How can modern satellites photos possibly be accurate to 20 centimeters in 10 kilometers?

3D WorldView-1 satellite view showing some of the ground survey points in PhotoSat’s Eritrea test area.

 From GIScafé by Gerry Mitchell (Photosat)

My intuition rebels at the notion that a satellite orbiting 750 kilometers above the earth, traveling at 7 kilometers per second could possibly take photos of the ground accurate to 20 centimeters in 10 kilometers.
When I realize that these satellites have scanning cameras which take their photos like push brooms, with the north end of the photo taken a few milliseconds before or after the south end, and that the whole satellite is vibrating while the photos are taken, my mind boggles.
It just does not seem that such high accuracy should be possible.
The satellite photos themselves, checked with tens of thousands of ground survey points, clearly demonstrate that the accuracy is real.

How do the satellites and cameras work?

We engineers and geoscientists in the commercial realm don’t actually know how these satellites and cameras really work.
Almost all of the technical details of the imaging satellites, their cameras and their ground processing stations, are classified.
Or if they are not classified they are certainly very difficult to discover.
I have had many conversations with satellite engineers who seem like they would love to tell me why their satellites perform so amazingly well.
However, sadly, they are not allowed to.
They simply can’t discuss classified technology with anyone who doesn't have the appropriate security clearances.

Whenever I have one of these conversations it always seems to me that part of what the engineer knows is public and part is classified but the engineer cannot be sure that he or she can remember what is still classified and what isn’t.
Since the engineers don’t have perfect memories it is safest to say nothing.
I have had satellite engineers decline to confirm information that is published on their own company’s websites.
This can make for some very awkward conversations.
We engineers and geoscientists in the commercial world only have access to the satellite photos themselves and very general public information about the satellites and their cameras.

How accurate are the satellite photos?

When the Digital Globe WorldView-1 ( WV1) satellite photos first became commercially available in 2008, PhotoSat acquired stereo photos for a test area in and Eritrea where we had over 45,000 precisely surveyed ground points.
When we shifted the WV1 photos three meters horizontally to match any survey point we were amazed to discover that all of the survey points within 10 kilometers matched the satellite photos to within 20 centimeters.
We eventually documented this discovery in an accuracy study white paper that we presented at conferences and is now published on our web site.

Now, eight years after that initial WorldView-1 accuracy study of the Eritrea test area, we have processed hundreds of satellite photos from the WorldView, Pleiades, SPOT and KOMPSAT satellites and have come to expect this incredible accuracy.
I am still in awe that this is possible and I still don’t know how it is achieved.
I do know that the photos are amazingly accurate.

WorldView-1 satellite photo over the PhotoSat test area in Eritrea.

The over 15,000 ground survey points used to confirm that the satellite photo accuracy is better than 20 centimeters in 10 kilometers are shown as black dots.

The completely black areas are survey points every 20 meters along lines separated by 100 meters.

Colour image of a one meter PhotoSat survey grid produced from the WorldView-1 satellite photos.  The ground survey points demonstrate that the PhotoSat grid is accurate to 35 centimeters in elevation.