Andy Garcia · freelance dev
FRENES
Back to the blog
Space

ESA Says Asteroid 2026 RW1 Was Found 7 Hours Before Impact. Its Team Timed the Entry to 0.1 s

October 20, 2026 7 min read

On September 6, 2026, an asteroid about one meter across entered Earth's atmosphere over the Indian Ocean. It is called 2026 RW1, and ESA's Near-Earth Object Coordination Centre lists it as the 13th asteroid ever detected in space before impact. In ESA's table, the time between its discovery and its impact is 7 hours 22 minutes.

With only a few hours left for follow-up, ESA's October newsletter says, its team rapidly obtained observations from telescopes in the US and Australia, determined the asteroid's rotation period and computed "its entry into the atmosphere to better than 100 m in position and 0.1 s in time." The prediction went out through Meerkat, ESA's automated warning system, and through a public web page, "leading to at least one ground-based detection of the resulting fireball from northern Australia."

I am not an astronomer. I build software, as the CTO of a LegalTech company built on Claude and as an indie developer. What I see here is an alerting pipeline exercised end to end, on an event that could hurt no one. So I read the newsletter, the paper that describes Meerkat, the account of the team that went looking for the fireball, and the raw observation record at the Minor Planet Center.

Why a few hours is the normal case

A one-meter rock is very faint. ESA's page for the object gives an absolute magnitude of 33.7 and a diameter of 0.5 to 1.1 meters, estimated from its brightness with an assumed reflectivity. An object that small becomes visible only when it is already close. In the Minor Planet Center's record, its reported magnitude goes from about 20 in the first images to about 15 in the last ones. Five magnitudes is a factor of 100 in brightness, mostly because it was getting closer.

ESA's list of all 13 cases shows discovery-to-impact times from 1 hour 46 minutes (2024 UQ) to 20 hours 46 minutes (2014 AA). None reached a full day. With 7 hours 22 minutes, 2026 RW1 sits exactly in the middle: it is the median of the 13.

All of them were small. "Asteroids of this size do not pose a significant hazard," the newsletter says. The event that pushed the field forward was never seen coming: in February 2013, a rock about 20 meters across broke up in the atmosphere near Chelyabinsk, Russia, and its shock wave injured more than 1,600 people. Twenty times the diameter means 8,000 times the volume.

A system that sleeps 20 hours a day and answers in under a minute

The paper, by Charlie Drury and colleagues, most of them at ESA, describes how the alert is produced. New candidate near-Earth objects reported by survey telescopes appear on the Minor Planet Center's NEO Confirmation Page. Meerkat checks that page about every 45 seconds. The hard part: a fresh detection is a short streak, a few positions taken minutes apart. It tells you where the object is on the sky and how fast it crosses it, not how far away it is or how fast it is coming toward you. (An everyday analogy: a light crossing a dark window shows you its direction, not whether it is a firefly or a plane.)

So Meerkat does not guess the missing two numbers. It lays out a grid of 160 by 160 possible distances and line-of-sight speeds, fits an orbit at each point, weights each one by how well it matches the observations, and propagates it 30 days forward. The share of the total weight that ends in a collision is the impact score. Then 1,000 random orbits drawn from that distribution map where and when the object could hit, and where telescopes should point next. Above a 10% impact score, external subscribers get an email.

Since version 2.0, the paper says, the average time to issue an alert is 56 seconds. Yet the system spends on average 20.5 hours a day asleep, waiting for new data. It is provisioned for latency, not for load. It also runs as two identical instances, so operations can switch if one fails, and a script compares the number of objects it is tracking with the count in NASA's independent Scout system, alerting operators by email and phone when they differ.

ESA does not say how it reached its 0.1 s, but the observation record points to one likely ingredient: geometry. When I queried the Minor Planet Center on October 20, the record for 2026 RW1 held 42 positions from seven stations. The last 27 came from two Australian telescopes: the Zadko Observatory at Wallingup Plain, in Western Australia, and Faulkes Telescope South at Siding Spring. Seeing the same object from another continent adds parallax. The paper describes the same effect for 2023 CX1, over the English Channel, where widely spread observations brought the uncertainty on the impact time under one second.

