3 October 2026 · Adigrat University campus, eastern Tigray

True?unlikely~35%

Reported drone strikes inside Adigrat University

Updated 8 October 2026

One source carries this. ACLED's Ethiopia alerts for 3 October say the Ethiopian military "carried out four strikes inside Adigrat University, targeting wounded TDF and AFNM troops who were transported from Mekele to the university". Reuters covered strikes at Adwa and Sheraro the same week and does not mention Adigrat or the university at all.

Because the campus is a named, mapped place, this is one of the few claims here that imagery can actually test. OpenStreetMap holds the campus as a polygon, about 1.4 km by 1.0 km. Over that footprint, 1 October is completely cloud-free and 4 October is 6.5 per cent obscured — good enough, even though the wider town was half under cloud on both later passes.

Comparing them, the campus buildings are unchanged. No roof is missing, no structure has collapsed, there is no burn scar. Most of the apparent change is cloud shadow, and what survives masking sits in open ground rather than on the buildings.

That is a real result and a narrow one. Four strikes aimed at people inside buildings would not necessarily mark the roofs at all, and ten-metre imagery cannot see a hole in a wall, a vehicle, or a casualty. We can say the campus was not flattened. We cannot say it was not struck.

Claim

15204555809599 ← ruled out established →

Ethiopian forces carried out four drone strikes inside Adigrat University on 3 October 2026, targeting wounded fighters moved there from Mekelle.ACLED Ethiopia alerts, 3 October 2026roughly even chance

roughly even chance

The strikes caused structural damage to university buildings visible from orbit.Implied by reports of four strikes inside the campusunlikely

unlikely

Satellite

Before and after

Satellite image after the reported event — 4 October, 08:06Z · after

1.8 km across · Sentinel-2 (S2A) MSI L2A · 10.0 m per pixel · identical extent and scale

What this can and cannot show. Sentinel-2 true colour at ten metres per pixel, acquired about a day after the reported strikes. The catalogue lists this scene at 47.7 per cent cloud, which is the figure for the whole hundred-kilometre tile; over this campus window the classification layer flags 6.5 per cent. That difference is the reason this comparison was possible at all. The same limits apply as to the earlier frame: a destroyed building or a large burn would show, a munition strike on people inside a building need not show at all.

Observations

What the evidence shows

What we see

Every row of campus buildings visible on 1 October is present on 4 October, including the distinctive teal roof on the western side and the red-roofed blocks at the centre. No building footprint has disappeared, changed outline or turned to rubble tone, and no dark patch of the size and shape of a burn scar has appeared on any roof or courtyard. Of the samples darkening by more than four robust sigma, 83 per cent lie inside the area the 4 October classification layer flags as cloud or shadow. Restricted to the 94 per cent of the window clear on both dates, 415 samples of 21,560 exceed that threshold, in 77 groups; the two largest sit in open ground at the edge of the shadow area rather than on any structure, and the rest are isolated single samples.

What it is consistent with

The campus was not destroyed, and the thing that looked most like damage was cloud shadow. That is the whole of what this shows. Four munitions aimed at people sheltering inside buildings can leave no mark a ten-metre sensor registers, so this is consistent with the strikes having happened exactly as described and with their not having happened at all. It is published because a careful negative is worth as much as a careful positive, and because the same imagery read carelessly would have produced two burn scars that are really shadows.

What we see

27 September and 1 October are both cloud-free over this window, giving a quiet pre-event interval to measure against. Across it, 0.55 per cent of the window changes by more than four robust sigma. Across 1 to 4 October, which contains the reported strikes, the figure is 1.92 per cent once cloud-masked samples are excluded.

What it is consistent with

The event interval does change somewhat more than the quiet one, and that excess is not nothing. It is also not localised on buildings, and sits beside a six per cent cloud field whose shadow mask is known to under-reach. Thin unflagged haze is a sufficient explanation and we prefer it to a damage reading we cannot support.

Assessments

Claim by claim

Ethiopian forces carried out four drone strikes inside Adigrat University on 3 October 2026, targeting wounded fighters moved there from Mekelle.

Claimed by ACLED Ethiopia alerts, 3 October 2026 · source rating B3

roughly even chance 45–55%

ACLED reports it, specifically and with a date. We found no second source, no official confirmation and no denial.

Supporting evidence

ACLED

What complicates it

One source. Communities around Adigrat have been cut off from outside contact since late September, which makes independent checking harder and is itself part of why this stands alone.

What would change our mind

Any second account, an official statement either way, or dated photographs from the campus.

The strikes caused structural damage to university buildings visible from orbit.

Claimed by Implied by reports of four strikes inside the campus · source rating B3

unlikely 20–45%

The campus buildings are all still there on 4 October. Every row visible two days before the reported strikes is visible the day after, with no collapsed structure and no burn scar.

Supporting evidence

Satellite image · 1 Oct 2026 Satellite image · 4 Oct 2026

What complicates it

This rules out the campus being wrecked. It does not rule out the strikes. Four munitions aimed at people inside buildings need not mark a roof at all, and at ten metres per pixel we cannot see a hole in a wall, a vehicle, a crater under a tree or anyone hurt. Most of the change we did measure turned out to be cloud shadow, which is exactly how a careless reading produces a burn scar that is not there.

What would change our mind

Higher-resolution imagery of the campus, dated ground photographs, or a located account naming which building was hit.

