4

I have 5 Landsat images for the same study area. One of them in 1975 (Landsat 2 MSS), two in 1985 and 1995 (Landsat 5 TM) and one in 2005 (Landsat 7 ETM+) and the last one in 2014 (Landsat8 OLI). I want to produce land cover for every time and produce Normalized Vegetation Index (NDVI) and compare the results between for the periods (1975, 1985, 1995, 2005, and 2014). LandsatTM5, Landsat 7 ETM+ and Landsat8 OLI are easily comparable because they have common bands. I am wondering how can I make change detection and deals with Landsat 2 MSS? To be informed the resolution for (Landsat 2 MSS) is 60m and the other sensors their resolution 30m how can I deal with this problem?

1

In terms of the question about the different resolutions, you would have to re-sample all of the imagery to a resolution of 60 meters, except for the Landsat 2 imagery of course.

As for accomplishing this, you will have to provide us with the software that you are using. Also, it is unclear whether or not you are asking for help with how to conduct the change detection/NDVI or simply how to solve the problem with the resolutions.

-- EDITS --

Ok, so based on your comment, there are a few things you need to do. I only have experience using ENVI Classic (which should be installed with 'complete' version of ENVI...just go into the start menu, navigate to ENVI, and you should see ENVI Classic in one of the subfolders), but either one should work for this. If you are going to use classic, make sure that your image to be resampled in loaded into ENVI. There should be a 'Transform' menu at the top; click on that, and then I believe it will be under 'Transform' then 'Resample' or 'Spectral Resample'. It's been a while since I've used the software so it may not be exactly where I said, but just do a little bit of searching around and you will find it.

Or, you can use the complete version of ENVI and follow this procedure: http://www.exelisvis.com/docs/SpectralResampling.html

Now, to get your NDVI's, you will want to click on 'Transform' again and click on NDVI (in ENVI classic). With this tool, you should be able to leave all parameters as default, but MAKE SURE that you correctly select the Red and NIR bands because they are numbered differently in your different sets of imagery.

Lastly, for your change detection, I'm not sure if you want change detection for NDVI values or for Land-Use/Land-Cover. If it's for NDVI, make all of your NDVI layers and then you will have to figure out how to basically subtract one layer from another to generate a new layer representing changes in NDVI values. If you are looking for Land-Use/Land-Cover change detection, you will first have to classify your imagery (which I will not explain as it is a whole process in itself), and then do some sort of subtraction. Don't get confused though when I say subtraction - this is just one of many methods of change detection that you can do within ENVI. I'm not sure exactly how to do change detection with the software, but this link should help get you started: http://www.exelisvis.com/docs/ChangeDetectionAnalysis.html

You can also do a simple Google search for 'ENVI change detection' that returns many results.

  • Thank you very much for your reply. The software that I use is ENVI5.1 so please how can I make this process (resample) in this software and how can I conduct the change detection/NDVI and how can I solve the problem with the resolutions??? your reply is highly appreciated. – A M D Dec 12 '14 at 2:06
-1

In remote sensing class we work a lot with NDVI to compare the changes in landcover over several dates. As far as I can tell, using imagery from sensors with different band width is going to throw off the final results slightly. If you are dead set on using these images and dates for whatever reason, small changes may be off. It should be fine to get a rough estimate of changes over a larger area. If you can change your dates, try to use imagery from the same sensor. This was the advice of my professor with 20 years experience working for NASA in remote sensing. I wanted to use landsat 5 and 8. The bands are not the same for these two.Also note NIR is different on landsat 7. I would suspect most the difference will be for the classes on the high and low end, water and high biomass. Normally, you would establish class breaks from one image. Then just apply those to all your additional images. In this case, you clearly will need to establish different class breaks for each sensor. Since this is a manual and subjective method, I see no way you can ensure the breaks are exactly the same.

Have you looked closely at the landsat 7 image? You say it is 2005, but the SLC broke in 2003. The images are mostly worthless and filled with holes after 2003. One of these damaged images is certainly not suitable for comparison with the others. It might be worth something on its own, but not for producing an NDVI landcover change matrix. Consider changing the date to sometime between 1999 and when the SLC broke(may 31, 2003).

Landsat 5

Band 3 0.63-0.69 30m

Band 4 0.76-0.90 30m

Landsat 8

Band 4 - Red 0.64 - 0.67 30m

Band 5 - Near Infrared (NIR) 0.85 - 0.88 30m

Landsat 7

Band 3 0.63-0.69 30m

Band 4 0.77-0.90 30m

  • 1
    But, with ETM+7, you are completely ignoring the interpolation image matching methodology that USGS came up with for the SLC off problem. Granted, this is an estimated, and not observed value, but is not without merit and does not completely negate NDVI images with the SLA error. Back in the day (pre ETM), it was common to correct for scan-line dropout. Thanks for pointing out the spectral differences, in the composite bands, across the landsat sensors although, you skipped MSS (RED 0.7-0.8 , NIR 0.8-1.1). – Jeffrey Evans Mar 14 '16 at 22:49
  • That is true. The worst part of ETM + 7 images is on the sides of the scene too. If the area of interest is in the center and it is going to be subset, then that area may be of very good quality. The only time I wanted to use ETM + 7, my area of interest was on the very edge of the scene. – GeographyNerd Mar 16 '16 at 6:01

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.