I am playing around with ERDAS and was implementing image difference on two rasters separated by interval of time. I was using a blue band raster and I created a default threshold for percentage change to be 20% and also did a standard deviation of 2.0.

The image gives two completely different outputs for the two types of measuring the difference, using the percentage change and standard deviation. This is the image.

The green describes the lower extreme, describing the reduction in change and the red describes the higher extreme, indicating the increase in change. Though I am not able to understand the difference between the standard deviation output and the percentage output and what is the significance of using one over the other?

Hope I was able to explain the question well. In the image the upper one is the percentage change and the lower one is the standard deviation.

  • Can you provide more information about the input images? What kind of change detection are you running? Are they vegetation indices or some other type of index? Can you also provide more information regarding your workflow? – Fezter Oct 10 '12 at 1:05
  • I am simply using the Image Difference tool in Erdas and providing the input as the raster of 1991 and raster of 2002. The image difference then returns the difference raster with the percentage change of 20% as I am setting the threshold to 20%. After that I am using the Attribute Editor to provide green for all the rows below the lower threshold and red to the rows above the upper threshold. Thereby identifying the change in the land cover in that particular region – Sam007 Oct 10 '12 at 1:08
  • 1
    When you say raster of 1991 and 2002, are these straight aerial or satellite images? I don't think this is going to give you much information. What do you want to measure the difference of? Increase or decrease of vegetation or urban areas for example? Or different land classes? Subtracting one aerial/satellite image from another doesn't really tell you much. – Fezter Oct 10 '12 at 1:11
  • Well honestly I am going thru my Lab work and this is one of them. So even I am not completely sure. But this is a LandSat image and in the blue band. So I need to find out that what exactly does the change that is reflected by the green and red in the blue band mean. Like for example I have another image difference of NDVI which clearly indicates the change in the vegetation. But I am not sure what does the blue band explain? – Sam007 Oct 10 '12 at 1:23
  • Oh, I get it now. I think I misread your question. So you're subtracting the blue bands? Blue band mostly represents water reflection from memory. But your text book should tell you that. ;-) – Fezter Oct 10 '12 at 1:28

Though I am not able to understand the difference between the standard deviation output and the percentage output and what is the significance of using one over the other?

Those refer to the threshold used to decide whether there has been any change between two images. For percentage change, it uses a symmetric relative difference formula to measure change:

enter image description here

Where T1 is the value in the first image and T2 is the value in the second image. So for example, if the values are 100 before and 120 after and we used your threshold of 20%, then according to the formula

(120-100)/100 + (120-100)/120 * 100 = 30

Which means this particular pixel will be marked as changed if you used percentage change.

For standard deviation on the other hand, the threshold is the number of deviations away from the mean.

enter image description here

So if for example, the standard deviation for your data is 15 and you chose two standard deviations as the threshold, then only those pixels where the change values are +/-30 would be marked as changed. So our example, with T2 = 120 and T1 = 100 would result in:

120 - 100 = 20

Which is lower than +30 and hence, won't be marked as a changed pixel if we use the standard deviation threshold.


From the USGS FAQ: the blue band is useful for "Bathymetric mapping, distinguishing soil from vegetation and deciduous from coniferous vegetation".

It's my experience that you get better results by using band combination, however.

  • Thanks that is great. This was though just the answer to the first part of the question. Also wanted the explanation on when percentage change makes sense and when does using standard deviation make sense? – Sam007 Oct 10 '12 at 1:37

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.