted. This information not only facilitates academic research but is also particularly useful for governments and international organisations, as it could help to direct humanitarian aid and plan for post-conflict reconstruction.
It was important for us to make this detection accessible to a wide range of users and not just to those who can afford to buy commercial imagery. This is one of the reasons why we undertook this study.
Armed conflicts and wars are, unfortunately, becoming increasingly frequent, and it is extremely important to detect the destruction of buildings to observe and measure the consequences of these wars. “It was important for us to make this detection accessible to a wide range of users (researchers, journalists, international organisations, etc.) and not just to those who can afford to buy commercial imagery. This is one of the reasons why we undertook this study.”
“Let us hope that, in future, increasingly ‘ready-to-use’ tools that do not require in-depth knowledge of data or the algorithms underpinning them will be developed to highlight the devastation caused by war. This information could then be easily incorporated into government decisions and news articles, thereby raising awareness of the reality of what is happening in conflict-torn regions.”
