To identify damaged buildings, analysts from Vertical52 analyzed radar satellite data from before and after the fighting. They identified changes on the Earth's surface. To determine which of these were building damage, the data was automatically compared with building outlines. If the roof of a building—whether a residential building, office building, or shed—has changed significantly before and after fighting, this is considered damage.
Detecting building damage with satellites
The model has been validated for destruction in Gaza. Tests for the Ukraine results showed an accuracy of around 80 percent. The analysis is experimental with regard to Ukraine. It is possible that minor damage may be overlooked or intact houses next to destroyed ones may be incorrectly identified as damaged. In addition, certain types of destruction cannot be detected from the air. For example, if a building is completely burned out from the inside but the facade and roof are not severely damaged.
To avoid false detections, the model has been calibrated so that only major changes are classified as damage. Otherwise, there is a risk of classifying changes such as missing parked cars as damage. This is a rather conservative estimate. In the case of Mariupol, for example, the UN assumes significantly higher figures: 90 percent of multi-story buildings and 60 percent of smaller residential buildings in Mariupol have been destroyed.