Radar damage test misfires at airbase

TLDR

• A Secjuice writer reran the radar test behind Bellingcat's Iran damage map at Borisoglebsk airbase and got misleading scores.¹²
• On 25 October the runway outscored the reported strike site; a week later the site scored below distant fields.¹
• Bellingcat says the tool narrows where to look, and the test's paper warns airports rarely score high.²³

A Secjuice writer using the name NetSniffer published a test on 2 November of the radar method behind Bellingcat's free Iran damage map. The writer ran a cut-down version at Borisoglebsk air base in Russia, where Ukraine's Navy claims six Su-34s were hit overnight into 15 October, and found the scores did not track the strike.¹

Why it matters: investigators lean on free Sentinel-1 radar tools when commercial imagery is restricted. The run shows how one unexplained anomaly in a single satellite track can create false damage signals, and how aircraft that come and go may leave no steady signal for the test to measure.¹²³

Driving the news:

• The Pixel-Wise T-Test compares each pixel's radar backscatter after a conflict began with a year of earlier passes. Its paper reports an AUC of 0.88 across eight Ukrainian cities and 0.81 for Gaza.³ Bellingcat applied it to Iran in April and cites an AUC of 0.87 across its full test sample.²
• The Secjuice run used Microsoft's Planetary Computer, about 2,000 point queries at twelve locations, with passes from 15 October 2025 to 1 November 2026 on three satellite tracks.¹ It read single raw 10-metre pixels, while the full test smooths speckle with a Lee filter and averages scores over building footprints.¹³
• On 25 October the runway scored 31.7, infield grass 29.4 and a north-west apron 29.6, while the reported strike coordinate scored 6.3, tenth of twelve points.¹
• One track (orbit 123) read roughly 16 dB brighter in the VH polarisation on 18 and 24 October, and on 18 and 24 September, across a band about 20 km wide. The jump is present in ESA's uncorrected files, and the writer says the cause is unknown.¹
• On 2 November, with a clean pass added, the runway fell to 2.0 and the grass to 1.5. The strike coordinate scored 2.8, below three fields 11 to 22 km away that scored 4.1, 3.8 and 3.3. The paper's example cutoff is 2.7.¹³
• One stand that scored 5.8 had already darkened on 20 September, a month before the strike. The writer guesses an aircraft was moved off it.¹

The big picture: Bellingcat says a high reading "doesn't make it definite" and that what the tool "excels at is highlighting and narrowing down areas" for corroboration by other imagery or geolocated footage.² The paper itself says airports and train stations are likely to have high pixel variance over time, so these areas are unlikely to reach high scores.³ The Secjuice writer concludes that, on this evidence, nobody should quote a reading from the test as a count.¹

The other side: the Secjuice run is not Bellingcat's tool. The writer notes the cut-down method, and that Borisoglebsk lies outside the area covered by Bellingcat's Iran map, so the test examined the underlying arithmetic and not the map.¹ None of the sources reviewed carries a response from Bellingcat or the paper's author.

Single-source note: every Borisoglebsk figure above comes from one self-run analysis by one writer and was not reproduced here. The cause of the track anomaly and the explanation for the stand are the writer's own reading.¹