-2

I performed an NDVI analysis using Qgis 2.18 'Las Palmas' of an area of Landsat 7 (wrs_path 168, wrs_row 061),date 2007-12-25 and Landsat 8 (same path and row) date 2017-12-28. I clipped both images to x max and x min (36.59588, 36.79113)and y max and y min (-1.29795 and -1.46917).thereafter i used raster calculator to perform the NDVI formula -NDVI=(NIR-RED)/(NIR+RED)for each image,to generate two image outputs,one for 2007 and the other for 2017.

The issue is that the NDVI values for vegetation of the 2017 output image are too low, with negative even and shockingly water has a positive value of +1, which is a dam located at latlong (-1.44070, 36.69519). 2007 values were sensible but why this weird values in 2017?

enter image description here

enter image description here

  • 1
    Have you checked the individual bands that are used to compute NDVI in that image? – Spacedman Jul 28 '18 at 8:16
  • 3
    Please do not use ALL CAPS in question titles. Doing so is a downvote magnet because of reduced legibility. Even though you've been here a while, you stll haven't taken the Tour. Please be sure to specify the exact GIS software in use within the body of the question, and to specify the exact inputs and procedure that led to the problem. – Vince Jul 28 '18 at 11:10
  • Yes I checked all individual bands of each image.I used the NIR band5 and Red band 4 for the 2017 Landsat 8 image and NIR band 4 and Red band 3 for the 2007 Landsat 7 image.plus I ve reedited my question for more clarity. – Samuel Gachuhi Jul 29 '18 at 18:41
2

Since you have provided enough details, I made several tests on Landsat 8 data to better understand the situation (I don't mean to insult you).

(1) NDVI from Landsat 8 bands (B5 - B4) / (B5 + B4)

enter image description here

(2) If Landsat 7 and 8 bands were mixed up... (B4 - B3) / (B4 + B3)

enter image description here

(3) For some reason, this one (below) looks similar to your case... (B3 - B5) / (B3 + B5) (cf. Open Water Index by Green and NIR).

enter image description here

So, as Spacedman has commented, you might as well check the input bands again.

  • The problem still remains.I rechecked my input bands and they are all correct. My stacked image is named 'clippedtiff'. So my band 5 was clippedtiff@5 and my band 4 was clippedtiff@4 and I performed the NDVI formula of NDVI=(NIR-RED)/(NIR +RED). So in my case in the raster calculator expression box it was-- (clippedtiff@5-clippedtiff@4)/(clippedtiff@5+clippedtiff@4) and the output was the same in the above 2017 img. In the SCP plugin under bandset it was set to Landsat 8 quickwavelength. The image info is LC08_L1TP_168061_20171228_20180103_01_T1, wrs_path 168, wrs_row 061 for 28-12-2007. – Samuel Gachuhi Aug 3 '18 at 14:58
  • @SamuelGachuhi I think you are doing correctly. Only difference I had in my workflow was that I used each band as separate files (i.e. single-band raster). – Kazuhito Aug 5 '18 at 8:08
  • 1
    Hello, thanks for your innovative comment.I now have a reasonable NDVI Image of 2017. Click here gis.stackexchange.com/questions/290956/… – Samuel Gachuhi Aug 8 '18 at 10:25
2

You might want to check NDVi results for your scene using this satellite image browser or this one. They calculate NDVI as custom index with custom parameters but you can use menu to adjust it to your liking. I tried to run analysis with your palette, classes, levels etc. and here's what I got.

1

enter image description here

The comment from kazuhito did the work,and therefore the NDVI output was more reasonable. I calculated the NDVI using single individual bands as he suggested.

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.