How I check a strike location with free satellite data, and when I refuse to publish

A reader asked how we decide whether a claimed strike location holds up. Here's the free toolkit and the actual decision rule. Nothing here needs a paid account.

The free sources

• Sentinel-2 (ESA, optical): 10 m resolution in the visible bands, a new image roughly every 5 days at mid latitudes. Big burn scars, collapsed large buildings, craters in fields. Not cars, not people.
• Sentinel-1 (radar): sees through clouds and at night. Change detection between two passes can show damage when optical is useless. Harder to read, I'm still learning.
• NASA FIRMS: active fire detections from VIIRS (375 m pixels) and MODIS, near real time. A thermal anomaly at the right place and time is supporting evidence. Not proof.

The process

1. Pin the claim: exact place, time window, what supposedly happened.
2. Geolocate any imagery that came with the claim (Marta has a great post on this in this topic).
3. Pull FIRMS detections for the area and window.
4. Compare the last clear Sentinel-2 image before with the first clear one after.
5. Look for independent sources: local reporting, other footage from different angles, official statements from more than one side.

Step 3 in pandas, from a FIRMS CSV download:

import pandas as pd

lat, lon = 48.0, 37.8      # the claimed location
df = pd.read_csv("firms_viirs.csv")
df["acq"] = pd.to_datetime(
    df["acq_date"] + " " + df["acq_time"].astype(str).str.zfill(4),
    format="%Y-%m-%d %H%M", utc=True)

window = df[(df["acq"] >= "2026-07-01T00:00Z") & (df["acq"] < "2026-07-03T00:00Z")]
near = window[window["latitude"].between(lat - 0.05, lat + 0.05)
              & window["longitude"].between(lon - 0.05, lon + 0.05)]
print(near[["acq", "latitude", "longitude", "frp", "confidence"]])

FIRMS times are UTC, and acq_time comes as an integer like 113 for 01:13, hence the zero padding. That one bit me once.

When I refuse to publish

• One source plus ambiguous imagery. That's an unconfirmed claim, and we say so or we say nothing.
• FIRMS hit but no visible change and no second source. Fires happen for a hundred reasons: industrial flares, field burning, a warehouse.
• Any doubt about the date of the "before" image. Clouds hide a lot, and a two-week gap hides even more.

The free stuff gets you surprisingly far. It also teaches humility: a 10 m pixel is the size of a small house, and most of what matters in a war happens at a smaller scale than that.