Satellite analysis finds 274 damaged structures at US bases in Iran war

Pentagon has declined to give specific assessments of base damage

The Wall Street Journal has identified 274 damaged structures at 24 installations with a U.S. military presence across the Middle East during the first six months of the war with Iran — an accounting the newspaper described as the most comprehensive public tally of that damage to date.

The count fills a gap left by official silence and wartime imagery restrictions. The Pentagon has declined to give specific assessments of base damage, and major U.S. commercial satellite companies stopped releasing high-resolution imagery of military installations in the region early in the war. Journal reporters instead developed methods for analyzing lower-resolution public imagery from European Union satellites, inspected tens of thousands of images and reviewed NASA data tracking fires, Iranian state media and other sources.

The analysis shows where damage appeared and roughly when, not what caused it, though Iran announced strikes that coincided with nearly all of it. Many of the installations are shared with, or owned by, allied nations, so not every damaged building is American or used only by U.S. forces. Because the work relies mainly on lower-resolution imagery, the Journal said it almost certainly undercounts the damage.

To develop and validate its methodology, the Journal consulted five independent analysts: Brady Africk of the American Enterprise Institute, a conservative-leaning think tank; Sam Lair of the Foreign Policy Research Institute, a national security think tank; and Joseph Bermudez Jr., Jennifer Jun and Chris Park of the Center for Strategic and International Studies.

Lair, who reviewed and confirmed the Journal’s findings, said: “The restrictions on satellite imagery in the region have made it pretty hard to get a real sense of the scale and extent of the damage done by Iran.”

Reporters began with a list of more than 40 Middle East installations with a U.S. military presence. Some came from a 2024 Congressional Research Service report. Others were named in more than 5,400 posts the Journal archived from four Iranian state-affiliated media outlets since the war began: the Iranian Students’ News Agency, the Islamic Republic News Agency, Defa Press and Sepah News.

To search those installations for damage, reporters used publicly available imagery from the European Space Agency’s Copernicus Sentinel-2 mission, which captures images roughly every five days at a resolution of about 10 meters per pixel. They then implemented a statistical method called anomalous change detection. For each installation, the algorithm first established a baseline distribution from imagery taken one year earlier, from February through August 2025, and then flagged spots in imagery from the start of the war on Feb. 28 through its six-month mark on Aug. 28 that fell far outside the baseline — more than 30,000 locations in all.

Reporters reviewed the flagged locations in a custom-built application, comparing imagery from before and after each detected change to identify damage. They looked for telltale signs of damage, such as irregular dark spots, across several bands of light, including infrared wavelengths invisible to the human eye. Signs of damage had to persist for at least two weeks, which ruled out short-lived changes such as vehicle movements, and reporters checked earlier available images to pin down the window in which each instance of damage first appeared.

Each damage assessment was corroborated with a contemporaneous fire or sudden flare in brightness detected at the same spot by NASA or EU satellites, on-the-ground imagery posted publicly and verified by the Journal, or higher-resolution satellite imagery showing damage. For 26 buildings, the corroboration came from a statement by Iranian state-affiliated media of a strike on that installation.

To sort the damage into the war’s phases — before, during or after the ceasefire — reporters identified the date of the first image in which each building appeared damaged and, when the window spanned phases, narrowed the date using fire, flare and Iranian-claim data.

Some damage sites contained multiple buildings. To count them, reporters used early- and pre-war high-resolution images captured by California-based Planet Labs. They also measured each damaged building’s original single-story square footage and found the average size fell from 12,000 square feet before the ceasefire to less than 5,000 after it.

To identify building types, reporters consulted current and former servicemembers, analysts and Defense Department regulations. Fifty-four of the structures were identified using only high-resolution satellite imagery or on-the-ground footage.

Iranian state-affiliated media claimed a strike that coincided with damage to more than 250 of the structures identified by the Journal. Iran-affiliated outlets also often release high-resolution images of allegedly damaged structures, which the Journal did not use in its verification.

To analyze Iran’s claims, reporters used a large language model to extract from each post which installations Iran said it had targeted, any specific claims about what on each base was damaged and the date of the attack. Statements in Farsi were machine-translated, and a human translator checked a 10% sample for accuracy; all were without meaningful errors, the Journal said. Reporters also manually checked a random 20% sample of the extracted claims and found no errors that affected the analysis. They calculated the share of claims that named a specific type of building and, among those, the number that cited troop housing.