Comparing NDVI values across Landsat 5 and 8 sensors

276
1
09-10-2021 07:39 AM
MervynLotter
Regular Contributor

The background is that I need to provide an opinion on the impact of grazing on an area using satellite imagery. I need to look at impacts after 2008 so I want to compare imagery taken before 2008 with imagery taken in 2021. I want to use the Normalized Difference Vegetation Index (NDVI) to do so and so I opted to use the Landsat imagery from Landsat 5 TM and Landsat 8 OLI. A quick analysis of the AGOL Living Atlas Multispectral Landsat imagery, comparing June of 2005 with June of 2021, showed that for my study area the average NDVI value June 2005 was 0.3374 and in June 2021 it was 0.3989. So based on this, I was thinking that there had not been any significant impacts.

However, I chose to download additional Landsat imagery from Earth Explorer and their Landsat Collection 2 Level-2 datasets. After doing so and manually computing these indices, I realised that the differences in values is actually much larger and attributed to differences in sensors, and not so much to grazing (as the impact would have been reversed – lower NDVI values in 2021). Using the Earth Explorer datasets, the mean NDVI value for 2007 is 0.1877, while for 2021 it is 0.5343.


This leads me to the purpose of my query.


1) Are the NDVI values from the Multispectral Landsat image service actually comparable across time as they are derived from different sensors measuring slightly different parts of the spectral range? Or has Esri applied some form of normalisation or correction to make them comparable? The metadata did not state anything like this from what I could tell.


2) Are there any approaches I can consider to standardise or correct the bands so that they are comparable? If I can’t correct for these differences, then I cannot use this approach to quantify changes.

I am thinking of moving across to the coarser-scale MODIS NDVI imagery at 250 m as they use the same sensor throughout.

Any assistance would be much appreciated.

Best wishes
Mervyn

0 Kudos
1 Reply
Elijah
by
Occasional Contributor

This is an important issue. It is widely regarded that there is no significant difference between the 2 sensors in what the NDVIs should be given your context. However, research result in this area is not very consistent. Having said that, some factor that may complicate matters are the projection and preprocessing. You may need to be sure that these preprocessing follows same procedure and projections are accurate. 

I have dropped below comments from this source which I think may help your decision in this regard. 

https://www.researchgate.net/post/Is-it-possible-to-have-NDVI-equivalence-between-images-from-Landsa...