In February I read the headline that a solar storm had destroyed 40 Starlink satellites. I pictured something large and took the story as it was, without looking at a single number. This month I finally did, because I wanted to know 2 things. How strong was that storm compared with the rest of 2022? And was it special for Starlink or did other launches fly through the same weather without losses?
Kataoka and colleagues posted a preprint in March that is now accepted by a journal. Dang and colleagues published theirs in August. A NOAA and CIRES team wrote a third one together with SpaceX engineers. All of them are about how dense the upper air was during those days. I did not repeat their models, I only wanted to check the story against public tables that anyone can download.
The tables
Kp from GFZ Potsdam is one number for every 3 hours, from 0 to 9, that says how disturbed the Earth’s magnetic field is globally. NOAA’s storm scale is built on it: Kp 5 is a G1 storm, 6 is G2, 7 is G3. Kp goes in thirds, so 5+ means 5 and a third. The same file has ap, a linear version of Kp that can be summed. Its daily mean is Ap.
The satellite catalog from CelesTrak lists every tracked object with its launch date and, if it has reentered, its decay date. Launch times I took from the Wikipedia list of Starlink launches, because the catalog only has dates. The analysis is about 110 lines of Python 3.10 with only the standard library.
Before counting I wrote down what I expected. The February storm would be a weak one, with more than 10 similar days in 2022. Group 4-7 would be the only launch with a large early loss. And at least one other launch would meet Kp 5 or more in its first days and lose nothing.
How strong it was
Group 4-7 launched on February 3 at 18:13 UTC with 49 satellites. SpaceX puts new satellites into a low orbit of about 210 km and raises them later, so that a broken one burns up fast.
On February 3 Kp reached 5+ in the morning, hours before the launch. From the launch until the next afternoon it stayed between 2.7 and 4. Then on February 4 between 15 and 18 UTC it went to 5+ again, about 21 hours after the satellites were released. Kyoto’s real time Dst, the other usual storm index, bottomed at −66 nT on February 3 and −61 on February 4. A strong storm goes below −100.
Between January 1 and December 21 this year there were 35 days with Kp of 5 or more, about 3 a month. 20 of them reached 5+, like February 3 and 4. Only 5 days reached G2 and none reached G3. Some services count 5 minus as G1 already. With that rule the year has 52 storm days, 9 of them G2 and 2 G3. Either way February is not near the top. By daily Ap, February 4 shares the 5th place among the most disturbed days of the year and February 3 the 11th. For this point of the solar cycle it is an ordinary count: 2012, also on the rise of a cycle, had 35 such days too.
Where I made a mistake
Then I counted early losses. For every Starlink launch since 2019 I took the satellites that reentered within 30 days. My first table showed Group 4-7 with 6 early losses. A calm launch on March 19 also had 6.
It took me a while to look at the other column. The catalog has between 46 and 54 objects for every 2022 Starlink launch except 2. The September 11 launch shared the rocket with BlueWalker 3 and has 34. Group 4-7 has 17. Of these, 11 are the satellites that survived. The other 6 are objects named only by a letter, like STARLINK A. They reentered between February 6 and 12. The other 32 reentered before they got public catalog numbers.
49 minus 11 is 38, the same number the papers give. So the catalog does confirm the loss, but only as a hole. If you count reentries row by row, as I did first, you see 6. The March 19 launch went up in quiet weather and has all its satellites in the catalog. Its 6 reentries came 13 to 30 days later, a few days after G2 storms on April 10 and 14. Satellites that failed their checks and were deorbited on purpose would fit as well, which I did not check.
Was the weather special for Starlink
For each of the 33 Starlink launches in 2022 I summed ap over the 3 hour slots from launch to about one day after. Group 4-7 had a sum of 256 with a maximum Kp of 5+.
It is not the worst. The launch on January 19, 2 weeks before, went up at 02:02 UTC straight into a Kp 5+ slot. Its first day sum is 199 and all 49 satellites are in the catalog with no losses in the first month. On July 7 the sum was 266 with Kp 5+. On September 5 it was 319 with Kp 5. The solar radio flux at 10.7 cm tracks how much the Sun heats the upper atmosphere. It was 130 that day, about the same as 127 and 130 on February 3 and 4. None of these 3 launches lost a satellite in the first month.
The third expectation also held. That is the part I cannot explain with Kp. By the numbers I had, Group 4-7 had a bad day, but 2 other launches had worse days and survived. SpaceX wrote at the time that onboard GPS showed drag up to 50% higher than on previous launches. Kp does not show that. It is built from magnetometers on the ground and does not measure the density of air at 210 km above a specific orbit. The papers estimate that density with models of the upper atmosphere.
Maybe the release orbit of January 19 and September 5 was higher. For September 5, maybe SpaceX handled new batches differently after February. My tables cannot tell which.
What I did not check
The release altitude of each launch. First orbits are on the US Space Force catalog site, which needs an account. The public catalog keeps only the latest orbit. Without them the comparison covers only the weather. How exposed each batch was I cannot say.
The Kp for the last weeks of the year is still preliminary and a few late counts can move. I used 30 days as the line for an early loss. A satellite that failed in orbit and reentered later is not counted and a planned deorbit within 30 days is counted as a loss.
I did not check the Wikipedia launch times against SpaceX. An error of an hour would move a window by one 3 hour slot at most.
I also did not look at the ground side, power grids and radio. The only network I could check with public tables was the one in orbit. Why January 19 and September 5 survived the same weather I still do not know.