NDVI differencing

[RachelHeadley]

I am teaching some GIS pros about NDVI differencing with Landsat data. Is there a model built to do this in ArcGIS? We managed to do it, but the logic failed on our first few attempts.

[DanPatterson_Retired]

Were you using What is the Image Analysis window?—Help | ArcGIS for Desktop  and some of its related options?

Using the NDVI button on the Image Analysis window—Help | ArcGIS for Desktop 

as your starting point?​

[RachelHeadley]

Seems the NDVI dialog. We wanted to compare an Landsat 5 image with a Landsat 8 image, so we had to correct the default bands for Landsat 8. The trouble came when we wanted to compare them.

A related side-note. Users should be careful to create an NDVI out of a mosaic dataset (as per the instructions for the NDVI button) - seemed to give us unexpected results for the pixel values. But that is probably a different thread.

[DanPatterson_Retired]

Ah yes...there are many links but

How do Landsat 8 band combinations differ from Landsat 7 or Landsat 5 satellite data?

What are the band designations for the Landsat satellites?

is pretty good with visuals etc and there are tons of links from Dr Google using 'compare landsat 5 and landsat 8' related to specific land cover types.

[RachelHeadley]

Yes, all good links, but they don't address NDVI in ArcGIS. I imagine it doesn't exist. Thanks for trying to help!

[JosephMcGlinchy]

Hi Rachel,

There are a few things to consider here. The bands between Landsat 5 and 8 are for sure different. For L5, bands 4 and 3 are needed for NIR and Red, while for L8 it is bands 5 and 4. Also, ideally, each image you are using to produce the NDVI product should be corrected to a surface reflectance product otherwise there will be lots of variability in the result that is false.

You asked about a model, how do you mean? If you have a mosaic dataset for L5 data, and a mosaic dataset for L8 data, you can construct raster functions to produce the NDVI product for each of those, and then difference those NDVI products. That can be done either in python or model builder, I believe.