After nearly two months under siege by Russian forces, what was left of the Ukrainian city of Mariupol?
Using freely-available satellite data from the European Space Agency, we detected damage to building across the city. The location of damage was matched up to individual buildings using Open Street Map. We verified the technique by comparing a list of damaged buildings from earlier in the war, compiled by the UN, to the results from our pipeline using satellite data from the same time.
We found that nearly 45% of the built up area had suffered damage, including 33% of residential buildings.
The article generated 51.5K views. It was in the top 10 most read of all of our Graphic Detail stories for the year. The story was widely shared on social media.
The story also generated discussion with other researchers in the fields of remote sensing and satellite intelligence. One researcher from UCL (Ollie Ballinger) is writing a paper about responsive remote sending in war zones and is using the piece as a case study. He has replicated and extended the pipeline for other dates (see link 2). Another company who are working on using SAR imaging in African war zones, Masae Analytics, got in touch to discuss the technique also, and offered to share data and insights for future stories.
We relied on freely available data and tools throughout this project.
Our main data source was the synthetic-aperture radar (SAR) scanner aboard Sentinel-1, a European satellite. SAR emits microwaves which bounce off the surface of the Earth and return as an energy “echo”, which is then recorded.
Damage to buildings influences the returning echo signal. Rubble creates stronger signals than flat roads or roofs as more microwaves bounce back towards the satellite. This changes the amplitude of the returning signal. A collapsed building can also affect where the echo wave is in its cycle when it reaches the satellite — a change in the signal’s phase.
We analysed pairs of SAR images, two images from before the start of the war were used as a reference. We measured the coherence of these pairs of images, similar to measuring the spatial correlation, but accounting for both the amplitude and phase. Undamaged, built-up areas, tend to have high coherence between images, whereas greenery has low coherence.
We then subtracted this reference coherence map to another coherence map created using one image from before the war and one during (17th April). Large drops in coherence indicate damage to built-up areas. This analysis was performed using SNAP, a freely accessible remote-sensing analysis platform created by the European Space Agency.
To determine a threshold value for ‘damage’ we compared a coherence-difference map, calculated using images from earlier in the war, to a list of damaged buildings compiled by the UN from the same time (March 14th). The ‘damage’ map, and the geocoded UN data, were matched to building footprints on Open Street Map. The threshold for deciding which buildings were damaged was optimised for accuracy (92%) and precision (85%) of match to the UN data. This was done in QGIS and R.
Context about the project:
At the time of publishing this article very little was known about the fighting in Mariupol. Journalists and aid workers could not enter the city safely and so simple facts like the death toll or conditions in the city were unknown.
One of the only reliable sources of information was satellite images, but simply viewing visual images of small patches of city does not tell us much about its overall state. Our article was the first to quantify, remotely, the number of destroyed buildings and their composition.
While we are not the first people to analyse SAR images for the purpose of detecting building damage, I believe we are the first newsroom to do so. This is also the first time I have seen any method to threshold the damage maps to determine how much signal change constitutes damage. This is also the first time I have seen such damage maps merged with records of individual buildings and building types.
What can other journalists learn from this project?
I hope that this project demonstrated how journalists can do their own quick-turnaround, responsive satellite investigations using freely available data and tools.
The European Space Agency provides many online tutorials on how to perform this type of analysis. We also published a two-part series in our _Off The charts_ newsletter, talking though the process we used step-by-step (see links 3 and 4).
Rosamund Pearce, the visual data-journalist on the project, went on the ExploreExplain podcast to discuss making the maps for this project (see link 5).