1

Is there a general rule of thumb that relates the ability of satellite remote sensors to resolve an object and the spatial resolution of the sensors?

For example, a whale that's 15m in length and 7-8 meters in width can not possibly be identified on sentinel-2 imagery which offers 20m spatial resolution but can be identified easily on Worldview 3 images with 10cm spatial resolution.

In general what is the relative spatial resolution required to resolve an object of of a given size? Is it 1:10? 1:5?

13
  • 2
    You really need to read up on the basics of remote sensing. In your example you are ignoring the concept of spectral discrimination in leu of purely visual discrimination. If we relied solely on visual discrimination historically, there is very little that could have been accomplished. Aug 4 '20 at 23:25
  • I can remember mapping bores in outback Australia from 10m monochrome satellite imagery, it wasn't the bore that was seen but the 'green' patch around the associated trough and a track that would often terminate at the bore site, no such luck with whales which are indistinguishable in choppy seas. I think you'd have bigger problems mapping whales than the cell size, they like being under water, deep under water, except mother and calf that tend to stick to the surface, except for this you'd only see them if they're coming up for air on in particularly shallow water. Aug 4 '20 at 23:46
  • I think technology has progressed to a point where it has become possible to visually distinguish whales from satellite images. Here is a picture if you don't believe me: smithsonianmag.com/smart-news/…
    – Frank
    Aug 4 '20 at 23:51
  • 1
    We have successfully used fractional mixture modeling to identify elephants using 30m landsat data. Obviously you cannot "see" an elephant in a lansat pixel so, this is what j mean by relying in spectral, not visual discrimination. However, water interacting with active sensor wavelengths is an entirely different issue that, in fact would influence feature space but, this was not the the original impetuous of the question. Aug 5 '20 at 0:53
  • 1
    Thanks for the concern. We don't really need highly accurate GPS coordinates since we are just trying to identify the regions they like to go to
    – Frank
    Aug 5 '20 at 1:51
2

From the comments this question seems less about "resolving an object" than identifying individual objects from imagery. Here's a general approach.

Get a small sample of high-resolution imagery with which your identification algorithm works really well.

Next, reduce its resolution by subsampling. Your identification algorithm will probably work less well. Try different subsampling sizes to create different simulated source resolutions. Compute the accuracy of your algorithm at these resolutions. Keep track of false positives and false negatives or any other error measure appropriate to your scientific need. This subsampling is probably a best-case scenario, since lower resolution data might come from higher up and have more distortions than a purely subsampled high resolution image of the same area. But we're talking ball-parks here.

Now you have a rough measure of how good a particular resolution is at applying your algorithm, and can then decide whether data at a given resolution is good enough for you.

You can use this is a plan for further data collection - such as deciding whether to fly a balloon high and cover more area at a lower resolution or fly lower and cover less area but at a higher resolution. These efficiency trade-offs are your choice.

2
  • This approach definitely works but is a bit time consuming. I was more looking for a quick rule of thumb that will allow us to quickly determine whether or not something is feasible from from the get go without spending too much time on it.
    – Frank
    Aug 5 '20 at 16:02
  • But it seems like such heuristics don't exist in the community so I guess this is the next best thing.
    – Frank
    Aug 5 '20 at 16:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.