The first match was wrong

Seven hours 22 minutes is ESA's count from discovery to impact. By the time the alert reached the Global Fireball Observatory team in Australia, five were left. Hadrien Devillepoix's account begins on a Sunday afternoon in Perth, with a message from ESA's NEO Coordination Centre: Meerkat has flagged an impact. This time there is good follow-up, and the impact scenario firms up fast. "We have five hours."

They could not book a flight to South-East Asia, and a small plane would have arrived after the asteroid. Bali, about 500 km from the predicted point, was ruled out by low cloud. Broome, on the coast of Western Australia, had clear skies but sat 780 km away. The team found a local photographer through Broome Facebook groups, and he aimed his camera at the low northwest horizon, one long exposure every 15 seconds.

He reported a promising fireball. Looking at the shot, Devillepoix writes: "At this point I am nearly certain." Then came the checks: that fireball was 20 minutes too late and 45 degrees off in azimuth. Wrong event. A closer look at the sequence turned up a much fainter fireball at the right time, and its astrometry lined up with the impact corridor computed by NASA JPL's Scout. ESA's table records the validation for 2026 RW1 in one word: cameras.

Notice the shape of it. ESA's system raised the alarm, and the photograph was checked against the corridor from NASA's system. Two independent pipelines, one physical observation.

What the record does not show

The size is an estimate from brightness, not a measurement. The newsletter gives neither the rotation period nor the method behind the 100 m and 0.1 s, and as of October 20 I had found no paper on 2026 RW1.

The Minor Planet Center record holds two positions taken by Pan-STARRS 2, in Hawaii, on September 5, about 20 hours before the Mount Lemmon Survey observation that is credited as the discovery, on September 6. ESA's table lists that first observation too. Nothing I read says whether those positions were reported that night or linked afterwards, so I draw no conclusion from them.

And NASA JPL's public fireball database, which lists detections by US government sensors, had no entry for September 5 to 8 when I queried it on October 20. The documented confirmation I found is the one from Broome.

What I take from it

  1. Measure latency from the event to the person who can act. Meerkat's 56 seconds is the computing step. The paper is candid about the step before it: 2024 UQ was discovered by an ATLAS telescope in Hawaii roughly two hours before impact, but its observations reached the confirmation page only after the predicted impact time. The software equivalent, as an analogy, is an error that sits in a log and never pages anyone.
  2. Put the confidence inside the alert. On very short observation arcs, almost every orbit fits equally well, and the impact score can occasionally climb past the 10% threshold for an object that may not even exist. Meerkat prints a fit-quality metric in the email, flags arcs shorter than 0.1 hour (six minutes) at the top, and sends an "All-clear" email when follow-up observations mean the object no longer meets the alert criteria.
  3. Compare against something independent. Meerkat checks its object count against Scout. The fireball was matched against Scout's corridor, computed by a different system from the one that raised the alarm. One system agreeing with itself proves little.
  4. Check a match on two coordinates before accepting it. Here either check alone was enough to reject the first fireball, but the right one was accepted only once its time and its trajectory both matched the prediction. I fact-check my own posts with Jev as a judge, and the hardest part of that habit is the same one: rejecting the first answer that feels right.
  5. Use harmless events as rehearsals. ESA calls this kind of detection "a valuable opportunity to validate" its planetary defence capabilities. A response process that runs only during real emergencies gets tested only when it matters most.

ESA's newsletter describes a chain of "highly automated systems, supported by specialised expertise and human oversight," with Flyeye telescopes to add a link upstream "in the near future." A chain is only as fast as its slowest link, and only as trustworthy as its last check.

A one-meter rock threatened no one. That is exactly why it was worth handling like the real thing.

Sources

A project like this one?

I design and deploy products like this. Let's talk.

Let's talk