I have a a study area of 2 Landsat scenes from 2013 which were classified for a specific natural community, and these areas have been isolated and extracted and digitized into vectors so that they can be integrated into a GIS. These communities exist as features all throughout the scene, kind of like islands, if you will. I am working with landsat data, so I want to see if these areas have grown/expanded through time, preferably over two or three decades.

What is the best way to accomplish this? I was thinking turning the natural communities polygons into a raster, using that as a mask for landsat data, and running a pixel change detection to see how they have changed. But would that give me results for if they had shrunk from previous times, since the mas would only represent current size? Or perhaps is fragstats a better option? I have no experience with fragstats. Thanks for your help and ideas

Edit for being too broad: the vectorization was for creating a detailed map for a GIS using a basemap with much higher resolution than landsat, and the change detection is a separate part of the project. I ran a supervised support vector machine classification in ENVI to obtain the results. The images I classified were corrected for atmosphere, so I will do that with the following temporal years.

Sorry for the misunderstanding, I suppose my main question is: instead of reclassifying and performing a change detection on the whole scene, what would be the most efficient way to determine land change for only areas of interest? The areas in question are isolated swamp ecosystems, from about 1-10 hectares in size. I'd like to see if they have grown or shrunk over time. Thanks and I hope this isn't too broad or difficult to understand.

  • What where your reasoning that you vectorized the landcover? What has fragstat to do with that? How did you classify your land cover? Why not do it again for the next temporal year and then clip your resulting raster to the area of interest?
    – Curlew
    Jul 30, 2014 at 21:59
  • Between scenes there are subtle differences due to haze, angle of the sun, recent rainfall... numerous factors. In order to compare them you must colour balance the images or you will return false positive/negative values. Rather than digitizing which is prone to interpretation use supervised classification in software like ENVI or ERDAS to detect the areas; no need to colour balance if you are using software but you cannot use training areas from one image to another. Jul 30, 2014 at 22:25

1 Answer 1


First you need to convert DN to reflectance and then perform atmospheric correction on your Landsat scenes to reduce the radiometric differences between time periods. I think the approach to use the polygons as a mask is not the right path to take. Rather, perform a binary classification of these scenes so that your wetlands have a value of 1 and everything else has a value of 0. For a basic change detection algorithm in ArcGIS, use the Difference tool in the Image Analysis window. Erdas Imagine has a very sophisticated feature called DeltaCue that walks you through the process of change detection using multispectral imagery.

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.