What am I missing?
For this same reason, I don't see why the altitude of the satellite is important. But I've never really studied this so I'd be happy to be corrected.
Think about the optics in reverse. Imagine turning the satellite's sensor head into a hologram then sending it backwards through the optics until it reaches the Earth, where it would be magnified to an enormous size and each pixel would be the size of a human.
This is not how traditional photography works. There is actually a one-to-one correspondence of motion of the sensor with displacement of the imaged object (in one direction). This is similar to how a "photo finish" is done [1] Since the satellite moves much faster than the plane and the displacement between the sensors is small, it seems that the plane should not have traveled very far.
If the arrays all share the same lens, that would cause this effect.
That is entirely plausible, I hadn't considered that before and had just been assuming that the satellite construction was 'ideal'. This sounds like the most likely reason for such an artifact to occur, though I still wonder whether this particular imperfection would be done given the high level of precision necessary for the entire project.
(EDIT: I have changed 'lens system' to 'lens', because I think the former might be misleading. All I meant by 'lens system' is that to make a decent camera, you have to have several lens elements along the optical axis. Subsequently, I realized 'lens system' might be taken as implying an array of lenses. What I actually mean by 'lens' is simply as in 'telephoto lens'.)
did a quick BOE calculation based on this assumption. The idea was to see if the implied focal length for the lens is plausible.
I actually used the green-blue pair for the calculation, because I think it is slightly more likely that each pair share a lens, than that they all do.
The green and blue images seem to be something more than a wingspan apart. We don't know what type of aircraft these are, so I picked a 737 as a mid-sized example. The latest models have a 34m wingspan, so let's say the aircraft has moved about 50m between images.
Assuming the aircraft are moving relatively slowly, because they are searching, I used 100 m/s (200kts) for their speed, which means there is 0.5 sec between the images.
The satellite is moving at "almost 5 miles per sec", so let's say 7 km/sec. That means the views of the two sensor arrays have a 3.5 km separation at the surface.
Zizzer gives the satellite altitude as about 630 km. The ratio of the focal length of the lens to the separation of the sensor arrays is the same as the ratio of the altitude of the satellite to the distance between what part of the surface each array is imaging: f/a = 630/3.5, or f = 180a (a is the array separation).
From the picture of the sensors, and principally using the connectors' pin-holes for a sense of scale, I guess the blue and green arrays are about 0.5 cm apart, implying f = 90cm, which seems plausible to me.
If so, that shared lens would be exactly the part of this that I was missing. Honestly, it doesn't look like there's a shared lens between the red and the other two, as you point out, but I guess it could be sharing a lens with the "unused lines" marked in the diagram that puts it at a distinct angle.
As a test of my understanding - would I be correct in saying that when the article says "The red and infrared rows are paired on one CCD array, while the green and blue sensors are on a second array, which is why the red plane often seems much further ahead than the green and blue planes." it's going a little too far: there's no reason that the red line would have to be inherently at a greater angle than the other three? I used to interpret this line as just meaning "the green and blue are closer together than the red is, so the red image will be further apart" but that should be insignificant.
I'm curious - does anyone know how they calibrate this kind of equipment or are the tolerances good enough that they don't really need to? Do they take passes over known objects to calibrate or do some algorithmic test for how well the three colors match up (in a non-moving example).
The picture shows that the red sensor is further from the green one than the blue is, simply because it is on a separate sub-assembly. If the infra-red image showed on the pictures, it would be as far from the red one as the green one is from the blue. Given the order of the arrays on the camera, the IR image of the plane would be ahead of the red image, meaning that the IR is the last of the four images taken.
Incidentally, the pictures show the aircraft are experiencing considerable wind drift, especially in the case of the top-left picture, where you can see the contrail. This indicates strong winds, and probably also a relatively low airspeed for the aircraft. The orientation of the waves suggests a head- or tail-wind, but to get that drift, it must be different at the airplane's altitude (which probably is not high if it is searching.)
Imagine a different satellite that instead of looking down looked out toward the Earth's horizon, as seen from orbit. And imagine it has two sensors, one which looks forward and one that looks backward, very nearly a 180 deg. difference. At the height of a satellite in low Earth orbit that could be a 1,000 km difference in what each sensor sees. Now, every millimeter the satellite moves the view moves as well, but because the views are separated by 180 deg. or 1,000 km of ground that means it takes much longer for one view to catch up to where the other was, even though the track of the sensor elements themselves in space do the same thing much quicker.
Or, just imagine if the satellite rotated so the sensor was sideways, parallel to the direction of motion. It still views the same multiple kilometer wide slice of the Earth, but now it reimages the same thing line as it moves, shifting just one pixel forward each image. The sensor is only a few cm wide so it takes a tiny fraction of a second for the satellite to translate the position of the rearward most sensor pixel to where the forwardmost sensor pixel was, but it does not take the same amount of time for the viewed slice of the Earth to translate by an entire frame, because the frame is many kilometers wide.
For a full color image, three linear sensors are used, each displaced some distance along the direction of motion. So the fact that the rays are spreading out in the perpendicular doesn't change the distance the satellite must travel to image the next part.
For example, your second paragraph doesn't hold. The large separation of rays only occurs in the direction perpendicular to that of travelling. Note how your method would be imaging the same place multiple times on each pass whereas a push broom scanner would capture each one just one.
For your third paragraph, I agree that this could be done but that's simply not how the sensors are aligned and that difference is important. In particular, it's the very difference that I'm trying to point out.
The important part is that it's also true for the different color lines as well. That's what I'm trying to explain.
The view of each color line may traverse 1mm for every 1mm the satellite moves in orbit, but that has nothing to do with the separation of the lines. The important thing is that the view of, say, the green color line will not be identical to the view of the red color line when they are at the same relative physical position in orbit (e.g. a fraction of a second apart in the orbital motion) because the view of each color line is at a different angle through the optics. And because of that the portion of the Earth that each line is viewing at a given time is separated by a considerable distance, and there is then a considerable time separation between when a given spot on the Earth is imaged by each color line.
My other examples are trying to provide hypothetical scenarios to help get you to understand why your preconceived ideas about how push broom scanning works have important gaps, they're not meant to explain how things work in this particular example.