4

I'm working on a project in which I have to compare vegetation cover changes. So we have 1950's aerial photographs which must be compared to actual satellite images. Images must be properly adapted for the satellite image, so I can measure the "before" and "after" accordingly. Do you all know any tutorial that teaches how to do this? I'll use Quantum GIS for these procedures.

The area I'm working is a landscape of ca. 50.000 ha. The aerial photographs were already digitized and yes, they're black-white. So I have several images, each corresponding to a certain amount of landscape. These photographs are not georeferenced, so I'll have to locate their coordinates manually. This is not a hard task, the hardest part for me is to properly deal with the errors / or distortions, photography angle, etc. in order to match to the satellite images. I think the resolution is quite OK to work with, since my idea is to verify how much vegetation advanced into grasslands in the past 60 years.

  • Welcome to the club. How big of an area are you researching? For example, the entire state of New Jersey, or perhaps the land cover of a flood plain? Knowing that might change how folks perceive your question. – elrobis Sep 10 '15 at 14:05
8

Your biggest issue will be that aerial photographs from the 50s may well be in black and white and this doesn't provide good basis for standard classification, such as the tutorial linked by @user3338197

Instead, you will have two paths open for you:

  1. Manual digitization (possibly outsourced)
  2. Object-based image analysis. The professional standard software is Definiens eCognition but various opensource options are available. See OTB and InterImage.

I unfortunately don't know of any tutorials covering object-based image analysis on black and white imagery, so I can't point at any links for that. Personally, I have had some success with this type of task using eCognition, but it is a lot of work to get it exactly right.

  • ah good point on the B&W images, I didnt think of that – ed.hank Sep 10 '15 at 13:42
  • 2
    To add to this answer: Your approach may depend on how large your area is. I have done a similar task manually, just digitizing areas that were forested vs. non-forested between past and present. But it was not a large area and I did not need lots of detail. Also, be aware of the accuracy limitations of georeferencing 1950's aerial photos for use with areal measurements (taking in to account different distortions in the photo) and error in georeferencing. – Tangnar Sep 10 '15 at 13:53
  • Hi Tagnar, I think that it's exactly what I want to do. How did you deal with the distortions/angles? – John Myers Sep 10 '15 at 15:36
  • If you're working with black and white photos you will want to look into performing a textural classification, not spectral. Do a quick search and you'll find several studies (usually in PDF format, so I won't link here). – Radar Sep 10 '15 at 18:13

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.