Sources

Everything this rests on

research — ACLED Ethiopia alerts archive. The 3 October entry, headed "Drones strike Adwa and Adigrat in Tigray", states that multiple Ethiopian drone strikes targeted various locations in Adwa and that the military "also carried out four strikes insi archived copy

Deep dive

Sources & technical provenance

ACLED

ACLED Ethiopia alerts archive. The 3 October entry, headed "Drones strike Adwa and Adigrat in Tigray", states that multiple Ethiopian drone strikes targeted various locations in Adwa and that the military "also carried out four strikes inside Adigrat University, targeting wounded TDF and AFNM troops who were transported from Mekele to the university".

Original source Archived copy

Technical provenance

Timestamp basis: unknown
Archived at: 2026-10-06 00:46:00+00:00
Archive method: wayback-existing-snapshot
SHA-256: c32a6139f78b6075b3783afe285ed1422388920209ec904f401c4f726df6dc8e
Upstream sources: ["ACLED"]

Satellite image · 1 Oct 2026

Sentinel-2 true colour over the Adigrat University campus polygon, 1 October 07:56Z, two days before the reported strikes. The scene classification layer flags no cloud or shadow anywhere in this window.

Technical provenance

Captured at: 2026-10-01 07:56:00+00:00
Timestamp basis: platform
Licence: Copernicus Sentinel data 2026, processed by ESA. Free and open under the Copernicus licence; redistribution permitted with attribution.
Reproduction procedure: earth-search sentinel-2-l2a S2A_37PER_20261001_0_L2A, windowed to 39.458795,14.294595,39.474754,14.306711 (OSM campus polygon + 150 m).
Upstream sources: ["Element 84 Earth Search v1 / AWS Open Data sentinel-cogs"]
Artifact type: satellite_rgb
Sensor: Sentinel-2 (S2A) MSI L2A
Acquisition start: 2026-10-01T07:56:00Z
Source scene id: S2A_37PER_20261001_0_L2A
Source url: https://earth-search.aws.element84.com/v1/collections/sentinel-2-l2a/items/S2A_37PER_20261001_0_L2A
Source catalog: Element 84 Earth Search v1 / AWS Open Data sentinel-cogs
Artifact path: media/artifacts/225c2647ec69ba7d.png
Sha256: 225c2647ec69ba7d25047d2182964e9db5b125a8952b2d1323536e0bbcc320f2
Processing: ["earth-search STAC query (sentinel-2-l2a)", "window taken from the OpenStreetMap campus polygon plus a 150 m margin", "windowed read of asset 'visual' at native resolution (EPSG:32637)", "bands 1-3 as RGB", "PNG encode"]
Bbox: [39.458795, 14.294595, 39.474754, 14.306711]
Crs: EPSG:4326
Resolution m: 10.0
Cloud cover pct: 0.6
License: Copernicus Sentinel data 2026, processed by ESA. Free and open under the Copernicus licence; redistribution permitted with attribution.
Redistributable: True
Detectability note: Sentinel-2 true colour at ten metres per pixel over the mapped campus, 172 by 134 samples. One sample covers 100 square metres, so a lecture block spans a handful of samples and a vehicle less than one. This pair can show a destroyed or unroofed building, a collapsed wing, a burn area of roughly thirty metres or more, or heavy ground disturbance. It cannot show a hole in a wall, a crater under tree cover, a vehicle, a casualty, or who fired. Absence of change in it is not evidence that no strike occurred.

Satellite image · 4 Oct 2026

Sentinel-2 true colour over the same campus window, 4 October 08:06Z, the day after the reported strikes. The scene classification layer flags 6.5 per cent of this window as cloud or shadow, against 47.7 per cent for the whole tile.

Technical provenance

Captured at: 2026-10-04 08:06:00+00:00
Timestamp basis: platform
Licence: Copernicus Sentinel data 2026, processed by ESA. Free and open under the Copernicus licence; redistribution permitted with attribution.
Reproduction procedure: earth-search sentinel-2-l2a S2A_37PER_20261004_1_L2A, windowed to 39.458795,14.294595,39.474754,14.306711 (OSM campus polygon + 150 m).
Upstream sources: ["Element 84 Earth Search v1 / AWS Open Data sentinel-cogs"]
Artifact type: satellite_rgb
Sensor: Sentinel-2 (S2A) MSI L2A
Acquisition start: 2026-10-04T08:06:00Z
Source scene id: S2A_37PER_20261004_1_L2A
Source url: https://earth-search.aws.element84.com/v1/collections/sentinel-2-l2a/items/S2A_37PER_20261004_1_L2A
Source catalog: Element 84 Earth Search v1 / AWS Open Data sentinel-cogs
Artifact path: media/artifacts/aeab0db82eb466fe.png
Sha256: aeab0db82eb466fe23e2593f88b3f579583d239cdb82a75e0c09fd8763d10ab2
Processing: ["earth-search STAC query (sentinel-2-l2a)", "window taken from the OpenStreetMap campus polygon plus a 150 m margin", "windowed read of asset 'visual' at native resolution (EPSG:32637)", "bands 1-3 as RGB", "PNG encode"]
Bbox: [39.458795, 14.294595, 39.474754, 14.306711]
Crs: EPSG:4326
Resolution m: 10.0
Cloud cover pct: 47.7
License: Copernicus Sentinel data 2026, processed by ESA. Free and open under the Copernicus licence; redistribution permitted with attribution.
Redistributable: True
Detectability note: Sentinel-2 true colour at ten metres per pixel, acquired about a day after the reported strikes. The catalogue lists this scene at 47.7 per cent cloud, which is the figure for the whole hundred-kilometre tile; over this campus window the classification layer flags 6.5 per cent. That difference is the reason this comparison was possible at all. The same limits apply as to the earlier frame: a destroyed building or a large burn would show, a munition strike on people inside a building need not show at all.