Seeing sub-pixel cars (and planes) in the brand new Sentinel 2 Earth Observation Satellite images!
19 de October 2015
A new Earth Observation satellite with 10 meter resolution able to see elements that were not visible in equivalent satellites of the same kind. In moving elements, the information of the three pixels that form the three channels that form color images (red, green and blue) are seen separated and objects become visible even if are smaller than a single pixel. This could allow for opening new Research lines in the field of mobility and road transport.
When trying to memorize the new Sentinel fleet of the European Space Agency (ESA), you are told that Sentinel 2 is like a Landsat (and Sentinel 3 like an Envisat, Sentinel 4 like a Meteosat and Sentinel 5 like a Modis). Sentinel 2 was launched in June 23rd, 2015 and I have been just waiting to see the new images coming after the calibration phase. Finally, first same images are available and after looking carefully I was confronted with an image that clearly shows an airplane three times in red, green and blue over the landscape. We have not seen anything like that before in all Landsat imagery we have examined for decades. I was amazed, and since flying machines are one of my passions I immediately started to think about it.
The first explanation for it is that Sentinel 2 has 10m resolution (a pixel in the image is a square of 10x10m) in contrast with the 30x30m resolution of the Landsat series. This doesn’t seem very much but it makes the difference. In the image we see the four trails created by four engines in the airplane. There are not so many four engine planes; let’s assume that it is an Airbus 380. A380 has 79.7m the wingspan and this fits with the 8 pixel wide plane that we see.
Why three planes in three different colors? It seems that the three RGB are captured not exactly in the same moment. We see approximately a 12 pixel separation in the east-west direction between the red and the green plane, so 120 m separation. Supposing that the plane flies at 900km/h of cruise speed (that are 250m/s), this means 0.5 seconds of separation between red and green channel.
Now the magic of sub-pixel resolution is relevant: Can we see cars in the images? It is clear that a car is much smaller than one 10x10m pixel so in principle we cannot see them, but sometimes we have heart about sub-pixel resolution. Since the pixel intensity represents an average of the signal received from the 10×10 m square area, it could partially contain the signal coming from a car. This means that if a car is moving at 80 km/h (22m/s) in 0.5 it will move one pixel (10m) between red channel and green channel capture. This means that, a car at 80km/h will be seen as a sequence of a bluish pixel, a greenish pixel and a reddish pixel.
And, this is exactly what we see in the images with roads!
I have the intuition that this is more than a simple game but another demonstration that, with imagination, unexpected applications can emerge from the new sensors such as road transportation studies with Sentinel 2! (and forget about the planes; I’m sorry to say this but this is going to be only an annoying anomaly).
Acknowledgements to Oscar Gonzalez that make us realize this anomaly.